Chapter 1 — Definition, Landscape and Stakes

A business that has customers and a decision to make asks what evidence would change the decision before it buys any, so this chapter defines market research and competitive intelligence, keeps them apart from the reading of the business's own numbers and from the voice of its customers, sorts the decisions that need research from those that do not, weighs evidence against opinion and the cost of being wrong against the cost of a study, draws the lines with the neighboring books, dates the landscape of public data, panels, platforms and firms, calls the rules that apply to research and to watching competitors, and sets the decision, the figures and the plan of Nordec Entretien inc. on day 1 of this book.

What You'll Be Able to Do

• Define market research and competitive intelligence in the words this book uses, and say what each one can and cannot answer.
• Separate four activities that are often confused — market research, competitive intelligence, the business's own numbers and the voice of the customer — and name the book that holds each one.
• Decide whether a given decision needs research, using three tests: whether an answer would change the action, whether being wrong is costly or hard to undo, and whether the business's own numbers or its customers already answer.
• Weigh evidence against opinion, and compare the cost of being wrong with the cost of a study before any study is commissioned.
• Size a study to the decision it serves, with the five steps of the method and the four functions a business needs to keep its research in one place.
• Call the rules that apply to research and to watching competitors to the books that own them, and keep self-regulation apart from law.

Why It Matters

A business that has customers already knows a great deal. It knows which buildings it serves, which customers renew, which technician is fastest on a call and which month of the year brings the most inquiries. That knowledge is real, and it is the reason the business exists. What it does not know is the part of the market the business has never touched: the buildings it has never called on, the managers who have never heard its name, the competitors it has never met, and the buyers who chose somebody else for reasons nobody recorded. Market research and competitive intelligence exist to reach that part of the market with evidence rather than with assumption, and the first task of this book is to say clearly what those two activities are, what they are not, and when a decision is worth the cost of either.

The cost of skipping that first task is a business that buys research it does not need and skips research it does. A business that orders a survey to settle a question its own records already answer has spent money to learn what it knew. A business that opens a new territory on the strength of a hunch, without counting the buildings or watching the tenders, has taken a risk it could have measured. The two mistakes look opposite and come from the same place: nobody wrote down, before the work began, which answer would change the action. This chapter gives the tests that separate the decisions that need research from the decisions that do not, and it gives the order in which the work is done: the decision first, the rule second, the evidence third.

There is a third reason the definitions matter, and it is the one that keeps a business out of trouble. Research touches people, and watching competitors touches the law. A conversation with a property manager involves personal information, and the rules on consent and on the use of what is said are set out in the books that own them. A watch on a competitor reads only what the competitor publishes, and the line between reading what is published and obtaining what is not is drawn by statute. A business that treats research as an informal activity, with no consent noted and no source recorded, will eventually discover that the informal version carries the same obligations as the formal one. This chapter names the rules and sends the reader to the books that hold them, and it says plainly that this book describes published rules and is not legal advice.

In This Chapter

Section 1.1 defines market research and competitive intelligence in the words this book uses, and introduces the decision rule, the desk study and the watch. Section 1.2 keeps four activities apart — market research, competitive intelligence, the business's own numbers and the voice of the customer — and names the book that holds each one. Section 1.3 gives the tests that separate the decisions that need research from those that do not. Section 1.4 weighs evidence against opinion and compares the cost of being wrong with the cost of a study. Section 1.5 sizes a study to the decision it serves, with a five-step method and the four functions a business needs. Section 1.6 draws the lines with the neighboring books. Section 1.7 dates the landscape of public data, panels, platforms and firms. Section 1.8 calls the rules that apply to research and to watching competitors. Section 1.9 sets the case on day 1 and walks the plan of this book.

1.1 What Market Research and Competitive Intelligence Are

Two activities sit at the center of this book, and neither of them is the reading of the business's own figures. The first is market research, and this book uses the phrase to mean the answering of a question about a market with evidence collected or read for that purpose. The question might be how many buildings of a certain size stand in a territory, what a group of property managers looks for when choosing a maintenance provider, or what share of a market has heard of a business. The evidence might come from a public register, from a conversation, from a questionnaire put to a sample, or from a digital trace. What makes the activity market research is not the method but the purpose: the evidence is gathered to answer a question about a market, and the question is written down before the evidence is gathered.

The second activity is competitive intelligence, and this book uses the phrase to mean the lasting, lawful reading of what competitors publish and do. The word lasting matters, because competitive intelligence is not a one-time exercise but a routine that continues: a competitor changes a public price page, wins a public tender, posts a job for a role that suggests a new service, and each of those is a signal that a business can read without ever crossing a line. The word lawful matters just as much, and section 1.8 sets out the rules that bound the activity. Competitive intelligence reads what competitors publish; it never reads a price exchanged between competitors, never uses a pretext, and never obtains a secret.

The literature on competitive intelligence is cross-disciplinary, and one review read for this book at abstract only describes the field as mainly an activity of environmental scanning and strategic management (Calof & Wright, 2008). That review is a bibliometric assessment of the competitive intelligence literature, drawing on a business database and on practitioner publications, and its abstract reports that practitioners need a keen understanding of all business functions, especially marketing and strategic planning. The abstract reports no prevalence figure, and this book prints none: no source read for this book gives a share of businesses that practice competitive intelligence, and no source read gives the size of the field.

Three words carry the method of this book, and they are worth defining here because they return in every chapter. A decision rule is the test written before the fieldwork that says which answer leads to which action. A desk study is the reading of public sources before any fieldwork begins. A watch is the routine reading of those sources over time. The three words describe a sequence: a business writes the rule, reads what is already public, and then keeps reading as the market moves. The sequence is what keeps research honest, because a rule written in advance cannot be quietly adjusted to fit a result that has already arrived.

The distinction between the two activities is not academic. A business that confuses them will ask a market research question of a competitor watch, or expect a competitor watch to say how many buyers exist, and the two answers differ in kind, in source and in the decision they support.

Both activities serve a decision, and the decision is named before the work begins. This book's whole method rests on that order. A business that begins with a method — let us run a survey — will produce a document that answers a question nobody asked. A business that begins with a decision — should next year's growth go deeper into the two territories or into a third territory — will produce a study whose result can be applied on the day it arrives. The difference is whether the study had a decision waiting for it.

DEFINITION The words this book uses for market research, competitive intelligence, a decision rule, a desk study and a watch

Market research answers a question about a market with evidence collected or read for that purpose. Competitive intelligence is the lasting, lawful reading of what competitors publish and do, described in the literature read for this book as mainly an activity of environmental scanning and strategic management (Calof & Wright, 2008). A decision rule is the test written before the fieldwork that says which answer leads to which action. A desk study is the reading of public sources before any fieldwork begins. A watch is the routine reading of those sources over time. No source read for this book defines market research, and the first two definitions above are this book's own working words, offered so that the rest of the book can use them consistently. They are not a standard and not a rule; they are the vocabulary of the method that Chapters 2 to 13 apply.

1.2 Research, Intelligence, Analytics and Customer Feedback Kept Apart

Four activities are often spoken of in one breath, and keeping them apart is the first practical skill of this book. The first is market research, which answers who buys, how many, what they need and what they think, and reads public data, conversations, surveys and digital traces. The second is competitive intelligence, which answers what competitors do, offer and signal, and reads public records, tenders, offers, reviews and customers. The third is the reading of the business's own numbers, which answers what the business's visitors and customers did, and reads its analytics, its sales and its records. The fourth is the voice of the customer, which answers what the business's own customers experience, and reads its surveys, interviews and complaints.

The four are not competitors for the same job; they are four different jobs. The business's own numbers can say how many proposals were issued last year and how many became contracts, and it cannot say who never called. The voice of the customer can say what a customer thought of the last service call, and it cannot say what the rest of the market thinks of the business. Market research can say how many buildings of a certain size stand in a territory, and it cannot say what a competitor will do next. Competitive intelligence can say that a competitor has begun advertising a service it did not advertise before, and it cannot say how many buyers exist.

Two of the four belong to other books, and this book cites them rather than redoing them. The business's own numbers belong to Book B7 (Marketing Analytics, Attribution and Dashboards), which holds analytics and attribution. The voice of the customer belongs to Book B10 (Customer Experience, Retention and Loyalty), whose chapter 4 holds the interviews and surveys of the business's own clientele. A business that needs to know what its own website visitors did is not doing market research; it is reading its analytics, and Book B7 is where that work is described.

Two neighbors continue into this book without being redone by it. Book A3 (Strategy and Competitive Positioning) holds the one-day strategic analysis in its chapter 2, and this book makes the watch continuous rather than annual. Book A1 (From Idea to Business Model) holds the discovery research of a venture that has no customers yet, in its chapters 3, 5 and 8, while this book holds research for a business that already has customers, with records, a clientele and a reputation to read.

Figure 1.1 keeps the four activities apart, market research, competitive intelligence, the business's own numbers and the voice of the customer, says what each asks, where it reads and what it cannot say, and sets the case's decision beneath them.

Figure 1.1 — Four activities kept apart. Market research and competitive intelligence are this book's; the business's own numbers belong to Book B7 and the voice of the customer to Book B10. On day 1, Nordec Entretien inc. holds Book B4's figure of about 1,400 reachable buildings and 72 buildings under contract, and the decision this book serves is where next year's growth should go.

The figure is read point by point. The first column is market research, which asks who buys, how many, what they need and what they think, and reads public data, conversations, surveys and digital traces; it cannot say what a competitor will do next. The second column is competitive intelligence, which asks what competitors do, offer and signal, and reads public records, tenders, offers, reviews and customers; it never reads a price exchanged, a pretext or a secret. The third column is the business's own numbers, which ask what the business's visitors and customers did, and read its analytics, its sales and its records; they cannot say who did not come. The fourth column is the voice of the customer, which asks what the business's own customers experience, and reads its surveys, interviews and complaints; it cannot say what the rest of the market thinks. Beneath the four columns, the figure shows the case's starting point and the decision this book serves.

The table below sets out the same four activities in the order in which a business should consider them, with the question each one asks, where it reads, what it cannot say and the book that holds it. Read it as a routing table: the first work of any study is to name the question, and the table says which activity can answer it and which book describes that activity in full.

Table 1.1 — Four activities kept apart

KEY CONCEPT Research is evidence bought before a decision, and its worth is what the decision would cost if it were wrong

The central idea of this chapter is that research is evidence bought before a decision, and that its worth is what the decision would cost if it were wrong. This is this book's own reading and never a finding of a source. Research is worth its cost only where an answer would change an action and where being wrong is costly. The four activities of this section answer four different questions, so the first work of any study is to name the question and to pick the activity that can answer it. Research does not replace judgment; it tests judgment where the test is affordable and the error is expensive. A business that buys evidence for a decision it has already made has bought a decoration, and a business that buys evidence for a decision that no answer could change has bought a ritual. The test is written down before the fieldwork, and it is short.

1.3 The Decisions That Need Research and Those That Do Not

Not every decision needs research, and a business that treats every decision as a research question will spend its time studying instead of acting. This book uses three tests, and they are its own criteria, untested by any source. A decision needs research when the answer would change the action, when being wrong is costly or hard to undo, and when neither the business's own numbers nor its customers already answer the question. A decision does not need research when the answer is already known, when the action is cheap to reverse, or when no answer could change what the business will do.

The three tests are applied in order, and the first one does most of the work. If no possible answer would change the action, the study is pointless regardless of how interesting its result would be. A business that has decided to open a third territory and orders a count of buildings in that territory is not researching a decision; it is documenting one. The second test asks what being wrong would cost. A decision that can be reversed in a week for the price of a letter is not worth a study, and a decision that commits a year of a technician's time and a lease is worth one. The third test asks whether the answer already exists inside the business, in its records or in the mouths of its customers, and if it does, the work is to read it rather than to collect it again.

The tests are not a rule and not a standard, and a business may apply them differently; what matters is that the sorting happens before the money is spent and that its reason is written down. A business that writes "no research: the answer is in the service records" has made a decision it can revisit.

The table below sets out five decisions of the kind a small service business faces, and for each one says whether it needs research, why, and what to do instead when it does not. Read it as a set of worked examples of the three tests: the first row is the case's own decision, and the four rows that follow show the tests applied to decisions that are cheaper, already answered, or answered by a watch rather than by a study.

Table 1.2 — Decisions that need research and decisions that do not

The first row is the decision this book follows, and it is worth saying why it passes all three tests. The answer would change the action, because the rule says to open the third territory only if its adjusted count reaches a threshold and the regional integrator won fewer than half of the public maintenance tenders seen there. Being wrong is costly, because a territory opened on a wrong count commits a year of travel, letters and technician time. And neither the business's own numbers nor its customers answer the question, because the third territory is a market the business has never served. The second and third rows fail the tests for opposite reasons: one is cheap to reverse, and the other is already answered by records the business keeps.

The fourth row shows why a watch is not a study: a competitor's new offer is read from public sources over time, not asked of a sample. The fifth row shows that a complaint is a service matter, and the work it calls for is a reply and a fix, not a study.

The trap this section warns against is the study ordered after the decision, or on a question that no answer could change, and the box below names its signs. The remedy is a single sentence written before any fieldwork: which answer would change the action. If no answer would, the study stops there.

WATCH OUT A study ordered after the decision, or on a question no answer could change, only decorates the choice

The trap of this section is researching a decision that has already been made, or one that no result could reverse. The signs are easy to recognize once they are named. The first sign is a study ordered after the choice: the territory has been opened, the service has been launched, and the research is commissioned to confirm that the choice was sound. The second sign is a question that no answer could contradict: the business asks whether its customers value quality, and every possible answer leads to the same action. The remedy is to write, before any fieldwork, which answer would change the action, and to stop if none would. The reasoning here is this book's own and no source tests it. A study that cannot change an action is not research; it is a record of a decision already taken, and the money it costs would be better spent on the action itself.

1.4 Evidence, Opinion and the Cost of Being Wrong

Evidence is a fact with a source, a date and a method that someone else could check. Opinion is a belief held with or without evidence, and the owner's opinion is included in that description. The distinction is not a judgment about people; it is a statement about what can be examined. A statement that the market holds about 1,400 reachable buildings is evidence if it comes with the count it was drawn from and the date it was drawn. A statement that the market feels bigger than that is opinion, and it may be right, and it cannot be checked by anyone else.

Opinion is not worthless, and this book says so plainly. Experience is how a business knows its field, and an owner who has served a territory for many years holds knowledge that no study will reproduce. The point of research is not to replace that knowledge but to test it where being wrong is costly. An owner's belief about what buyers want is a hypothesis, and a hypothesis that has never been contradicted is not the same as a hypothesis that has been tested. Research is the test, and it is worth its cost only where the test matters.

The cost of being wrong is described in this book as the revenue and the effort that ride on the decision. The cost of a study is described as the hours and the cash it takes. The comparison between the two is this book's own and untested by any source, and it is a comparison of size rather than a calculation of return. A study that costs a small share of what rides on the decision is a study whose cost is easy to justify, and a study that costs more than the decision is worth is a study that should be smaller or should not happen. The comparison says nothing about what the study will earn, and nothing is ever credited to a study.

The case gives the comparison a shape, and the figures are posed by this book. The decision rides on last year's signings multiplied by the average contract, and the study costs the hours and the cash it takes to run. The arithmetic is set out in the box below, and it is labeled illustrative because the case is fictional. The box shows only the size of the study next to the size of what rides on the decision.

DO THE MATH The case's study costs a small share of the revenue that rides on the decision

Illustrative figures; the case is fictional. Step 1 — The internal cost is $1,603: the administrator 38 hours at $35, the lead technician 4 hours at $42 and the dispatcher 3 hours at the administrator's rate, posed by this book. Step 2 — The cash cost is $2,300 paid to the fieldwork service, posed by this book. Step 3 — The total is $3,903, the owner's 22 hours being counted and not priced, as Book B5 does. Step 4 — The revenue at stake is 19 signings times the average contract of $8,541 a year, which is $162,279. Step 5 — $3,903 divided by $162,279 is 2.4 percent, rounded to one decimal, and the comparison says only how large the study is next to what rides on the decision, never what the study earns. Nothing is credited to the study: the decision rule decides, and no contract, sale or share is attributed to the research.

1.5 Time, Budget and the Size of the Question

A small study answers one narrow question with sources that are already public and a few conversations. A large study answers several questions with fieldwork, a sample and a longer calendar. This book gives no typical duration, no typical budget and no typical number of interviews or answers, because no source read for this book supports one, and because a typical figure would be borrowed from a setting that is not the reader's. The case counts its own, and the counts are posed by this book. The size of a study follows from two things: the size of the decision and the time before the decision must be taken.

The time before the decision is a constraint that shapes the method. A decision that must be taken in a month cannot wait for a survey whose fieldwork is not finished, and a decision that can wait a quarter can afford a method that takes longer and answers more. The order is therefore to fix the date first and choose the method second, rather than to choose a method and discover that its results arrive after the choice. A business that has fixed the date can also decide how much of the question it needs answered, because a partial answer delivered on time is worth more than a complete answer delivered late.

The case's plan runs 30 days in three slices of ten days, with gates on days 10, 20 and 30, and a gate that is not met is tried again 3 days later. The plan is set out in Chapter 13, and its shape is worth noting here because it shows how a study is bounded. The first slice writes the decision, the question, the rule and the hypotheses, and reads the desk study. The second slice runs the interviews and the survey and reads the digital signals and the watch. The third slice writes the synthesis, counts the market by two paths, gives the forecast as a range and produces the decision brief. The gates are what keep the study from expanding, because a slice that has not met its gate does not hand its work to the next one.

The method below is this book's own and no source tests it. It is written as five steps because five is enough to keep a study bounded, and it is applied in the case in the order given. The steps are not a standard and they are not a rule; they are the sequence this book recommends and the sequence its chapters follow. A business that applies them will find that the fourth step, choosing the smallest set of methods that can contradict the opinion, is the one that saves the most money, because it is the step that refuses methods whose results could not change anything.

The four functions a business needs to keep its research in one place are described in the box below, and none of them is a specialist product or a brand.

METHOD Sizing a study to a decision in five steps

First, write the decision and the date by which it must be taken. Second, write which answer would change the action. Third, list the evidence that already exists inside and outside the business. Fourth, choose the smallest set of methods that can contradict the opinion. Fifth, set the hours and the cash, and stop when the decision rule can be applied. The sequence is this book's own and no source tests it. The example is the case's decision between deepening two territories and opening a third, and the sequence runs in the order given: the decision and its date, the answer that would change the action, the evidence already held, the smallest set of methods, and the hours and the cash. A study that reaches the fifth step and finds that the rule cannot be applied has learned something worth writing down, and the finding belongs in the decision file.

TOOLS OF THE TRADE What a business needs to keep its research in one place: a decision file, a source log, a shared sheet and a calendar

Four functions carry a study, and no brand name belongs on this list. The first is a decision file that holds the question, the decision rule and the brief, so that the rule written before the fieldwork is still readable on the day the result arrives. The second is a source log that records where each figure came from, with its date and its unit, so that a number printed in a brief can be traced to the page it was read on. The third is a shared sheet for counts, where the case's buildings, completes and signals are kept in one place rather than in three. The fourth is a calendar that holds the gates of the plan and the recurring watch, set by hand. Chapter 12 gives the functions in full, and this chapter names them so that the reader knows what the rest of the book assumes exists.

1.6 Where the Lines Fall With the Neighboring Books

This book sits inside a library, and the lines between its chapters and the neighboring books are drawn deliberately. The first line is with Book A1 (From Idea to Business Model), which holds the discovery research of a venture that has no customers yet, in its chapters 3, 5 and 8. This book holds research for a business that already has customers, and the difference is not the method but the starting point. A venture with no customers is testing whether a market exists at all. A business with customers is testing where its next growth should go, and it brings records, a clientele and a reputation to the question.

The second line is with Book A3 (Strategy and Competitive Positioning), which holds the one-day strategic analysis and the five forces in its chapter 2, and the annual strategy day in its chapter 10. This book makes the watch continuous rather than annual, and it does not redo the analysis. A business that has run the one-day analysis has a picture of its industry; this book keeps that picture current by reading what competitors publish, week by week, and by turning a signal into a decision rather than into a note.

The third line is with Books B3 (Branding and Visual Identity) and B4 (The Marketing Plan, Budget and KPIs). Book B3, section 9.7, prints the margin formula of an awareness measure, and Book B4, chapter 2, prints the reachable market of about 1,400 buildings. This book refreshes both: it measures awareness with a sampled measure and its own margin, and it counts the market by two paths rather than one. The figures of those books are cited as theirs, with their chapters, and they are never recomputed here as new.

The fourth and fifth lines are with Book B7 (Marketing Analytics, Attribution and Dashboards), which holds the business's own analytics, and Book B10 (Customer Experience, Retention and Loyalty), whose chapter 4 holds the voice of the business's own customers. This book samples a market rather than reading a clientele, and the distinction is the one drawn in section 1.2. The sixth line is with Books C3 (Search Engine Optimization), C7 (Social Media Marketing and Syndication) and C9 (Reviews, Testimonials and Social Proof), which hold keyword research for a page, listening to the business's own accounts, and the collection of reviews. This book reads search interest and reviews as market signals, and it does not redo the work those books describe.

The seventh line is with Books H2 to H5, which hold the law, and the rules are called to them rather than rewritten here. The eighth line is with Book C8 (Local Marketing, Directories and Classifieds), which treats local listings, directories and classified ads, and with Book E4 (Consultative Selling, Discovery and Demos), which treats discovery calls and demonstrations; it also names Book E9 (Tenders, RFPs and Public Procurement), Book J8 (Data Management and Business Intelligence), Book K7 (Web Scraping, Data Enrichment and Market Signals), Book L2 (Product Development and Innovation Management) and Book L3 (New Markets and Internationalization).

The table below sets out the lines in full, with the neighbor, what it holds and what this book holds. Read it as a map of the library rather than a hierarchy: the rows are drawn so that a reader who has a question can find the book that owns it, and so that this book does not repeat what another book has already printed.

Table 1.3 — Where this book stops and its neighbors begin

WHO DOES THIS The owner sets the question and decides, and one person runs the study day to day

The work of a study is divided the same way in every size of business, and only the number of people changes. A solo owner does both jobs, sets the question and runs the work, and asks a trusted outsider to challenge the decision rule before the fieldwork begins, because a rule written alone is rarely challenged. A small team has one person run the study day to day while the owner decides, and the split is what keeps the study moving when the owner is on a roof. An established business adds an analyst who prepares the numbers and a reviewer who reads the brief before it is sent. In the case, Sophie Nadeau sets the question and decides, and the administrator runs the study day to day. No hour and no dollar is printed here, because the case counts its own.

1.7 The Landscape in 2026: Public Data, Panels, Platforms and Firms

The landscape a business reads is made of four kinds of thing, and this section describes them by kind. The first is public data: statistics agencies, industry classifications, business counts, enterprise registers, property assessment rolls, public tender systems and open data with their licenses. The second is panels: groups of people who have agreed to answer questions, recruited by a firm that keeps them. The third is platforms: tools that host a questionnaire, a form, a dashboard or a listening feed. The fourth is firms: research firms, fieldwork services, independent researchers and analysts, described by the form of their fee rather than by an amount. This book names none of them and ranks none of them.

The public data is the part of the landscape that a business reads at the page of its publisher. A statistics agency publishes counts of businesses and the classifications that organize them. An enterprise register publishes the public file of a registered business. A property assessment roll lists every property of a municipality, assessed on the same basis and at the same date. A public tender system gathers the calls for tenders published by public bodies. Each of these is a source with a license, a date and a unit, and each is read at its own page on the day it is used.

The counts are worth reading carefully, because the unit matters. In December 2025 there were 1.37 million employer businesses in Canada and 3.67 million non-employer businesses with annual revenues greater than $30,000, and the counts are of businesses active on the Business Register, compiled mainly from Canada Revenue Agency records, usually counted by location, each operating location separately, with employer status depending on payroll remittances in the previous 12 months (Statistics Canada, 2026). The agency asks for cautious interpretation of period-to-period changes, and the counts say nothing of the number of distinct firms, because a business with three operating locations is usually counted three times.

The industry profile is read the same way, with its unit in view. For the industry group 5419, the page read for this book, with data labeled 2025, counts 11,556 employer establishments and 48,145 non-employer or indeterminate establishments in Canada, and 1,576 and 11,397 in Québec, with Canada's employer establishments at 63.5 percent micro, meaning 1 to 4 employees, 35.6 percent small, meaning 5 to 99, and 0.8 percent medium, meaning 100 to 499 (Innovation, Science and Economic Development Canada, n.d.). The industry of marketing research and public opinion polling is listed under that group, and its own page returned no data on the reading day. No count specific to research firms was read, and this book prints no size of the research industry.

The panels and the platforms are the part of the landscape that a business pays for, and the firms are the part it hires. A panel is a group of people who have agreed to answer questions, and this book treats a sample drawn from an opt-in panel as a non-probability sample, for which it prints no margin of error. A platform is a tool that hosts a questionnaire or a form, and a platform is not a sample. A research firm or an independent researcher is hired for a study, and the form of the fee, a fixed price per study, time and materials, or a price per complete, describes the arrangement. Book I8 (Freelancers, Agencies and Outsourcing) holds the work of hiring, and Chapter 11 of this book describes the roles.

The standards that govern the firms are self-regulation, and they are described in section 1.8. The council whose page was read for this book publishes a code adopted from an international code, a pledge meant to help people tell legitimate research requests from other requests, standards for public opinion research, a guideline for research with children and other vulnerable people and guiding principles on artificial intelligence, and its accredited and associate agency members must comply with a standard whose text was not read for this book (Canadian Research Insights Council, n.d.-b). No requirement of that standard is printed here, because the text was not read.

WHAT CHANGED The release of February 13, 2026 counts Canadian businesses as of December 2025, by location

The dated change is a count. The release of February 13, 2026, read on October 10, 2026, counts 1.37 million employer businesses and 3.67 million non-employer businesses with annual revenues greater than $30,000 in December 2025, usually by location, each operating location counted separately, compiled mainly from Canada Revenue Agency records, with the agency asking for cautious interpretation of period-to-period changes (Statistics Canada, 2026). The counts are of businesses active on the Business Register, and they say nothing of the number of distinct firms. Pages and counts change, and this book dates its reading as of October 10, 2026. A business that uses a count should read it on the day of use, note the unit and the date, and print both beside the figure, because a count without its unit and its date cannot be checked by anyone else.

1.8 The Rules That Apply to Research and to Watching Competitors

Research touches people and watching competitors touches the law, and this section calls the rules to the books that own them. The rules are called, never rewritten: one sentence gives the provision, its jurisdiction and its date, and the book that owns it. This book describes published rules on research, personal information, competition and copyright and is not legal advice. A reader who needs to know what their own recording, invitation, call or watch must be is sent to the book that owns the rule and to a lawyer.

The first rule concerns personal information, and it is federal and provincial. The Personal Information Protection and Electronic Documents Act, S.C. 2000, c. 5, Schedule 1, clauses 4.2 to 4.5, current to September 21, 2026, requires that purposes be identified at or before collection, that knowledge and consent be present, that collection be limited to what the identified purposes need, and that use and disclosure be limited to those purposes with retention only as long as necessary; the rule is called to Book H2 (Privacy and Personal Information). The Act respecting the protection of personal information in the private sector, CQLR c P-39.1, ss. 13 and 14, up to date to August 12, 2026, is printed only at the level reused from Books C9 and D2: no communication of personal information to a third person without consent, and consent that is clear, free, informed and specific; the rule is called to Book H2.

The second rule concerns electronic messages and calls. An Act to promote the efficiency and adaptability of the Canadian economy by regulating certain activities that discourage reliance on electronic means of carrying out commercial activities, and to amend the Canadian Radio-television and Telecommunications Commission Act, the Competition Act, the Personal Information Protection and Electronic Documents Act and the Telecommunications Act, S.C. 2010, c. 23, ss. 1(1), 1(2), 1(3) and 6(1), current to September 21, 2026, defines a commercial electronic message and requires consent to send one, and no provision of it names surveys or research; whether a given survey invitation is a commercial electronic message is Book H3's question, and this book poses it rather than deciding it. The national do-not-call exemption for survey, poll and market research calls, read at the commission's page modified April 9, 2018, provides that calls made solely for those purposes are not telemarketing, and whether a call that also sells is covered is Book H3's question; the rule is called to Book H3 (Canadian Radio-television and Telecommunications Commission, 2018).

The third rule concerns competitors. The Competition Act, R.S.C. 1985, c. C-34, s. 45(1) and (2), current to September 21, 2026, makes it an offense for a person, with a competitor, to conspire, agree or arrange to fix prices, allocate sales, territories, customers or markets, or to control production or supply, and the Bureau's guidelines, dated May 6, 2021, say that exchanging pricing information, costs, trading terms, strategic plans or marketing strategies between competitors can raise concerns, and that an exchange of current or future information is riskier than an exchange of historical information; the rule is called to Book H4 (Consumer Protection, Advertising Law, Competition and Language Requirements) (Competition Bureau, 2021).

The fourth rule concerns secrets and copying. The Criminal Code, R.S.C. 1985, c. C-46, s. 391, current to September 21, 2026, makes it an offense to obtain, communicate or make available a trade secret by deceit, falsehood or other fraudulent means, and the rule is called to Book H5 (Intellectual Property). The Copyright Act, R.S.C. 1985, c. C-42, s. 29, current to September 21, 2026, provides that fair dealing for the purpose of research, private study, education, parody or satire does not infringe copyright, and whether a given copy of an industry report is fair is a lawyer's question; the rule is called to Book H5.

The standards that govern research firms are self-regulation, and they bind their members and the associations that adopt them, never a business by law. The ICC/ESOMAR international code, read at its fundamental principles and articles 1 to 8 on October 10, 2026, is self-regulation recognized by over 60 associations in more than 50 countries, revised in 2025 after 2016, and it applies to everyone involved in research, with the client responsible for making its contractors comply (International Chamber of Commerce & ESOMAR, 2025). The council's standards are also self-regulation (Canadian Research Insights Council, n.d.-b). No requirement of a standard whose text was not read is printed, and no amount of a penalty is printed, because no text read for this book prints one.

The table below sets out the rules in the order in which a study meets them, with the jurisdiction and the date read, whom each rule binds, and the book that owns it. Read it as a routing table for the legal questions a study raises: the first two rows concern the people who answer, the third and fourth concern how they are reached, the fifth and sixth concern competitors, and the last three concern secrets, copying and self-regulation.

Table 1.4 — The rules that apply to research and to watching competitors

THE RULE A research code is self-regulation, and the statutes beside it are called to the books that own them (International Chamber of Commerce & ESOMAR, 2025)

The ICC/ESOMAR international code, read at its fundamental principles and articles 1 to 8 on October 10, 2026, is self-regulation recognized by over 60 associations in more than 50 countries, revised in 2025 after 2016, and it binds its members and the associations that adopt it, never a business by law; the client answers for the contractors it uses (International Chamber of Commerce & ESOMAR, 2025). The Canadian council's standards are also self-regulation (Canadian Research Insights Council, n.d.-b). No requirement of a standard whose text was not read is printed. The statutes beside the codes are called to the books that own them: personal information to Book H2, electronic messages and calls to Book H3, agreements between competitors to Book H4, and trade secrets and copying to Book H5. No amount of a penalty is printed, because no text read for this book prints one.

QUÉBEC & CANADA NOTE This book describes published rules on research, personal information, competition and copyright and is not legal advice

This book describes published rules on research, personal information, competition and copyright and is not legal advice. The Canadian texts that frame research are named in one sentence each, with the book that owns them. Personal information is governed federally by the Personal Information Protection and Electronic Documents Act, Schedule 1, clauses 4.2 to 4.5, and in Québec by the Act respecting the protection of personal information in the private sector, ss. 13 and 14, at the level reused from Books C9 and D2, and both are called to Book H2. Electronic messages and calls are governed by the long-titled anti-spam act of 2010 and by the national do-not-call exemption for research calls, and both are called to Book H3. Agreements between competitors are governed by the Competition Act, s. 45(1) and (2), and the Bureau's guidelines, called to Book H4. Trade secrets and copying are governed by the Criminal Code, s. 391, and the Copyright Act, s. 29, called to Book H5. This book explains where the rules stand, recommends no vendor, and a lawyer answers the reader's own case.

1.9 The Case and the Plan of This Book

Day 1 of this book is the first day of the month after Book B4's plan year, and the case is counted in days, never in years or months. Nordec Entretien inc. is a fictional business created for this encyclopedia; its owner, its employees, its customers, its competitors, its figures and the events described are invented, and any resemblance to a real business is coincidental. The business provides preventive maintenance and on-call technical service to commercial buildings on Montréal's north shore, as Book B4 prints in its chapter 1. The owner is Sophie Nadeau, who sets the question, writes the decision rule, approves the research brief and reads the decision brief.

The figures the case starts with are imported from published books, each with its book and chapter. Book B4, chapter 13, prints a year that produced 19 new contracts against a target of 20, 3 buildings lost against a maximum of 3, and 72 buildings against a target of 73, on 35 proposals against 36. Book B4, chapter 2, prints about 1,400 commercial buildings of 2,000 to 9,000 square metres within forty kilometres of the base, an average contract of about $8,541 a year, and the statement that nobody knows how many of the 1,400 reachable buildings are managed by managers the business already knows. Book B3, section 9.7, prints an awareness of 19 of 60, which is 31.7 percent, with a margin of about plus or minus 11.8 points, which cannot be distinguished from chance. Book A3, chapter 1, prints the regional integrator, a regional facility-services firm expected to bid a maintenance component at about $560 a month. Book B5 prints the rates of $35 an hour for the administrator and $42 for the lead technician.

The decision this book serves is posed by this book: where should next year's growth go, deeper into the two territories, or into a third territory next to the eastern one. The rule is written before any fieldwork, and it is printed here word for word: open the third territory only if its adjusted count is at least 400 buildings in the size band and the regional integrator won fewer than half of the public maintenance tenders seen there; otherwise deepen the two territories. The rule is the test that the study exists to apply, and it is written before the first source is read so that no result can quietly change it.

The plan of this book follows the rule. Chapter 2 sets the research design, the question and its decision rule, the hypotheses that research can contradict and the one-page brief. Chapter 3 covers the desk study, the public sources read before any fieldwork. Chapter 4 covers the interviews and focus groups. Chapter 5 covers the survey, its frame and its sample. Chapter 6 covers the digital signals, search interest and reviews read for what they are. Chapter 7 covers the watch of the three competitors. Chapter 8 covers the market count and the forecast. Chapter 9 turns the research into a decision brief. Chapter 10 sets the routine, the weekly round and the monthly review. Chapter 11 covers the help, the roles and when to get it. Chapter 12 covers the tools by function. Chapter 13 gives the plan of thirty days, and day 365 is in section 13.9. The case keeps its own clock in each book that carries it, and this book counts days of its own.

La Machine can host a simple questionnaire as a form whose answers become contacts, keep competitors as company records with dated notes, remind a team of its weekly watch through recurring alerts and tasks, and stage public business listings for prospecting; it draws no sample, computes no margin of error, offers no rating scale, records or transcribes no interview, listens to no social network, monitors no competitor and holds no market data. The case runs its survey outside the application, through the fieldwork service, and its answers stay in the research file.

CASE IN POINT Nordec Entretien inc. starts day 1 with a decision, about 1,400 reachable buildings and no study yet

Nordec Entretien inc. is a fictional business created for this encyclopedia; its owner, its employees, its customers, its competitors, its figures and the events described are invented, and any resemblance to a real business is coincidental. On day 1 of this book, Sophie Nadeau holds the figures Book B4 printed: about 1,400 reachable buildings, 72 buildings under contract, and 19 contracts signed on 35 proposals. Sophie Nadeau also holds Book B3's measure of 19 of 60, which is 31.7 percent with a margin of about plus or minus 11.8 points, and which cannot be distinguished from chance. No study has been run yet. The decision is posed by this book: where should next year's growth go, deeper into the two territories, or into a third territory next to the eastern one. The rule is written before the fieldwork: open the third territory only if its adjusted count is at least 400 buildings in the size band and the regional integrator won fewer than half of the public maintenance tenders seen there; otherwise deepen the two territories.

IN LA MACHINE La Machine hosts a form, keeps company records and rings alerts, and it has no research function

In the application copy read for this book, there is no study or research module: no sampling, no quota, no weighting, no margin of error, no rating scale, no panel, no recruitment of participants, no recording, transcription or coding of interviews, no listening to social networks, no reading of competitors' reviews, no monitoring of competitors' pages, no alert on public tenders, no market data, no market count and no forecast. The Consent page business/marketing/consent and CRM → Do Not Reach business/marketing/do-not-reach govern who may receive a message, and there is no research consent distinct from marketing consent. Control center → Workflows /registre is the register of automated workflows and not a register of enterprises; the enterprise register of Québec is read on its public site. Check it in your installation. The case does not use La Machine for its study. La Machine is described, never sold.

Costly Mistakes

• Buying a study for a decision that has already been made. A study ordered after the choice decorates the choice, and the money it costs would be better spent on the action itself.
• Asking a question that no answer could change. A question whose every possible result leads to the same action is not a research question, and the fieldwork it calls for is a ritual rather than a test.
• Confusing the four activities. A business that asks a market research question of a competitor watch, or expects its own analytics to say who never called, will collect the wrong evidence and read it as though it were the right kind.
• Treating opinion as evidence, or evidence as opinion. An owner's belief about what buyers want is a hypothesis, and a figure with a source, a date and a unit is evidence; the two are useful in different places and are not interchangeable.
• Printing a figure without its source, its date and its unit. A count of businesses by location is not a count of firms, and a figure that arrives without its unit cannot be checked by anyone else.
• Treating a research code as a law, or a law as a suggestion. The ICC/ESOMAR code and the council's standards are self-regulation and bind their members; the statutes beside them are called to Books H2 to H5 and are answered by a lawyer.
• Watching a competitor through anything other than what the competitor publishes. A watch that uses a pretext, or that reads a price exchanged between competitors, has left the activity this book describes and entered the one the law governs.

Key Takeaways

• Market research answers a question about a market with evidence collected or read for that purpose, and competitive intelligence is the lasting, lawful reading of what competitors publish and do; neither definition comes from a source read for this book, and both are this book's own working words.
• Four activities are kept apart: market research and competitive intelligence belong to this book, the business's own numbers belong to Book B7, and the voice of the customer belongs to Book B10, chapter 4.
• A decision needs research when an answer would change the action, when being wrong is costly or hard to undo, and when neither the business's own numbers nor its customers already answer; it does not need research when the answer is known, the action is cheap to reverse, or no answer could change it.
• Evidence is a fact with a source, a date and a method that someone else could check, and opinion is a belief held with or without evidence; research tests opinion where being wrong is costly and leaves it alone where being wrong is cheap.
• The cost of a study is compared with what rides on the decision as a matter of size, and in the case the posed study of $3,903 is 2.4 percent of the $162,279 that rides on the decision, a comparison that says nothing about what the study earns.
• The rules that apply to research and to watching competitors are called to Books H2 to H5, the ICC/ESOMAR code and the council's standards are self-regulation rather than law, and this book describes published rules and is not legal advice.

Your Action Plan

1. Write the decision and its date before anything else. Name the decision in one sentence, name the date by which it must be taken, and put both at the top of a decision file. Deliverable: a decision file holding the decision and its date. Verification: the decision is written as a choice between two or more actions, and the date is a day, not a season.
2. Write which answer would change the action. State, in one sentence, the result that would lead to each action, and stop the study if no result would change anything. Deliverable: a decision rule written before any fieldwork. Verification: the rule names a threshold or a condition, and it can be applied on the day the result arrives.
3. Sort the decision with the three tests. Ask whether an answer would change the action, whether being wrong is costly or hard to undo, and whether the business's own numbers or its customers already answer. Deliverable: a one-line reason for researching or for not researching. Verification: the reason names which of the three tests decided the matter.
4. List the evidence that already exists and choose the smallest set of methods. Read the public sources first, then the business's own records, then decide which methods can contradict the opinion, and set the hours and the cash. Deliverable: a source log and a method list with the hours and the cash. Verification: every method on the list could produce a result that would change the action.
5. Call the rules and keep the research answers in the research file. Name the rules that apply to the study, send each to the book that owns it, and keep the answers and the names of those who gave them out of every list used for selling. Deliverable: a one-page note naming the rules and the book that owns each one. Verification: no person who answered appears in any selling list, and no rule is stated as a conclusion about the reader's own case.

In La Machine

The application copy read for this book carries three pages that a study touches, and it carries none of the study itself (La Machine, 2026). The Consent page business/marketing/consent and CRM → Do Not Reach business/marketing/do-not-reach govern who may receive a message, which matters to a study because an invitation to answer a questionnaire is a message like any other, and because there is no research consent distinct from marketing consent in the copy read. Control center → Workflows /registre is the register of automated workflows, and it is not a register of enterprises; the enterprise register of Québec is read on its public site. The neutral sentence printed in section 1.9 frames the platform in this book. Book N4 (CRM, Contacts and Prospect Data) holds the rules of contacts, and Book H3 (Anti-Spam and Telemarketing Compliance) holds the rules on electronic messages and calls.

What the platform does not do belongs in the same breath, because a business that expects a research function to be built into its software will not build the method. La Machine does not run a study, draw a sample, compute a margin of error or hold market data. The reader keeps the decision file, the source log, the shared sheet of counts and the research data outside the product, and the case runs its survey outside the application, through the fieldwork service. Check it in your installation. La Machine is described, never sold.

Self-Check

1. What does this book mean by market research and by competitive intelligence, and where do those definitions come from?
2. Which four activities does section 1.2 keep apart, and which book holds each one?
3. What are the three tests that decide whether a decision needs research, and what are the three conditions under which it does not?
4. What is the difference between evidence and opinion, and what does the comparison of the cost of a study with the cost of being wrong actually say?
5. Which rules does section 1.8 call, to which books, and what is the difference between a research code and a law?

Answers. 1. Market research answers a question about a market with evidence collected or read for that purpose, and competitive intelligence is the lasting, lawful reading of what competitors publish and do; no source read for this book defines market research, and both definitions are this book's own working words, with the literature read at abstract only describing competitive intelligence as mainly an activity of environmental scanning and strategic management (Calof & Wright, 2008). 2. The four are market research and competitive intelligence, which belong to this book, the business's own numbers, which belong to Book B7 (Marketing Analytics, Attribution and Dashboards), and the voice of the customer, which belongs to Book B10 (Customer Experience, Retention and Loyalty), chapter 4. 3. A decision needs research when an answer would change the action, when being wrong is costly or hard to undo, and when neither the business's own numbers nor its customers already answer; it does not need research when the answer is already known, when the action is cheap to reverse, or when no answer could change it. 4. Evidence is a fact with a source, a date and a method that someone else could check, and opinion is a belief held with or without evidence; the comparison of the cost of a study with the cost of being wrong is a comparison of size, and in the case the posed study of $3,903 is 2.4 percent of the $162,279 that rides on the decision, which says nothing about what the study earns. 5. In this chapter, section 1.8 calls the Personal Information Protection and Electronic Documents Act, Schedule 1, clauses 4.2 to 4.5, and the Act respecting the protection of personal information in the private sector, ss. 13 and 14, to Book H2; the long-titled anti-spam act of 2010 and the national do-not-call exemption for research calls to Book H3; the Competition Act, s. 45(1) and (2), and the Bureau's guidelines to Book H4; and the Criminal Code, s. 391, and the Copyright Act, s. 29, to Book H5; a research code such as the ICC/ESOMAR code and the council's standards is self-regulation that binds its members and the associations that adopt it, never a business by law.

Chapter References

Act respecting the protection of personal information in the private sector, CQLR c P-39.1, ss. 13 and 14 (Québec). LégisQuébec, up to date to August 12, 2026 — read October 9, 2026. https://www.legisquebec.gouv.qc.ca/en/document/cs/P-39.1

An Act to promote the efficiency and adaptability of the Canadian economy by regulating certain activities that discourage reliance on electronic means of carrying out commercial activities, and to amend the Canadian Radio-television and Telecommunications Commission Act, the Competition Act, the Personal Information Protection and Electronic Documents Act and the Telecommunications Act, S.C. 2010, c. 23, ss. 1(1), 1(2), 1(3) and 6(1) (Canada). Justice Laws Website, current to September 21, 2026 (last amended January 1, 2026) — read October 10, 2026. https://laws-lois.justice.gc.ca/eng/acts/E-1.6/page-1.html

Calof, J. L., & Wright, S. (2008). Competitive intelligence. European Journal of Marketing, 42(7/8), 717–730. https://doi.org/10.1108/03090560810877114

Canadian Radio-television and Telecommunications Commission. (2018). Exemptions [National Do Not Call List, information for organizations]. (Page modified April 9, 2018.) Retrieved October 10, 2026, from https://lnnte-dncl.gc.ca/en/Organization/Exemptions

Canadian Research Insights Council. (n.d.-b). Standards. Retrieved October 10, 2026, from https://www.canadianresearchinsightscouncil.ca/standards/

Competition Act, R.S.C. 1985, c. C-34, s. 45(1) and (2) (Canada). Justice Laws Website, current to September 21, 2026 (last amended March 26, 2026) — read October 10, 2026. https://laws-lois.justice.gc.ca/eng/acts/C-34/

Competition Bureau. (2021). Competitor collaboration guidelines. Government of Canada. (Enforcement guidelines dated May 6, 2021; page modified June 27, 2024.) Retrieved October 10, 2026, from https://competition-bureau.canada.ca/en/how-we-foster-competition/education-and-outreach/competitor-collaboration-guidelines

Copyright Act, R.S.C. 1985, c. C-42, s. 29 (Canada). Justice Laws Website, current to September 21, 2026 (last amended November 7, 2024) — read October 10, 2026. https://laws-lois.justice.gc.ca/eng/acts/C-42/section-29.html

Criminal Code, R.S.C. 1985, c. C-46, s. 391 (Canada). Justice Laws Website, current to September 21, 2026 (last amended July 18, 2026) — read October 10, 2026. https://laws-lois.justice.gc.ca/eng/acts/C-46/section-391.html

Innovation, Science and Economic Development Canada. (n.d.). Other professional, scientific and technical services – 5419: Businesses [Canadian Industry Statistics]. Retrieved October 10, 2026, from https://ised-isde.canada.ca/app/ixb/cis/businesses-entreprises/5419

International Chamber of Commerce & ESOMAR. (2025). ICC/ESOMAR international code on market, opinion and social research and data analytics. ESOMAR. Retrieved October 10, 2026, from https://community.esomar.org/uploads/public/knowledge-and-standards/codes-and-guidelines/ICCESOMAR-International-Code_English.pdf

La Machine. (2026). Documentation — route reference inventory. Internal product documentation, read October 10, 2026, together with the code of the pages this book describes: business/marketing/consent, business/marketing/do-not-reach and /registre.

Personal Information Protection and Electronic Documents Act, S.C. 2000, c. 5, Schedule 1, clauses 4.2 to 4.5 (Canada). Justice Laws Website, current to September 21, 2026 (last amended March 4, 2025) — read October 10, 2026. https://laws-lois.justice.gc.ca/eng/acts/P-8.6/page-7.html

Statistics Canada. (2026, February 13). Canadian business counts, December 2025. The Daily. Retrieved October 10, 2026, from https://www150.statcan.gc.ca/n1/daily-quotidien/260213/dq260213c-eng.htm

Chapter 2 — Research Design

A business that designs its research before it collects anything starts from the decision, writes the question and the decision rule while no result exists, states hypotheses that the evidence can contradict, chooses between primary and secondary research and between qualitative and quantitative evidence, defines its population, frame and unit, anticipates the errors a design can carry, and puts all of it on one page, and this chapter does that for Nordec Entretien inc. and its choice between two territories and a third.

What You'll Be Able to Do

• Start from the decision rather than from the method, and write the question that the decision puts to research before any evidence is collected.
• Write a decision rule that names its threshold, the source of each test and the action each outcome leads to, and have the owner approve it in writing before the fieldwork begins.
• State hypotheses that a result could prove false, and write beside each one the result that would reject it.
• Choose between primary and secondary research and between qualitative and quantitative evidence, and say in what order the two are used.
• Define the population, the sampling frame and the unit of analysis, and name the coverage error that a frame can carry.
• Draft a one-page research brief that an owner can read and approve in a single sitting, with the errors the design expects written into it.

Why It Matters

A business that collects evidence without a design does not end up with research. It ends up with a pile. There are notes from a conversation with a property manager, a spreadsheet of buildings drawn from a public roll, a handful of reviews read on a competitor's page, a number somebody remembers from a trade show, and a strong impression that the market is moving in one direction. Every piece of that pile is real, and the pile as a whole answers nothing, because there is no question it was assembled to answer. When the moment to decide arrives, the pile is sorted by whoever feels most strongly about the outcome, and the evidence that survives the sorting is the evidence that agrees with the decision already taken. This is not dishonesty. It is what happens when a business collects before it designs.

The design is what prevents that sorting, and the part of the design that does the most work is the decision rule. A decision rule is written while no result exists, and it says in advance which answer leads to which action. It has a threshold, a source for each test and an action for each outcome. Because it is written first, it cannot be adjusted to fit a result the business likes, and because it is approved by the owner before the fieldwork begins, it is a commitment rather than a preference. A business that writes its rule first will sometimes be told by its own evidence to do something it did not want to do, and that is the point. The rule is the instrument that makes the evidence capable of changing a mind.

The cost of a design is small and the cost of skipping it is not. Writing a question, a rule, three hypotheses and a one-page brief takes an afternoon, and the afternoon is spent before any money is committed to fieldwork. Skipping it costs the fieldwork itself, because a study that answers the wrong question is not a partial success; it is a full expense with no result. Nordec Entretien inc., the fictional business followed through this book, spends a posed $3,903 on its first market study, of which $2,300 is cash paid to a fieldwork service and $1,603 is internal time. The design is the part of that spending that decides whether the other parts produce a decision. A business that writes the brief in an afternoon and reads it before the fieldwork begins has spent the cheapest hour of the whole study.

In This Chapter

Section 2.1 explains why a study begins with the decision rather than with the method, defines the five words this book uses, and sets out where this chapter stops and its neighbors begin. Section 2.2 shows how a decision becomes one research question and how the question gets a decision rule written before any fieldwork. Section 2.3 covers hypotheses that a result could prove false, and the three hypotheses of the case. Section 2.4 compares primary and secondary research and places the case's work in a grid of four cells. Section 2.5 compares qualitative and quantitative evidence and gives the order in which a design uses them. Section 2.6 sorts exploratory, descriptive and causal designs and says why the case claims no cause. Section 2.7 defines the population, the sampling frame and the unit of analysis. Section 2.8 lists the five sources of error a design must anticipate. Section 2.9 presents the one-page research brief and the code that asks a study to disclose its limits. Section 2.10 shows the brief of the case, drafted by the administrator and approved by Sophie Nadeau by gate 1, on day 10.

2.1 Starting From the Decision, Not From the Method

A business that begins with a method begins with a preference. Somebody in the business likes a survey, or has heard of a focus group, or has read that a competitor watches its rivals' job postings, and the method is chosen before the question is asked. The method then answers the question it happens to fit, which is rarely the question the business needed answered. A survey can tell a business how many buildings are managed by a property management firm, and it cannot tell the business why an owner renewed a contract. A set of interviews can tell the business why owners switch providers, and it cannot tell the business how many owners are about to switch. A method chosen first is a method that will produce an answer, and the answer will be to a question nobody asked.

The alternative is to begin with the decision. A decision is a choice the business must make, at a date, with consequences that can be named. Where should next year's growth go? Should the business hire a second technician? Should the price of the entry offer rise? Each of those is a decision, and each one implies a question that evidence could answer. The decision comes first because the decision is what makes the research small. A business that asks what it could learn about its market will learn a great deal and decide nothing. A business that asks what it must know before it chooses between two territories will learn a little and decide.

The decision also fixes the deadline. A study that has no date by which it must be finished expands to fill the time available, and a study that expands is a study that arrives after the choice has been made. The decision has a date because the business has a calendar: a budget season, a hiring moment, a contract renewal, a lease. The research inherits that date, and the date is what tells the business how much research it can afford to do. A question that needs a year to answer is not a question for a decision that must be taken in a month, and a business that discovers this early saves itself the year.

The decision also names the person who decides. In a small business that person is the owner, and the owner is the person the research must serve. This matters because research designed for a committee produces a report, and research designed for one decider produces a brief. Nordec Entretien inc. is a fictional business created for this encyclopedia, and its owner is Sophie Nadeau. The decision this chapter serves is Sophie Nadeau's decision, and the brief of section 2.10 is written to be read and approved by Sophie Nadeau in a single sitting.

The five words this book uses for the work are worth fixing before anything else is written, because a business that uses them loosely will design loosely. They are defined in the callout below, and they are this book's own working words rather than the vocabulary of any source. Book A1 (From Idea to Business Model) defines the discovery research of a venture that has no customers yet, in chapters 3, 5 and 8 of Book A1, and that definition is not repeated or replaced here.

The table below sets out where this chapter stops and where its neighbors begin. It has three columns: the neighbor, what that neighbor holds, and what this chapter holds. Read it as a boundary rather than a division of labor, because the point of the table is that a business reading this book should know which book owns which question before it starts writing.

Table 2.1 — Where this chapter stops and its neighbors begin

DEFINITION The words this book uses for a research question, a decision rule, a hypothesis, primary research and secondary research

A research question is the decision put as something evidence can answer. A decision rule is the test, written before any fieldwork, that says which answer leads to which action. A hypothesis is a statement that a result could prove false. Primary research is evidence the business collects for the question. Secondary research is evidence that already exists and is read for the question. These five are this book's own working words, chosen so that a business can write a brief without borrowing the vocabulary of a discipline, and they are used in the same sense in every chapter that follows. Book A1 (From Idea to Business Model) defines the discovery research of a venture that has no customers yet, in chapters 3, 5 and 8 of Book A1, and that definition stands where it is; this chapter does not restate it and does not compete with it.

KEY CONCEPT A decision rule is a promise made while no result exists, and it keeps the evidence from bending to the first opinion

The central idea of this chapter is that a business writes its decision rule before it collects anything, and that the rule is what makes the evidence capable of changing a mind. Starting from the decision rather than from the method keeps the work small, because a decision names what must be known and a method names only what can be done. Starting from the rule keeps the work honest, because a rule written in advance cannot be adjusted to fit a result the business likes. The rule is approved by the owner in writing, and it is changed only by writing the change and the reason, never after a result has been seen. This reading is this book's own; it is not a finding of any source, and no study is cited for it. The rule is a promise, and like every promise it is worth most at the moment it is inconvenient.

2.2 Writing the Research Question and Its Decision Rule

The research question is the decision put as something evidence can answer. The transformation is smaller than it sounds and it is easy to get wrong. A decision is written as a choice: where should next year's growth go, deeper into the two territories or into a third territory next to the eastern one? The question that follows is narrower: is a third territory worth opening next year? The narrower question is the one evidence can answer, because it has a yes and a no, and because the business can name what it would have to see to say either one.

The decision rule is written next, and it is written before any fieldwork. A rule has three parts. It has a threshold, which is the value at which the answer changes. It has a source for each test, which is where the number comes from. And it has an action for each outcome, which is what the business does if the test passes and what the business does if the test fails. A rule without a threshold is an intention. A rule without a source is an argument waiting to happen. A rule without an action is a report.

The case's rule is posed by this book and is printed here word for word. The decision is where next year's growth should go, deeper into the two territories or into a third territory next to the eastern one. The question is whether a third territory is worth opening next year. The rule is to open the third territory only if its adjusted count is at least 400 buildings in the size band and the regional integrator won fewer than half of the public maintenance tenders seen there, and otherwise to deepen the two territories. The rule has two tests, and both must pass. The first test has a threshold of 400 buildings and a source, which is the property assessment roll read and adjusted. The second test has a threshold of half the tenders seen and a source, which is the public tender system read for the third territory. The action for a pass is to open the third territory. The action for a fail is to deepen the two territories.

Two features of the rule are worth naming. The first is that the rule is conjunctive: a count of 400 buildings with a tender record that fails the second test does not open the territory, and a tender record that passes with a count below 400 does not open it either. A conjunctive rule is stricter than a single test, and it is chosen here because the decision is expensive and the evidence is thin. The second feature is that the rule is written in the language of the business rather than the language of a method. It says buildings and tenders, not sample and incidence, because the person who approves it is the owner and the owner decides in buildings and tenders.

The rule is approved by Sophie Nadeau before any fieldwork. The approval is in writing and it is dated, and the date is a day of this book rather than a calendar date. The reason for the writing is not ceremony. A rule approved in writing is a rule that the business can point to when the result is unwelcome, and a rule approved in a conversation is a rule that can be remembered differently by two people who were both in the room. The written approval is also what makes the rule changeable in a disciplined way: a change is made by writing the change and the reason, and the reason is recorded beside the original rule rather than in place of it.

The rule is not changed after a result has been seen. This is the hardest part of the discipline and the part that a business is most tempted to abandon, because the moment a result arrives is the moment the business knows what it would prefer the rule to have said. A rule changed after a result is not a rule; it is a description of the result. The remedy is procedural rather than moral: the rule is stored where the result will be read, and the change, if there is one, is written on the same page with its date and its reason, so that the change is visible to whoever reads the file a year later.

METHOD Writing a decision rule in five steps

First, write the decision and the action each answer would lead to, so that the business knows what it is choosing between before it knows what the evidence says. Second, turn the decision into one question, narrow enough to have a yes and a no. Third, write the tests with their thresholds and the source of each, naming where the number will come from and who will read it. Fourth, have the owner approve the rule in writing before any fieldwork, with the date of the approval recorded beside it. Fifth, change the rule only by writing the change and the reason, never after seeing a result. The sequence is this book's own and no source tests it. The example is the case's rule with its two tests, a threshold of 400 buildings and a threshold of half the tenders seen, and an action for each outcome.

2.3 Hypotheses That Research Can Contradict

A hypothesis is a statement that a result could prove false. The definition is narrow on purpose. A statement that every possible result would confirm is not a hypothesis; it is a slogan, and a business that writes slogans into its brief has written a set of conclusions and called them questions. The test of a hypothesis is whether the business can name, in advance, the result that would reject it. If the business cannot name that result, the statement is not yet a hypothesis and it is not yet ready to be researched.

The case's three hypotheses are posed by this book. The first is that the third territory holds at least 400 buildings in the size band after a check of the records. The evidence that would contradict it is a count below 400 after the same check. The second is that the regional integrator has won fewer than half of the public maintenance tenders seen in the third territory. The evidence that would contradict it is a record showing the regional integrator winning half or more of the tenders seen. The third is that owner-run buildings are a reachable base at home, in the two territories the business already serves. The evidence that would contradict it is a set of conversations and a survey in which owner-run buildings are consistently harder to reach and less likely to change provider than buildings under a property management firm.

Each hypothesis names its evidence, and each one is written so that a single result can settle it. That is what makes the three of them useful. A hypothesis that needs three studies to test is a research program, and a business with a decision to take in a month does not have a research program. A hypothesis that can be settled by one count, one tender record or one set of conversations is a hypothesis that fits inside the brief.

The section's other point is about what a result can and cannot do. Research can contradict a hypothesis, and research can fail to contradict it. Failing to contradict is not proof. A count of 414 buildings in the third territory is consistent with the first hypothesis and it does not establish that the territory is worth opening, because the second hypothesis may fail and because the count itself carries a coverage error that section 2.8 describes. A business that treats a surviving hypothesis as a confirmed one has moved from research to belief in a single step, and the step is invisible because the evidence was real.

The three hypotheses also have an order, and the order matters for cost. The first two are answered by secondary research, which costs hours of reading. The third is answered by primary research, which costs conversations and a survey. A business that tests the cheap hypotheses first will sometimes find that it does not need the expensive one, because a failed first test settles the decision before the fieldwork begins. The case's rule is written so that this is possible: a count below 400 in the third territory ends the question, and the interviews and the survey are then spent on the two territories the business will deepen.

WATCH OUT A hypothesis that no result could reject is a slogan, and one written after the data is a rationalization

The trap is a statement that every result confirms, and the trap is also a hypothesis that appears after the data has been read. The signs are easy to recognize once a business knows to look for them. The first sign is the absence of a rejecting result beside the statement: if the brief says the market is growing and nothing in the brief says what would show that it is not, the statement is a slogan. The second sign is a rule or a hypothesis that is rewritten in the week a result arrives, with the new version matching the result. The remedy is procedural: write, beside each hypothesis, the result that would reject it, and write it before the fieldwork, so that the rejecting result is on the page when the result arrives. Failing to contradict a hypothesis is not proof of it, and the reasoning in this callout is this book's own.

2.4 Primary and Secondary Research Compared

Secondary research reads what already exists. Primary research collects what does not exist yet. The distinction is not about quality and it is not about difficulty; it is about where the evidence comes from. A property assessment roll, a set of business counts, a public tender notice and a public review were all written by somebody else, for another purpose, and they exist whether or not the business reads them. A conversation with a property manager and an answer to a survey question exist only because the business asked, and they would not exist otherwise.

The case's work divides cleanly across the two, and the division is posed by this book. The desk study reads the property assessment roll, which gives 1,512 buildings in the size band before any check, and it reads the business counts published by the statistics agency. The interviews are 14 conversations, and the survey collects 92 completed answers. The digital signals are read from what is published: a relative search-interest index and 40 public reviews. The watch reads what the three competitors publish. The market count is done by two paths, one from the roll and one from walking a sample of commercial zones.

The cost of the two kinds of research is different in kind, and this book prints no amount for either. Secondary research costs hours of reading, and the hours are the hours of somebody who can find the source, understand what it covers and judge whether it is current. Primary research costs hours and cash, because a conversation takes time to arrange and to write up, and because a survey may be run by a fieldwork service that is paid per completed answer. The case's posed figures are $1,603 of internal time and $2,300 of cash paid to the fieldwork service, and the split is not a rule; it is what this case's design happened to cost.

The order in which the two are used is a design decision, and the order this book recommends is secondary first. The reason is not that secondary research is better. The reason is that secondary research is cheaper, and a business that reads the public sources first will sometimes find that the question is already answered, or that the population is smaller than expected, or that the frame it intended to use does not exist. Each of those findings changes the primary research that follows, and each of them is found before the fieldwork is paid for.

The table below sets out the four cells that the two distinctions produce, with the case's work in each cell and the limit of each kind of evidence. It has four columns: the kind of evidence, what it answers, the case's work and what it cannot say. Read it as a map of the study rather than a ranking, because no cell is better than another and each one is blind to something the others can see.

Table 2.2 — Primary and secondary research, qualitative and quantitative evidence, with the case's work

Figure 2.1 shows the chain from the decision to the method in four steps, then a grid of primary and secondary research against qualitative and quantitative evidence, with the case's work in each cell. The four steps run left to right along the top: the decision, which is where next year's growth goes; the question, which is whether a third territory is worth it; the decision rule, which is two tests written first; and the hypotheses, each of which can be contradicted. Below the four steps sits a grid of two columns and two rows. The columns are secondary research and primary research, and the rows are qualitative and quantitative evidence. The secondary and qualitative cell holds the public reviews and the tender notices read for themes. The secondary and quantitative cell holds the property assessment roll, with 1,512 buildings before checking, and the business counts. The primary and qualitative cell holds the 14 interviews. The primary and quantitative cell holds the survey of 92 completed answers. The figure is read as a whole: the top row is the design, and the grid is the work the design produced.

Figure 2.1 — From the decision to the method. A decision gives a question, the question a decision rule written first, and the rule hypotheses that can be contradicted; the grid then places the case's work: public reviews and tender notices, the property assessment roll with 1,512 buildings before checking, 14 interviews and a survey of 92 completed answers.

2.5 Qualitative and Quantitative Evidence

Qualitative evidence says what and why, in the words and themes of the people who were asked. Quantitative evidence says how many and how much, in counts and shares. The two are not rivals and this book does not rank them. A business that has only counts knows the size of something and not its meaning, and a business that has only themes knows the meaning and not the size. A design that needs both will use them in an order, and the order this book recommends is conversations first and a survey second.

The reason for the order is vocabulary. A survey question is written by the business, and the business writes it in the words the business uses. If the business calls its service a maintenance contract and its customers call it a service plan, the survey will ask about a maintenance contract and the answers will be about something the respondents had to translate. A set of conversations held before the questionnaire is written gives the business the words its customers actually use, and those words go into the questions. The same conversations give the business the themes it will later try to count, and a theme that nobody mentioned in a conversation is a theme the business invented.

The case's interviews are 14 conversations, posed by this book, and their results come in Chapter 4. The case's survey collects 92 completed answers, also posed by this book, and its results come in Chapter 5. The order is the one just described: the conversations are held first, the themes are coded, and the questionnaire is written with the themes and the words that the conversations produced. The survey then counts how many buildings fall into each theme, and the count carries a margin because the survey is drawn as a probability sample, which is the only kind of sample for which this book prints a margin.

The literature on this stage of the work asks questions rather than settling them (Griffin & Hauser, 1993). The paper studies the customer-needs stage of quality function deployment, which is the work of identifying, structuring and prioritizing what customers need, and the abstract reports that the paper asks how many customers to interview, how many analysts should read transcripts, which needs analysts miss, and whether focus groups or one-on-one interviews are better. The abstract states the questions and not the answers, so this book prints no number of interviews and no verdict on focus groups from it. What the abstract shows for a business is the shape of the problem: the number of conversations and the choice between a group and a single conversation are decisions a design has to make, and the paper treats them as open questions.

WHO DOES THIS The administrator drafts the research brief and the owner approves it

The design work is small enough to be done by one person and important enough to need a second reader, and how the two roles are filled depends on the size of the business. A solo owner writes the question, the rule and the brief, and has a trusted outsider read them before the fieldwork begins, because the person who wrote a rule is the person least able to see its gaps. A small team has one person draft the brief and the owner approve it, which is the arrangement at Nordec Entretien inc.: the administrator drafts the one-page brief, and Sophie Nadeau approves it. An established business adds a reviewer who did not take part in the design, so that the rule is read by somebody with no stake in the method chosen. In every size, the person who approves is the person who decides, and the approval is written and dated.

2.6 Exploratory, Descriptive and Causal Designs

A design can be sorted by what it is trying to do, and the sorting below is this book's own. It is offered as a working vocabulary rather than as a classification with a source behind it, and no study is cited for it. An exploratory design looks for the shape of a question when little is known. A descriptive design counts or measures something as it is. A causal design asks whether one thing makes another thing happen, and it needs a comparison group or a comparison period to answer.

An exploratory design is what a business uses when it does not yet know what to ask. The work is qualitative, the sample is small, and the output is a set of themes and a set of words rather than a number. The case's interviews are an exploratory design: 14 conversations, posed by this book, held to learn why owners and property managers change providers and what they say when they describe the change. The design is exploratory because the business did not know, before the conversations, which reasons would turn out to matter, and a questionnaire written before the conversations would have asked about the reasons the business already believed in.

A descriptive design is what a business uses when it knows what it wants to count. The work is quantitative and the output is a count or a share with a stated precision. The case's market count is a descriptive design: the property assessment roll gives 1,512 buildings in the size band before any check, and the count is adjusted after a check of the records. The case's survey is also a descriptive design: 92 completed answers, posed by this book, giving the share of buildings in each category, with a margin because the sample is a probability sample. Neither design explains anything. They describe what is there.

A causal design is what a business uses when it wants to know whether one thing makes another happen, and it is the design a small business is most likely to claim without having. The requirement is a comparison group or a comparison period: a set of buildings that received the treatment and a set that did not, or the same set measured before and after with something else changing at the same time. The case uses no causal design. Nothing in this book credits a contract, a sale or a share to the study, and the reason is not modesty. A before-and-after comparison without a comparison group cannot show that the study caused anything, because the market, the season and the competitor all moved at the same time.

The sorting matters because it tells a business what it may claim. An exploratory design may produce themes and may not produce a share. A descriptive design may produce a count with a margin and may not produce a cause. A causal design may produce a cause and can cost more than a small business can spend. A business that knows which of the three it is running will write a brief that says what the study can and cannot show, and the brief will be read a year later by somebody who needs to know.

The sorting also tells a business when to stop. An exploratory design stops when the conversations stop producing new themes, which is a judgment rather than a number, and Chapter 4 treats it. A descriptive design stops when the sample is large enough for the margin the business is willing to accept, and Chapter 5 treats it. A causal design stops when the comparison has run long enough to separate the treatment from everything else that moved, and this book does not treat it, because the case does not run one. Small tests as research, which are a form of descriptive work with a comparison built in, are treated in section 6.8.

2.7 Populations, Frames and Units of Analysis

Three words decide what a study is about, and a business that gets them wrong will collect good evidence about the wrong thing. The population is everyone or everything the question is about. The sampling frame is the list from which units are drawn. The unit of analysis is what one row of data describes. The three are related but they are not the same, and the gap between the population and the frame is where a study's most common error lives.

The population is defined by the question, not by what is convenient to reach. If the question is whether a third territory is worth opening, the population is the commercial buildings in the size band in the municipalities of that territory. It is not the buildings the business can find, and it is not the buildings whose managers the business already knows. A business that defines its population as what it can reach has defined its frame and called it a population, and the study will then describe the reachable part of the market and report it as the market.

The frame is the list the business actually draws from, and it is always imperfect. The property assessment roll is a frame, compiled for a municipal purpose and not for this study, and like every frame it is imperfect: it may omit a building, include a building that has been demolished, list the same building twice under two cadastral numbers, or classify a building in the wrong category. The methods manual of the national statistical agency names these four kinds of defect as coverage errors, and the manual is a federal agency's methodology rather than a rule that binds a private business (Statistics Canada, 2010). The manual is read here at the sections named in its entry, and it is a Canadian reference for the words this section uses.

The unit of analysis is what one row of data describes, and choosing it is a decision with consequences. The case's unit is the building, posed by this book, and it is never the firm and never the owner. The reason is partly practical and partly legal. The practical reason is that the decision is about buildings: the business services buildings, prices per building and counts buildings. The legal reason is that the roll redacts the owners' names and addresses, and no owner hidden in the roll is ever identified. A study whose unit was the owner would need a frame the business does not have and must not build from the roll.

The frame of the case's survey is drawn in Chapter 5, and it is posed by this book: 1,259 buildings not under contract, of which 1,050 have a reachable decision-maker, a coverage of 83.4 percent. The gap between the 1,259 and the 1,050 is the coverage error of that frame, and it is named in the brief before the fieldwork begins rather than discovered after it. A business that names its coverage error in advance knows what its count can and cannot support, and it will not report a count of buildings as a count of decision-makers.

QUÉBEC & CANADA NOTE The national statistical agency's manual gives the Canadian words for a population, a frame and the errors of a survey

The archived methods manual of the national statistical agency, catalog no. 12-587-X, originally published in October 2003 and read on October 10, 2026, is a Canadian reference for the words this section uses. It defines coverage errors as omissions, erroneous inclusions, duplications and misclassifications in the frame, and it treats sampling error and the several kinds of nonsampling error in the same vocabulary. The manual is a federal agency's methodology, written for the agency's own surveys, and it is a method rather than a rule that binds a private business. A business that reads it is borrowing a vocabulary, not accepting an obligation. This book explains where the practice stands and recommends no vendor, no panel and no research firm; the manual is cited for its definitions, at the sections named in its entry, and nothing in it is presented as a requirement of law.

2.8 Sources of Error a Design Must Anticipate

Every study carries error, and a design that does not name its errors in advance will discover them as surprises. The methods manual of the national statistical agency draws the first distinction a business needs: sampling error exists only in sample surveys and decreases as the sample grows, while nonsampling errors exist in samples and censuses alike (Statistics Canada, 2010). The distinction matters because it tells a business where a larger sample helps and where it does not. A larger sample reduces sampling error. A larger sample does not reduce a bias, because a systematic error is reproduced in every additional unit the business adds.

The manual also names the kinds of nonsampling error a design must anticipate: coverage, measurement, nonresponse and processing (Statistics Canada, 2010). Coverage error is the defect in the frame that section 2.7 described. Measurement error is the difference between what a respondent meant and what the answer recorded, and the manual lists the respondent, the interviewer, the questionnaire, the collection method and the tool among its causes, with technical jargon as one of them. Nonresponse error is the difference between the people who answered and the people who did not. Processing error is what happens between the answer and the number: a transcription, a coding decision, a formula, a merge.

Measurement error is the kind a small business is most likely to introduce without noticing, and the literature on method bias treats it at a level of generality that is useful for a design. A critical review examines how far method biases affect behavioral research, their likely sources, the cognitive processes through which they shape answers, and the procedural and statistical remedies available, with recommendations for choosing remedies by research setting (Podsakoff et al., 2003). The abstract reports the review's scope and not its contents, so this book prints no list of sources and no list of remedies from it. What the abstract points to for a business is that the way a question is asked may itself be a source of error, and that a design which ignores this may attribute to the market what belongs to the questionnaire.

The table below sets out the five sources of error a design must anticipate, where each one comes from, what it does to a result and what the design does about it. It has four columns and five rows, and each cell is a complete sentence so that the table can be read aloud. Read it as a checklist to be worked through before the fieldwork begins, because every row is cheaper to handle in the design than after the results arrive.

Table 2.3 — Five sources of error and what a design does about them

Two consequences follow from the table and both are worth stating plainly. The first is that a business cannot remove these errors; it can only name them, reduce the ones it can and report the ones it cannot. The second is that the errors are not equal in their effect on a decision. A coverage error that omits a tenth of the buildings changes a count by a tenth, and a nonresponse error that brings the wrong tenth into the sample changes the count in a direction nobody can see. A design that names its errors in the brief gives the person who reads the result the information needed to weigh it.

2.9 The One-Page Research Brief

The brief is one page, and the one page is a design decision rather than a formatting preference. A brief that fits on one page can be read and approved by an owner in a single sitting, which is what makes the approval real. A brief of ten pages is read by the person who wrote it and skimmed by the person who approves it, and a skimmed approval is not an approval. The brief is the document that the decision rule, the hypotheses, the population and the errors all fit into, and the constraint of one page is what forces the business to say what it means.

The brief is written before the fieldwork and it is not rewritten during the study. A brief that changes while the study runs is a brief that no longer describes the study, and the person who approved the first version has approved something that no longer exists. The brief is amended only by the same procedure as the decision rule: the change is written, with its date and its reason, beside the original rather than in place of it. The result is a page that shows what the business intended and what it changed, which is more useful a year later than a clean page.

The code that governs research practice asks for the same discipline in its own language. The international code of the market research profession, read at articles 1 to 8 on October 10, 2026, requires that research be fit for purpose, that its limitations be disclosed, and that findings be distinguished from interpretation and recommendations (International Chamber of Commerce & ESOMAR, 2025). The same code requires that known or suspected biases be disclosed. The code is self-regulation, recognized by over 60 associations in more than 50 countries, and it binds its members and the associations that adopt it rather than a business by law. A business that writes its limits into the brief is doing voluntarily what the code asks of the firms that sign it.

The table below sets out the seven fields of the brief, what each field holds and the entry the case makes in it. It has three columns and seven rows, and the case's entries are posed by this book. Read it as the template itself: a business can copy the seven field names into a document and fill them in, and the result is a brief that an owner can approve.

Table 2.4 — The one-page research brief

THE RULE A brief states the limits of the design before the fieldwork, not after (International Chamber of Commerce & ESOMAR, 2025)

The international code of the market research profession, read at articles 1 to 8 on October 10, 2026, requires that research be fit for purpose, that its limitations be disclosed, and that findings be distinguished from interpretation and recommendations (art. 7), and it requires that known or suspected biases be disclosed (art. 8). The code is self-regulation, recognized by over 60 associations in more than 50 countries, and it binds its members and the associations that adopt it, never a business by law. The rule this chapter draws from it is operational: the brief states the limits of the design before the fieldwork and not after, and the example is the case's brief, which names the coverage error in the roll, the nonresponse in the survey and the self-selection in the reviews before any of them is read.

WHAT CHANGED The ICC/ESOMAR code was revised in 2025 after its 2016 edition and now reaches users of self-service platforms

The international code of the market research profession was revised in 2025 after its 2016 edition, and the 2025 text was read on October 10, 2026 (International Chamber of Commerce & ESOMAR, 2025). Two changes matter to a small business. The code now applies to everyone involved in research, including users of self-service platforms, which means that a business running its own questionnaire is inside the code's reach if it has signed on to it or if the association it belongs to has. And the code makes the client responsible for its contractors' compliance, which means that a business that hires a fieldwork service answers for how that service treats the people it calls. Codes change, and this book dates its reading as of October 10, 2026; a business that relies on a code should read the current text rather than this summary.

TOOLS OF THE TRADE What a business needs to design a study: a one-page template, a locked decision rule and a sheet of hypotheses

The design needs four functions and no more. The first is a one-page template for the brief, with the seven fields of the table above, so that the brief cannot grow past a page without somebody noticing. The second is a place for the decision rule that records the date the owner approved it and shows any later change beside the original rather than in place of it. The third is a shared sheet for the hypotheses and the result that would reject each, so that the rejecting result is on the page before the fieldwork begins. The fourth is a calendar for the approval, so that the rule is approved before the first call is made rather than after the first answer arrives. Chapter 12 gives the functions in full. No brand name and no price belongs on this list.

2.10 The Case's Research Brief

The brief of the case is posed by this book. It is one page, it is drafted by the administrator, and it is approved by Sophie Nadeau by gate 1, on day 10. The plan of Chapter 13 runs 30 days in three slices of ten days, with gates on days 10, 20 and 30, and a gate that is not met is tried again 3 days later. Gate 1 is the design gate: the decision, the question, the decision rule and the hypotheses are written, the one-page brief is approved by Sophie Nadeau, the desk study is read, and the survey frame is drawn. Nothing is collected before gate 1 is met, and the reason is the one this whole chapter has given: a rule written after the first result is not a rule.

The brief holds seven fields. The decision is where next year's growth should go, deeper into the two territories or into a third territory next to the eastern one. The question is whether a third territory is worth opening next year. The decision rule is the two-test rule of section 2.2, with its threshold of 400 buildings and its threshold of half the tenders seen. The hypotheses are the three of section 2.3, each with the result that would reject it. The population is the commercial buildings of 2,000 to 9,000 square metres in the municipalities of the territories, and the unit is the building. The methods are listed in the order they will be done: the desk study, 14 interviews, a survey, digital signals, the watch, and two counts of the market. The errors expected are named before any of the work begins: coverage in the roll, nonresponse in the survey, and the fact that reviews are written by those who chose to write.

The errors field is the easiest to leave blank, and it is the field that makes the brief worth writing. The coverage error in the roll is expected because the roll is compiled for a municipal purpose and the business is using it for another; the count of 1,512 buildings in the size band is a count before any check, and the check is part of the design rather than a correction to it. The nonresponse error in the survey is expected because a survey of buildings reaches the decision-makers who answer, and the ones who do not answer are not a random subset. The self-selection in the reviews is expected because a review is written by a person who chose to write one, and 40 public reviews are not a sample of a market; they are 40 statements by 40 people who had something to say.

The brief is kept as an ordinary document outside any research product, because no research product is needed to hold one page. It sits beside the decision rule and the sheet of hypotheses, and the three are read together at the end of the study, when the results arrive and the rule is applied. The brief is not rewritten during the study. If the rule changes, the change is written beside the original with its date and its reason, and the brief then shows both what the business intended and what it changed.

DO THE MATH The case's rule says in advance what a count of 400 buildings and 4 tenders seen would decide

Illustrative figures; the case is fictional. Step 1 — The rule has two tests and says yes only if both pass. Step 2 — The first test is an adjusted count of at least 400 buildings in the size band. Step 3 — The second test is that the regional integrator won fewer than half of the public maintenance tenders seen in the third territory. Step 4 — If 4 tenders were seen, half of 4 is 2, so fewer than half means 1 or none. Step 5 — A count of 400 or more with 2 of 4 tenders won fails the second test, so the rule would say to deepen the two territories, and no rounding is needed. The five steps are the whole of the arithmetic, and nothing is credited to the study: the rule decides, and the study only supplies the two numbers the rule reads.

CASE IN POINT Nordec Entretien inc. writes its research brief and its decision rule before any fieldwork

Nordec Entretien inc. is a fictional business created for this encyclopedia; its owner, its employees, its customers, its competitors, its figures and the events described are invented, and any resemblance to a real business is coincidental. The administrator drafts the one-page brief, and Sophie Nadeau approves it by gate 1, on day 10. The page holds the decision, the question, the two-test rule, the three hypotheses with the result that would reject each, the population and the unit, the methods in the order they will be done, and the errors expected. The brief is posed by this book, and the figures it names are posed by this book as well: the roll's 1,512 buildings before checking, the 14 interviews and the 92 completed answers, whose results come in Chapters 3, 4 and 5. Nothing is collected before the approval, and the rule is not changed after a result is seen.

IN LA MACHINE La Machine has no study module, so the brief is kept as an ordinary document outside it

La Machine can host a simple questionnaire as a form whose answers become contacts, keep competitors as company records with dated notes, remind a team of its weekly watch through recurring alerts and tasks, and stage public business listings for prospecting; it draws no sample, computes no margin of error, offers no rating scale, records or transcribes no interview, listens to no social network, monitors no competitor and holds no market data (La Machine, 2026). For the subject of this chapter, what is missing is the whole of it: there is no study or research module, no sampling, no quota, no weighting, no margin of error, no rating scale and no panel. The reader keeps the brief, the decision rule and the hypotheses outside the product, as ordinary documents, and checks the state of the application in the reader's own installation. The case does not use La Machine for its design.

Costly Mistakes

• Choosing the method before the question. A business that begins with a survey, a focus group or a monitoring tool will get an answer to the question the method happens to fit, and the answer will be to a question nobody asked.
• Writing the decision rule after the first result. A rule written after a result is a description of the result, and the business that writes one has spent the fieldwork and kept the opinion it started with.
• Stating a hypothesis that no result could reject. A statement that every outcome confirms is a slogan, and a brief full of slogans is a set of conclusions wearing the clothes of questions.
• Treating a surviving hypothesis as a confirmed one. Research can contradict a hypothesis or fail to contradict it, and failing to contradict is not proof; a count that is consistent with a hypothesis has not established it.
• Defining the population as what is convenient to reach. A business that defines its population as the buildings it can find has defined its frame and called it a population, and the study will describe the reachable part of the market as the market.
• Leaving the errors field of the brief blank. Coverage, measurement, nonresponse, processing and sampling error are cheaper to name in the design than to discover after the results, and a brief that names none of them is a brief that promises more than the study can deliver.
• Claiming a cause from a before-and-after comparison. Without a comparison group or a comparison period, a change that follows a study cannot be attributed to the study, because the market, the season and the competitor all moved at the same time.

Key Takeaways

• A study begins with the decision rather than with the method, because a decision names what must be known and a method names only what can be done.
• The decision rule is written before any fieldwork, it has a threshold, a source for each test and an action for each outcome, and it is approved by the owner in writing and changed only by writing the change and the reason.
• A hypothesis is a statement that a result could prove false, and each hypothesis is written with the result that would reject it; failing to contradict a hypothesis is not proof of it.
• Secondary research reads what already exists and primary research collects what does not, the two are used in that order, and the case's work divides into public reviews and tender notices, the property assessment roll with 1,512 buildings before checking, 14 interviews and a survey of 92 completed answers.
• The population is what the question is about, the frame is the list the business draws from, and the unit is what one row describes; the case's unit is the building, never the firm and never the owner.
• The one-page brief holds seven fields — the decision, the question, the rule, the hypotheses, the population and unit, the methods and sources, and the errors to anticipate — and it is approved by Sophie Nadeau by gate 1, on day 10, before anything is collected.

Your Action Plan

1. Write the decision and the action each answer would lead to. Name the choice the business must make, the date by which it must be made and what the business does if the evidence points each way. Deliverable: one paragraph naming the decision, its date and the two actions. Verification: the paragraph names an action for each outcome, and no method is named in it.
2. Turn the decision into one question and write the rule. Narrow the decision to a question with a yes and a no, then write the tests with their thresholds, the source of each test and the action for each outcome. Deliverable: a written decision rule with at least one threshold and a named source for every test. Verification: every test has a number and a source, and every outcome has an action.
3. Write the hypotheses with their rejecting results. State each hypothesis so that a single result could prove it false, and write beside it the result that would reject it. Deliverable: a sheet of hypotheses, each with its rejecting result. Verification: no hypothesis is written without a rejecting result beside it, and every rejecting result is something the planned methods could produce.
4. Define the population, the frame and the unit, and name the coverage error. Say what the question is about, which list the business will draw from, what one row of data describes, and what the frame is likely to omit, duplicate or misclassify. Deliverable: one paragraph with the three definitions and the named coverage error. Verification: the unit is stated in one word, and the coverage error is named before any record is read.
5. Draft the one-page brief and have the owner approve it in writing. Fill the seven fields of the brief, keep it to one page, and have the owner read and approve it with a date before any fieldwork begins. Deliverable: a one-page brief with the owner's dated approval. Verification: the brief fits on one page, all seven fields are filled, and the approval date falls before the first call, the first interview and the first record read.

In La Machine

This chapter names no page of the application, and the reason is a fact about the product rather than a gap in the chapter. The application copy read for this book on October 10, 2026 has no study or research module, so there is no page for a decision rule, a hypothesis or a research brief (La Machine, 2026). What the platform does carry is the work around a study rather than the study itself: a form can host a simple questionnaire whose answers become contacts, a competitor is a company record with dated notes, a watch is a recurring reminder and a task, and the prospecting page stages public business listings for the administrator and reseller roles. The case's survey is run by the fieldwork service outside the application, and its answers stay in the research file.

What the platform does not do belongs in the same breath, because a business that expects the design to be built into the software will not build it. La Machine does not design a study, draw a sample or compute a margin of error. The reader keeps the brief, the decision rule and the sheet of hypotheses outside the product, as ordinary documents that the owner can read and approve, and checks the state of the application in the reader's own installation. La Machine is described here and never sold, and this book is valid with any tool.

Self-Check

1. Why does a study begin with the decision rather than with the method, and what are the five words this book uses for the work?
2. What three parts does a decision rule have, and why is the rule approved in writing before any fieldwork?
3. What makes a statement a hypothesis, and what does it mean when research fails to contradict one?
4. What is the difference between the population, the sampling frame and the unit of analysis, and what is the case's unit?
5. What are the seven fields of the one-page brief, and which field is the easiest to leave blank?

Answers. 1. A study begins with the decision because a decision names what must be known and a method names only what can be done, so a method chosen first answers the question it happens to fit; the five words are the research question, the decision rule, the hypothesis, primary research and secondary research, and they are this book's own working words. 2. A rule has a threshold, a source for each test and an action for each outcome; it is approved in writing before any fieldwork so that it cannot be adjusted to fit a result the business likes, and so that a change is made by writing the change and the reason rather than by remembering the conversation differently. 3. A statement is a hypothesis when a result could prove it false, which means the business can name in advance the result that would reject it; failing to contradict a hypothesis is not proof of it, because the result may be consistent with the hypothesis without establishing it. 4. The population is what the question is about, the frame is the list the business draws from, and the unit is what one row of data describes; the case's unit is the building, never the firm and never the owner, because the decision is about buildings and because the owners' names and addresses are redacted in the roll and are never identified. 5. The seven fields are the decision, the question, the decision rule, the hypotheses, the population and unit, the methods and sources, and the errors to anticipate; the errors field is the easiest to leave blank, and it is the field that makes the brief worth writing.

Chapter References

Griffin, A., & Hauser, J. R. (1993). The voice of the customer. Marketing Science, 12(1), 1–27. https://doi.org/10.1287/mksc.12.1.1

International Chamber of Commerce & ESOMAR. (2025). ICC/ESOMAR international code on market, opinion and social research and data analytics. ESOMAR. Retrieved October 10, 2026, from https://community.esomar.org/uploads/public/knowledge-and-standards/codes-and-guidelines/ICCESOMAR-International-Code_English.pdf

La Machine. (2026). Documentation — route reference inventory. Internal product documentation, read October 10, 2026, together with the code of the pages this book describes: no route is named in this chapter.

Podsakoff, P. M., MacKenzie, S. B., Lee, J.-Y., & Podsakoff, N. P. (2003). Common method biases in behavioral research: A critical review of the literature and recommended remedies. Journal of Applied Psychology, 88(5), 879–903. https://doi.org/10.1037/0021-9010.88.5.879

Statistics Canada. (2010). Survey methods and practices (Catalogue no. 12-587-X; originally published October 2003; archived). Retrieved October 10, 2026, from https://www150.statcan.gc.ca/n1/pub/12-587-x/12-587-x2003001-eng.htm

Chapter 3 — Secondary Research and Data Sources

A business that reads before it asks starts with what the public already holds, so this chapter shows why secondary research comes first, what the national and provincial statistics agencies, the industry classification, the register of enterprises, open data and its licenses, the public tender system and the property assessment roll each say and do not say, how to judge an industry report, how to judge and document any source so that the work can be redone, and how Nordec Entretien inc. counts the buildings of its size band from the roll and checks the count.

What You'll Be Able to Do

• Explain why secondary research comes first in a study, and what a desk study, an administrative record, open data and a source log each mean in this book.
• Read a count of businesses published by a statistics agency with its unit and its date beside it, and say plainly what that count does not measure.
• Use the industry classification to find the class of an activity, and read an industry profile for what it shows rather than for what it leaves out.
• Consult a public register of enterprises, an open data portal and a public tender system, and record the license or the terms that govern each one.
• Judge a source by four questions — who made it, how, when and for whom — and write the answers into a source log that another person could follow.
• Count a market from a property assessment roll, check the count against a sample of records, and report the working count with its limits attached.

Why It Matters

A study that begins with a conversation begins in the wrong place. A conversation is costly evidence for a small business, because it costs scarce hours in the firm and it costs the goodwill of the person on the other end of the line. A business that books fourteen interviews before it has read anything will spend those fourteen hours asking questions that a public register, a property roll or a statistics release would have answered at a desk, and it will arrive at the fourteenth interview with the same understanding it had at the first. Reading first is not a courtesy to the reader. It is what turns a conversation into a test of something the business already half knows.

There is a second reason, and it is the one that decides whether a study is worth anything at all. Public sources are evidence that can be checked by somebody else. An interview produces a theme and a count of codes, and no second person can re-run it. A roll produces a number of buildings, and any second person who opens the same file, applies the same size band and reads the same license can arrive at the same number or find the mistake. That property — the ability of another person to redo the work — is what makes a desk study the backbone of a market count, and it is why this chapter spends as much space on documenting a source as it does on finding one.

The third reason is the least comfortable. A great deal of what is published about a market is published by somebody with an interest in it, and a business that reads without asking who paid for the reading will absorb a claim as though it were a fact. A statistics agency counts locations because that is what its register holds, and a business that reads the count as a count of firms will overstate its market. An industry profile counts establishments in an industry group, and a business that reads the group as its own industry will misjudge its competition. Reading well means reading the author before reading the number.

In This Chapter

Section 3.1 explains why secondary research comes first, gives the five working words this book uses and sets out the frontier between this chapter and its neighbors. Section 3.2 reads the national statistics agency's counts of businesses and the provincial institute's home page, and puts the unit and the date beside every count. Section 3.3 covers the industry classification, its six-digit structure and the industry profile as an example of what such a page shows. Section 3.4 covers the public register of enterprises and what a public file does and does not show. Section 3.5 covers open data and the conditions of an open license. Section 3.6 covers the property assessment roll and the public tender system as administrative records. Section 3.7 covers industry reports, associations and the trade press, and how to read them for method. Section 3.8 gives four questions to put to any source. Section 3.9 gives the fields of a source log so that the work can be redone. Section 3.10 runs the case's desk study, counts the buildings of the size band from the roll and checks the count against a sample of records.

3.1 Why Secondary Research Comes First

A business that already has customers and a decision in front of it may reach for the telephone first, and that instinct is understandable and expensive. The telephone reaches one person at a time, at a cost measured in hours, and what is already published tells a business which questions are worth a conversation, which are already answered, and which cannot be answered by anybody at all. A desk study is not a substitute for talking to people. It is what makes talking to people worth the hour.

Secondary research is the reading of evidence that already exists, and in a business that has customers it is the first move of every study this book describes. The evidence sits in public bodies, in industry associations, in the trade press and in the business's own files, and it is read before a single interview is booked and before a single questionnaire is drafted. Reading is not free, because hours are the scarcest thing a small business has, and a desk study of the kind this chapter describes takes hours. What reading buys is a question worth asking. A business that has read the roll knows how many buildings of its size band stand in a territory, and it can therefore ask a property manager a question about the buildings that are not under contract rather than a question about the market in general.

The order matters more than the reading. A study that reads first and asks second arrives at the interviews with a hypothesis and a count, and the interviews either support the count or contradict it. A study that asks first and reads second looks for the source that confirms its impressions, which is a way of buying reassurance rather than evidence.

Five working words carry the rest of this chapter, and they are this book's own working words rather than terms of art borrowed from a source. They are defined in the note below and used in the same sense throughout the book. A business that keeps them apart will find that most of the confusion in a desk study comes from using one of them where another belongs: a register read as though it were a count, a count read as though it were a market, a license read as though it were a permission to do anything at all with what was read.

The frontier between this chapter and its neighbors is drawn before the reading starts. The table below sets out five neighbors, what each of them holds, and what this chapter holds instead. Read it as a boundary rather than as a hierarchy: the neighbors are not alternatives to this chapter, and this chapter does not do their work.

Table 3.1 — Where this chapter stops and its neighbors begin

The judgment this chapter asks for is stated once and used everywhere after it. A source is evidence only once its author, its design, its date and its purpose are known. A number without an author is a rumor with a decimal point. A count without its unit is a count of nothing in particular. A page without its date is a page that may have been rewritten since it was read. A report without its purpose is a report whose selection of facts the reader cannot weigh. The four questions of section 3.8 are the working form of this judgment, and the source log of section 3.9 is the record that keeps the answers.

DEFINITION The words this book uses for secondary research, a desk study, an administrative record, open data and a source log

Secondary research is the reading of evidence that already exists, produced by somebody else for a purpose of their own, before any fieldwork is done. A desk study is the reading of public sources before any fieldwork, done at a desk, in a fixed order, and written down as it proceeds. An administrative record is a record a public body keeps for its own work — a property roll, a register of enterprises, a list of tender notices — which research reads for another purpose entirely, and whose fields were designed for that other purpose. Open data is information a public body publishes under a license that lets anyone reuse it, subject to the conditions the license states. A source log is the record of each source's author, address or file, date read, date or version of the source, license or terms, and what was taken from it and what was not. These five are this book's own working words, used in this sense throughout, and they are not terms of art of any standards body.

KEY CONCEPT A source is evidence only once its author, its design, its date and its purpose are known

The central idea of this chapter is that reading comes first because it is the cheapest evidence a business can obtain and because it tells the business what to ask people, and that what is read is judged and recorded before it is used. A number carries four things with it or it carries nothing: the author who produced it, the design or the record it came from, the date it describes, and the purpose for which it was produced. A count of employer locations produced by a statistics agency for the purpose of compiling production statistics is a good count of locations and a poor count of firms, and the difference is not a defect of the source but a fact about it. This reading is this book's own and is not a finding of any source. It is the reason the chapter spends its second half on four questions and a log rather than on a list of places to look.

3.2 National and Provincial Statistics Agencies

The first place a Canadian business looks is the national statistics agency, and the first thing it must learn there is what the agency counts. In December 2025 there were 1.37 million employer businesses in Canada and 3.67 million non-employer businesses with annual revenues greater than $30,000 (Statistics Canada, 2026). Those two numbers are the headline of a release read on October 10, 2026, and they are useful, and they are also easy to misuse. The counts are of businesses active on the Business Register, compiled mainly from Canada Revenue Agency records; businesses are usually counted by location, each operating location separately; and employer status depends on payroll remittances in the previous 12 months (Statistics Canada, 2026). A business that reads the first number as a count of firms, or as a count of the buildings it might serve, has read a good number as a different number.

The release is careful in a way that is worth copying. It asks for cautious interpretation of period-to-period changes, and it says nothing about the number of distinct firms, about openings or about closures (Statistics Canada, 2026). A business that wants to know how many competitors opened in its territory will not find the answer here, and a business that treats a change from one period to the next as a trend will be reading a movement that the agency itself declines to interpret. The discipline this chapter draws is small and it is repeated in every section that follows: write the unit and the date beside every count, in the same sentence, before the count is used for anything.

The provincial layer is a separate source with a separate purpose. The Institut de la statistique du Québec describes itself on its home page, read on October 10, 2026, as the province's statistics agency, with a mission to provide reliable, objective and relevant statistics on Québec society; it publishes by theme, including demography, economy, education, employment, health and environment, it runs surveys and offers access to research data, and its page features main indicators on Québec and its regions together with a tool to explore data by administrative region and by regional county municipality (Institut de la statistique du Québec, n.d.). The page carries no date of its own, and it was read on October 10, 2026. What it establishes is where the provincial statistics live and at what geography they can be read. It establishes no value of any indicator, and this chapter prints none.

The two agencies are read together for a reason that has nothing to do with rank. The national agency counts the whole country at a geography that may be too coarse for one shore of one island, and the provincial institute publishes at the region and the regional county municipality. Neither replaces the other, and neither answers the question of this case, which is how many buildings of a given size stand in a given set of municipalities. That question belongs to the property roll of section 3.6, because the roll is built building by building for municipal taxation, while a business register is built business by business for compiling statistics.

WATCH OUT A count of employer locations is not a count of firms, and neither is a count of buildings

The trap is to read the counts published by the national statistics agency as a count of firms, of customers or of buildings. The counts are of businesses active on the Business Register, compiled mainly from Canada Revenue Agency records, and businesses are usually counted by location, each operating location separately, with employer status depending on payroll remittances in the previous 12 months (Statistics Canada, 2026). The signs that a business has fallen into the trap are easy to spot in a draft brief: a number of businesses used as a number of customers' buildings, a number of businesses used as a number of competitors, and a change from one period to the next read as a trend without the caution the agency itself asks for. The remedy is a habit rather than a correction: write the unit and the date beside every count, in the same sentence, and write what the count does not measure in the sentence after it. This reasoning is this book's own and is not a finding of the agency.

3.3 Industry Classifications and Business Counts

An industry classification is the vocabulary that lets a business compare itself with a published count, and it is worth an hour of a study to learn how it is built. The North American Industry Classification System was developed by the statistical agencies of Canada, Mexico and the United States to provide common definitions of the industrial structure of the three countries; it is production-oriented, grouping establishments by similar production processes; and it is designed mainly for compiling production statistics and classifying establishments, though it can also classify companies and enterprises, the higher levels of the classification suiting that better (Statistics Canada, 2022). A business that understands those three sentences will use the classification correctly and will not ask it to do something it was not built to do.

The code is hierarchical and it is read from the left. Digits 1 and 2 give the sector, digit 3 the subsector, digit 4 the industry group, digit 5 the industry and digit 6 the Canadian industry, with a zero meaning that there is no further national detail (Statistics Canada, 2022). Version 1.0 of the 2022 version holds 20 sectors, 99 subsectors, 323 industry groups, 694 industries and 923 Canadian industries (Statistics Canada, 2022). The table below sets out those five levels in the order in which the code is read, with the number of digits each level uses, what it groups and how many of each the 2022 version holds. Read it as a map of the vocabulary rather than as a list of industries: the counts are of classes in the classification, not of businesses in the economy.

Table 3.2 — The levels of the industry classification

The counts of classes are one thing and the counts of businesses are another, and the second is what an industry profile publishes. The page for the industry group 5419, read on October 10, 2026, with data labeled 2025, gives Canada 11,556 employer establishments and 48,145 non-employer or indeterminate establishments, and Québec 1,576 and 11,397; Canada's employer establishments in the group are 63.5 percent micro, meaning 1 to 4 employees, 35.6 percent small, meaning 5 to 99 employees, and 0.8 percent medium, meaning 100 to 499 employees (Innovation, Science and Economic Development Canada, n.d.). The page also lists an industry named for marketing research and public opinion polling under the group, and its own page returned no data on the reading day (Innovation, Science and Economic Development Canada, n.d.). That last fact is the one this book must state plainly: the industry group read does not break research firms out, and no count specific to research firms was read.

What an industry profile shows is an industry group, its establishments and its year, and not a count of the businesses a reader has in mind, because the group contains several industries and the reader's own activity may sit in one of them, in another group, or across two. The classification is a vocabulary for finding published counts, and it is most useful when a business uses it to locate the profile closest to its activity and then reads the profile's own definition of what it counted.

Finding the class of an activity is a procedure rather than a lookup, and it is done the same way every time. Open the current version of the classification, search the activity by its ordinary words, read the class definition and its exclusions, and check the code in the version actually used. The exclusions matter as much as the definition, because a class is defined partly by what it does not contain. The class of building maintenance and the class of marketing research were not read on either page consulted for this chapter, so this book prints no code for the case's activity and none for research firms, and it says instead how the class is found and that the code is checked in the version used.

The version matters because classifications are revised. The introduction page of the current version was modified in 2023 and read on October 10, 2026 (Statistics Canada, 2022). The page of the 2017 version, modified in 2021 and also read on October 10, 2026, says that the 2017 standard was replaced by the 2022 version (Statistics Canada, 2017). A count published under one version and a class read under another are not the same object, and a business that records the version it used will not have to reconstruct the comparison a year later.

WHAT CHANGED The 2022 version of the industry classification replaced the 2017 version, and the work records the version it used

The change is dated by its source. The introduction page of the current version of the North American Industry Classification System, modified in 2023 and read on October 10, 2026, presents the 2022 version (Statistics Canada, 2022), and the page of the 2017 version, modified in 2021 and read on October 10, 2026, says that the 2017 standard was replaced by the 2022 version (Statistics Canada, 2017). Classifications are revised, and a revision changes the classes as well as the counts published under them, so a code read under one version and a count published under another are not comparable without saying so. The practice this chapter draws is small: record the version of the classification beside the code, the same way the date is recorded beside a count, and re-check the code when a new version is read. This book dates its own reading as of October 10, 2026, and prints no code for the case's activity.

3.4 Public Registers of Enterprises

A public register of enterprises is the second kind of administrative record a business reads, and it answers a narrower question than a statistics release. The register of enterprises of Québec can be consulted free of charge; an enterprise can be searched by its Québec enterprise number, by its name or by a related word, and a natural person can be searched by name to see the enterprises the person is linked to; an enterprise's public file shows its number, its registration status, the names it uses, the addresses of its establishments, the contact details of its directors and officers and the names of its ultimate beneficial owners (Gouvernement du Québec, 2025). The page was updated on September 23, 2025, and read on October 10, 2026.

What the register shows is existence, status and addresses. It shows that a company exists, that it is registered or struck off, where its establishments are, and who its directors and officers are. What it does not show is revenue, customers, prices, contracts or the number of buildings a company serves. A business that reads the register hoping to size a competitor will find a name and an address and nothing that measures the competitor's business, whereas a business that reads the register to confirm that a company named on a tender notice exists has used the register for exactly what it holds.

The register is read in this book's case for two narrow purposes, both of them posed by this book. The first is to watch for new establishments of the three competitors, which is one of the signals the watch of Chapter 7 collects. The second is to check that a company named on a tender notice exists and is registered, before the notice is recorded as a signal about a competitor. Neither purpose requires anything the register does not publish, and neither purpose involves contacting anybody whose name appears in the file.

The page states no rule on reusing the information in the register for solicitation, and this book does not supply one. A business that wants to use a name or a contact detail from the register in a message is asking a question about personal information and about electronic messages, which belongs to Book H2 (Privacy and Personal Information) and to Book H3 (Anti-Spam and Telemarketing Compliance), which alone answers whether a given message is a commercial electronic message. A business with a plan for the names it has read should put the plan to a lawyer.

One sentence is needed about a name that looks like this register and is not. In the application described in this book, Control center → Workflows /registre is the register of automated workflows; it is not a register of enterprises, and the enterprise register of Québec is read on its public site (Gouvernement du Québec, 2025). A reader who goes looking for the register inside a business platform will find a list of automations, and the confusion is worth preventing once.

A public file is a snapshot of a status at a moment, and a status changes, so a business records the date it read the file beside the fact it took from it and does not have to wonder a year later whether the fact was ever true.

3.5 Open Data and Its Licenses

Open data is information a public body publishes under a license that lets anyone reuse it, and the license is the part of open data that a business most often skips. The Open Government Licence – Canada, version 2.0, on a page modified on December 2, 2022 and read on October 10, 2026, lets a user copy, modify, publish, translate, adapt and distribute the information for any lawful purpose, commercial purposes included, worldwide, royalty-free and non-exclusive (Government of Canada, n.d.). That is a broad permission, and it is the reason open data is worth the reading time. It is also a permission with conditions, and the conditions are short enough to read in full.

The conditions are three, and the box below sets them out: attribution of the source, a set of exclusions that includes personal information, and a prohibition on suggesting an official status or an endorsement (Government of Canada, n.d.). A business that publishes a count drawn from an open data set, in a brief or on a page, is publishing under those conditions, and the attribution statement is the part of the publication that carries them.

The exclusions are the part that matters most to a market study, because a data set can be open and still contain personal information that the license does not cover. The federal license says so in terms, and a provincial or municipal portal has its own license rather than this one (Government of Canada, n.d.). The property assessment roll of Québec, for example, is published under an Attribution license, CC BY 4.0, which is not the federal license and is read in the next section (Ministère des Affaires municipales et de l'Habitation, 2026).

The practical consequence is a habit that costs a minute and saves a study: every time a source is opened, the license or the terms are read and recorded with the source, in the same entry, beside the date read. The source log of section 3.9 has a field for exactly this.

THE RULE An open license is a permission with conditions, and the conditions are recorded with the source (Government of Canada, n.d.)

The Open Government Licence – Canada, version 2.0, on a page modified on December 2, 2022 and read on October 10, 2026, lets a user copy, modify, publish, translate, adapt and distribute the information for any lawful purpose, commercial purposes included, worldwide, royalty-free and non-exclusive (Government of Canada, n.d.). The user must acknowledge the source with the provider's attribution statement where one is given, or with the general statement "Contains information licensed under the Open Government Licence – Canada" (Government of Canada, n.d.). The license does not cover personal information, third-party rights, the provider's names, crests and logos, or information under other intellectual property rights, and the user must not suggest an official status or an endorsement (Government of Canada, n.d.). A provincial or municipal portal has its own license rather than this one. The rule this chapter draws is operational: read the license when the source is opened, record it with the source, and carry the attribution statement into whatever is published.

3.6 Public Tenders, Property Rolls and Other Administrative Records

Two administrative records do most of the work in a market count, and they are different in kind. The first is the property assessment roll, which is built property by property for the purpose of municipal taxation. The second is the public tender system, which is built notice by notice for the purpose of public purchasing.

The property assessment roll of Québec lists every property of a municipality, assessed on the same basis and at the same date, and it is the central element of the municipal tax system; the data set holds one file per municipal roll, 1,140 files, together with georeferenced files for 2026, and it is updated yearly (Ministère des Affaires municipales et de l'Habitation, 2026). The catalog record was read on October 10, 2026, with metadata modified on August 24, 2026 and the 2026 files created on March 3 and April 1, 2026. The files themselves were not opened for this book, so the fields of a record are read on the day of use and this chapter prints none of them.

What the record does say is what has been removed: cadastral numbers, owners' names and postal addresses are redacted, and re-identifying the redacted personal information, meaning the owners' names and addresses, is prohibited (Ministère des Affaires municipales et de l'Habitation, 2026). The license is Attribution, CC BY 4.0 (Ministère des Affaires municipales et de l'Habitation, 2026). For a market count, that prohibition is a definition of the unit rather than an obstacle: the roll counts buildings, and a business that counts buildings from the roll is counting exactly what it publishes.

The public tender system is the second record, and it is read for a different question. The electronic tendering system of Québec gathers the calls for tenders and other notices published by most public bodies of Québec, so that businesses can find opportunities; notices come from ministries and public bodies, from Santé Québec and its establishments, from the education network and from the municipalities; and the system carries public calls for tenders, notices of intent for a contract by mutual agreement and notices of interest, with tender documents such as specifications and addenda (Gouvernement du Québec, 2026). The page was updated on August 31, 2026 and read on October 10, 2026.

What the page does not carry is as important as what it does. Awarded contracts, bidders' names and amounts, and any fee for consulting the system were not on the page read, so this chapter prints none of them, and a business should verify what the system shows on the day of use (Gouvernement du Québec, 2026). Book E9 (Tenders, RFPs and Public Procurement) treats finding opportunities in public tenders, including award results as a map of the market in its section 2.8. This chapter reads tender notices as a source of what public bodies buy and when, which is a signal about a market and not a sales process.

QUÉBEC & CANADA NOTE The Québec assessment roll redacts its owners and forbids re-identifying them, so the case counts buildings and never owners

The catalog record of the property assessment roll of Québec, read on October 10, 2026, says that cadastral numbers, owners' names and postal addresses are redacted, that the information can be consulted unit by unit on the municipality's site or at its clerk's office, and that re-identifying the redacted personal information is prohibited (Ministère des Affaires municipales et de l'Habitation, 2026). The license is Attribution, CC BY 4.0, which requires the source to be credited (Ministère des Affaires municipales et de l'Habitation, 2026). The consequence for a market count is simple and it is a definition rather than a restriction: the roll counts buildings, so the case counts buildings, and it never attempts to work out who owns a redacted property. Personal information is a question for Book H2 (Privacy and Personal Information), and this book explains where the rules stand and recommends no vendor. A business with a specific plan for the data it has read should put the plan to a lawyer.

3.7 Industry Reports, Associations and Trade Press

Industry reports, association surveys and the trade press are the sources a business reads when it wants to know what an industry thinks about itself, and a careful reader reads them for method. They are not dishonest by nature; they are produced by somebody with a purpose, and the purpose shapes what is measured, whom it is measured on and what is published. A report sold by a firm that also sells services to the industry has an author, a method and a purpose, and a business that reads the number without the purpose has read part of the source.

The first thing to read in a report is not the finding but the method: who paid for the report, whom it measured, how the measurement was made, when it was made and what was left out decide how much of it a business can use. A report that surveyed its own customers says nothing about a market, and a report whose method page is missing is a report whose numbers cannot be weighed at all.

The second thing to read is the license or the terms. A report that has been purchased is a report whose use is governed by the terms of its purchase, and those terms decide whether it can be copied inside the firm, quoted in a brief, or shared with a partner. The Copyright Act, R.S.C. 1985, c. C-42, s. 29, current to September 21, 2026, provides that fair dealing for the purpose of research, private study, education, parody or satire does not infringe copyright, and whether a given copy of an industry report falls inside that provision is a question for Book H5 (Intellectual Property), which alone answers it. Redistributing a purchased report inside or outside the firm is governed by the terms of its license, and those terms are read before the report is circulated.

The third thing to read is the trade press, and it is read for a limited purpose. A trade publication aimed at building owners and managers carries the names of the issues that the industry is talking about and the language it uses for them, which is useful for framing a question and finding the vocabulary a respondent will recognize, and it is not evidence of a market size. An article is one author's account of an event, and a business that treats an article as a count has confused a report with a record.

The case's own practice is posed by this book and it is deliberately thin. Nordec Entretien inc. buys no report, reads the trade press of building owners and managers only for the names of the issues, and records each reading in the source log with its date and its author. No figure from any report, no vendor study and no industry benchmark is printed in this book, because a report is its authors' own sample, method and claim, and a claim is not evidence until its method has been read.

3.8 Judging a Source: Who, How, When and for Whom

Four questions decide whether a source can be used, and they are asked in the same order every time. Who made it? How was it made? When was it made? For whom was it made? A number whose author cannot be identified is not a number that a brief can carry, and a business that cannot name the author of a figure has already learned everything it needs to know about the figure.

The second question is the one that separates a record from a claim. A record is produced by a process that can be described: a roll is compiled property by property for taxation, a register is compiled from filings, a tender system is compiled from notices published by public bodies. A claim is produced by a study whose design decides what it can support: a survey of a panel measures the panel, a survey of a customer list measures the customer list, and a report that does not say how it measured has not told the reader what it measured. The second question is also where the unit is settled. A count of locations, a count of establishments, a count of firms and a count of buildings are four different counts, and a source that does not say which one it produced has produced none that a business can use.

The third question is the date, and it has two parts. The date of the source is the period the source describes; the date read is the day the business opened it. Both are recorded, because a page can be rewritten between the two and because a business that re-reads a source a year later needs to know whether it is reading the same page. A source read on October 10, 2026 and describing December 2025 is a source about a period that has already passed, and a brief that carries it should say so.

The fourth question is the purpose, and it is easy to skip. A record produced for taxation is a good record of what it taxes and a poor record of what it does not, and a tender system produced so that businesses can find opportunities is a good record of what is being bought and a poor record of what was awarded, at least on the page read. The purpose is the boundary of what the source can support.

The research code that governs much of the profession says the same thing in its own language. Secondary data must fit the original purpose and respect contracts and intellectual property, and known or suspected biases are disclosed (International Chamber of Commerce & ESOMAR, 2025, arts. 5 and 8). The code is self-regulation, recognized by more than 60 associations in more than 50 countries, and it binds its members and the associations that adopt it rather than a business by law (International Chamber of Commerce & ESOMAR, 2025). It was read at its articles 1 to 8 for this book.

The table below sets out the four questions, what to look for in each, a warning sign that the question has not been answered, and the case's example for each. Read it as a checklist to be run before a source is used rather than after a brief is written, because a source that fails one of the four is a source whose number is removed from the brief rather than qualified in it.

Table 3.3 — Four questions to put to a source

3.9 Documenting Sources So the Work Can Be Redone

A source log is the record that makes a desk study repeatable, and it is the least glamorous and most valuable document a study produces. Its purpose is narrow and it is stated in one sentence: another person, opening the log a year later, should be able to find the same page, read it the same way and arrive at the same count or find the mistake. A log that does not permit that has recorded a reading rather than a source, and a reading cannot be checked.

The log has six fields, and each one answers a question that will be asked later. The table below sets out the fields, what each holds, why it matters and the case's entry for the property assessment roll. Read it as the shape of a single row: a log is a sheet in which each source is one row of these six fields, and the row is written when the source is read rather than when the brief is written.

Table 3.4 — The fields of a source log

The log is kept by a person rather than by a tool, and who keeps it changes with the size of the business, as the box below shows.

The exclusions field is easy to leave blank, and it is the one a reviewer reads first, because a row that says what was not taken prevents a later reader from treating a count of buildings as a count of owners or a count of locations as a count of firms.

METHOD Documenting a source in five steps so that the count can be redone

First, name the source and its author, in the words the source itself uses. Second, record the date read and the date or version of the source, so that a later reader knows which version of a page was opened. Third, record the license or the terms and the attribution statement the license requires, because reuse without the conditions is reuse without permission. Fourth, write what was taken, in what unit, and what was deliberately not taken, because the exclusions are what stop a later reader from overreading the count. Fifth, store the file or the page beside the log, so that the count can be redone from the same material rather than from a memory of it. This sequence is this book's own and no source tests it. The case's example is the property assessment roll: the source and its author, the catalog record and the municipal file, the date read, the metadata date, the attribution license, and the count of buildings in the size band with the record's fields left unread.

WHO DOES THIS The administrator keeps the source log and the owner reads it at each gate

Who keeps the log depends on the size of the business, and the work is the same in each. A solo owner keeps the log in a shared sheet and writes each row as the source is read, because a row written a week later is written from memory. A small team has the person who runs the study keep the log, and the owner reads it at each gate of the plan rather than at the end. An established business adds a reviewer who did not do the reading and who redoes one count from the log, which is the only test that shows whether the log is complete. In the case, the administrator keeps the log and Sophie Nadeau reads it at each gate, and the count that is redone is the count of buildings in the size band. No hour and no dollar is attached to this work in this chapter.

TOOLS OF THE TRADE What a business needs to keep its sources: a shared sheet, a folder of dated files, bookmarks and a calendar

Four functions carry a source log, and none of them is a specialist product. The first is a shared sheet with one row per source and the six fields of the log, reachable by the people who read it. The second is a place for the files that were read, kept with their dates in the file name, so that a count can be redone from the same material. The third is a set of bookmarks to the public sources the business reads regularly, so that the same page is opened rather than found again. The fourth is a calendar entry for re-reading the sources that are updated yearly, set by hand, so that a count is refreshed rather than carried forward. Chapter 12 gives the functions in full. No brand name and no price belongs on this list, and the requirement is the function rather than the product.

3.10 The Case's Desk Study

The desk study of Nordec Entretien inc. is posed by this book and it is illustrative, and it is read by gate 1 of the plan, on day 10. Nordec Entretien inc. is a fictional business created for this encyclopedia; its owner, its employees, its customers, its competitors, its figures and the events described are invented, and any resemblance to a real business is coincidental. The decision the study serves was written before any fieldwork: open the third territory only if its adjusted count is at least 400 buildings in the size band and the regional integrator won fewer than half of the public maintenance tenders seen there; otherwise deepen the two territories. The desk study exists to produce the two numbers that decision rule needs, and it produces them from public records rather than from conversations.

The first record read is the property assessment roll. The roll gives 1,512 buildings in the size band in the municipalities of the two territories, and a check of 50 records finds 44 eligible, which is 88 percent, so the top-down count is 1,331 (posed by this book). The same check on the third territory turns 470 raw buildings into 414 adjusted (posed by this book). The check is the part of the work that makes the count a working count rather than a raw count: a roll lists every property, and a size band applied to a roll catches properties that are not the ones the business means, so a sample of records is read one by one and the eligible share is applied to the whole.

The second record read is the register of enterprises, read for the two narrow purposes of section 3.4 (Gouvernement du Québec, 2025). It shows existence, status and addresses and no revenue, no customers and no prices, so it is used for what it holds and nothing else.

The third record read is the public tender system, and it is read for what public bodies buy. Four public tenders of the third territory are seen in the system, and their winners are read in Chapter 7 (posed by this book). The page read for this book carries public calls for tenders, notices of intent and notices of interest, and it does not carry awarded contracts, bidders' names or amounts, so a business should verify what the system shows on the day of use, as section 3.6 sets out. Book E9 (Tenders, RFPs and Public Procurement) treats finding opportunities in its chapter 2; this chapter reads the notices as a source and stops there.

The table below sets out the five steps of the case's desk study, the count each step produces, the kind of source it comes from and what the count is. Every figure in it is posed by this book and illustrative, and none of it is a measured result. Read it as the arithmetic of a market count rather than as a finding: the raw count comes from a roll, the check comes from a sample of records, and the working count is the raw count adjusted by the check.

Table 3.5 — The case's desk study, posed by this book

Figure 3.1 sets out the six public sources of the case's desk study, what each says and what it does not say, and shows how the property assessment roll was checked. The figure is drawn as six cards in two rows, each with a colored head, a line for what the source says and a dashed line for what it does not say, and an orange box below for the case's count. The business counts say how many active locations there are by industry and size, not how many distinct firms or buildings. The register of enterprises says names, addresses, directors and owners, not revenue or customers. The property assessment roll says each property, its use and its size, not the owner, which is redacted and never re-identified. The public tender notices say what public bodies buy and when, not the awards, at least on the page read. The open licenses allow reuse with attribution and do not cover personal information. Industry reports say what their authors measured and are not a fact until their method has been read. Below the cards, the orange box carries the case's count: 1,512 buildings in the size band, 44 of 50 records eligible, which is 88.0 percent, and a working count of 1,331.

Figure 3.1 — Six public sources and their limits. Business counts, the register of enterprises, the property assessment roll, public tender notices, open licenses and industry reports each say something and leave something unsaid. In the case, the roll gives 1,512 buildings in the size band; a check of 50 records finds 44 eligible, 88.0 percent, and the working count is 1,331.

The desk study finishes with a count and a set of limits, and the limits are written down with the count. The count is a count of buildings and not of owners, because the roll redacts the owners and forbids re-identifying them (Ministère des Affaires municipales et de l'Habitation, 2026). It is a count of buildings in a size band and not of buildings that need the service, because a size band is a proxy for the kind of building the business serves. It is a working count and not a measured one, because it is posed by this book and the check of 50 records is a sample rather than a census. The count of tenders is a count of notices seen and not of contracts awarded, because the page read does not carry the awards.

The desk study also tells the business what to ask people, and that is the part of it that pays for the reading. The roll says how many buildings of the size band stand in the two territories; it does not say how many of them are managed by a property management firm, how many are under a bundled facility contract, or how many of their owners are thinking about changing a provider. Those three questions are what the interviews of Chapter 4 and the survey of Chapter 5 are for, and they are questions that no public record can answer. The desk study has done its work when it has produced the count, the limits and the three questions.

DO THE MATH The case's roll gives 1,512 buildings, a check of 50 records leaves 88 percent, and the working count is 1,331

Illustrative figures; the case is fictional. Step 1 — The property assessment roll gives 1,512 buildings in the size band in the municipalities of the two territories, posed by this book. Step 2 — A check of 50 records, read one by one, finds 44 eligible, posed by this book. Step 3 — 44 divided by 50 is 88 percent, which is the eligible share of the records checked. Step 4 — 1,512 times 88 percent is 1,331, rounded to the nearest building, which is the working count for the two territories. Step 5 — The count is a working count of buildings and not of owners, because the roll redacts the owners and forbids re-identifying them, and the whole computation is posed by this book and illustrative. Nothing is credited to the study, and the decision rule decides.

CASE IN POINT Nordec Entretien inc. counts the buildings of its size band from the roll and checks 50 records

Nordec Entretien inc. is a fictional business created for this encyclopedia; its owner, its employees, its customers, its competitors, its figures and the events described are invented, and any resemblance to a real business is coincidental. The administrator reads the property assessment roll, the register of enterprises and the public tender notices, and keeps the source log with its six fields, one row per source. Sophie Nadeau reads the log at gate 1, on day 10, and the count that is redone from the log is the count of buildings in the size band. The roll gives 1,512 buildings in the size band in the municipalities of the two territories; a check of 50 records finds 44 eligible, which is 88 percent; the working count is 1,331; and the third territory turns 470 raw buildings into 414 adjusted. Four public tenders of the third territory are seen in the tender system, and their winners are read in Chapter 7. Every one of those figures is posed by this book and illustrative, and none of them is a measured result.

IN LA MACHINE La Machine keeps competitors as company records and public sources as bookmarks, and its Business Explorer is not a market count

Three pages of the application copy read for this book carry part of the desk study, and none of them counts a market (La Machine, 2026). CRM → Companies business/marketing/companies holds one record per company, with its sector, its size, its site and notes, and it has no field for a competitor's price, so a competitor profile is kept as dated notes on a company record. Reseller → Business Explorer business/marketing/business-explorer, reserved to the administrator and reseller roles, searches public business listings, stages them with their industry classification code and their source, and is a tool of prospecting and never a count of a market, and Book B1 (Marketing Fundamentals for Business Owners), chapter 3, sends the reader to this book on that point. Productivity → Links /liens keeps the team's bookmarks to the public sources, dated by the team. La Machine holds no market data, and the case does not count its market in the application; check it in your installation. La Machine is described, never sold.

Costly Mistakes

• Booking the interviews before reading anything. A study that begins with conversations spends its scarcest hours asking questions that a roll, a register or a statistics release would have answered, and it arrives at the last interview with the understanding it had at the first.
• Reading a count of employer locations as a count of firms or of buildings. The counts published by the national statistics agency are of businesses active on the Business Register, usually counted by location, and a business that reads them as a market count has read a good number as a different number.
• Quoting a number without its unit, its date and its author. A count of locations, a count of establishments, a count of firms and a count of buildings are four different counts, and a figure that does not say which one it is cannot be checked by anybody.
• Using a source outside the purpose it was produced for. A register produced so that the public can identify businesses shows existence and addresses and no revenue, and a tender system read for awards will not carry them on the page read.
• Reusing open data without reading its license. An open license is a permission with conditions: the attribution statement is required, personal information is excluded, and a provincial or municipal portal has its own license rather than the federal one.
• Trying to work out who owns a redacted property in the assessment roll. The catalog record prohibits re-identifying the redacted owners' names and addresses, and a market count does not need them, because the roll counts buildings.
• Keeping no source log, or writing it a week after the reading. A log written from memory cannot be checked, and a count that cannot be redone is a count that a reviewer has to take on trust.

Key Takeaways

• Secondary research comes first because it is the cheapest evidence a business can obtain and because it tells the business what to ask people, and a desk study is the reading of public sources before any fieldwork.
• A source is evidence only once its author, its design, its date and its purpose are known, and the four questions of who, how, when and for whom are asked before a number enters a brief.
• The national statistics agency counts businesses active on the Business Register, usually by location, and the provincial institute publishes by theme at the region and the regional county municipality; neither count is a count of firms or of buildings.
• The industry classification is a six-digit, production-oriented vocabulary for finding published counts, and an industry profile shows an industry group, its establishments and its year rather than a count of the businesses a reader has in mind.
• The property assessment roll counts buildings and redacts its owners, the register of enterprises shows existence, status and addresses, and the public tender system carries notices rather than awards; each is read for what it holds and no more.
• A source log with six fields — source and author, address or file, date read, date or version of the source, license or terms, and what was taken and what was not — is what lets another person redo the count or find the mistake.

Your Action Plan

1. Write the four questions into the top of a sheet before reading anything. Open a sheet with the columns of the source log, and put the four questions — who made it, how, when and for whom — at the top of the page. Deliverable: one sheet with six columns and four questions. Verification: every row later written into the sheet has an answer to all four questions or the source is not used.
2. Read the definitions page of the first statistics release you use. Before quoting any count from a statistics agency, read the definitions of the unit, the period and the population the release uses, and write them beside the count. Deliverable: one count with its unit, its date and its population in the same sentence. Verification: a reader who has not seen the release can say what the count measures and what it does not.
3. Find the class of your activity in the current version of the classification. Open the current version, search the activity by its ordinary words, read the class definition and its exclusions, and record the version you used beside the code. Deliverable: one code with its version and its exclusions. Verification: the code was checked in the version used, and the exclusions were read as well as the definition.
4. Record the license of every source as you open it. For each source, write the license or the terms and the attribution statement it requires into the log, in the same row as the date read. Deliverable: a license field filled for every source read. Verification: any figure published from the log carries the attribution statement its license requires.
5. Count your market from the roll and check the count against a sample of records. Apply your size band to the roll, read a sample of records one by one, compute the eligible share, apply it to the raw count, and write the working count with its limits beside it. Deliverable: one working count of buildings with its limits. Verification: another person can redo the count from the log and the stored files and arrive at the same number.

In La Machine

Three pages of the application copy read for this book carry part of a desk study, and the reader keeps the rest outside the product (La Machine, 2026). CRM → Companies business/marketing/companies holds one record per company, with its sector, its size, its site and notes, so a competitor profile is kept as dated notes on a company record. Reseller → Business Explorer business/marketing/business-explorer, reserved to the administrator and reseller roles, searches public business listings and stages them for prospecting; it is never a market database and never a count of a market. Productivity → Links /liens keeps the team's bookmarks to the public sources, dated by the team. Control center → Workflows /registre is the register of automated workflows and not a register of enterprises; the enterprise register of Québec is read on its public site (Gouvernement du Québec, 2025).

What the platform does not do belongs in the same breath, because a business that expects a desk study to be built into a product will not build it. La Machine holds no market data, counts no market and forecasts no demand. The source log, the files read, the licenses and the counts live outside the product, in a sheet and a folder that the reader keeps and dates. Check it in your installation, because only an installation can tell what a given copy of the application shows. La Machine is described, never sold, and the case runs its survey outside it.

Self-Check

1. Why does secondary research come first in a study, and what does a desk study produce besides a count?
2. What does the national statistics agency count, and what does its release say it does not support?
3. What are the five levels of the industry classification, and why does the version of the classification matter?
4. What does a public register of enterprises show, and what does it not show?
5. What are the six fields of a source log, and what is the log for?

Answers. 1. Reading comes first because it is the cheapest evidence a business can obtain and because it tells the business what to ask people; a desk study produces a working count with its limits attached and a short list of questions that no public record can answer, which is what the interviews and the survey are for. 2. The agency counts businesses active on the Business Register, compiled mainly from Canada Revenue Agency records, usually counted by location, with employer status depending on payroll remittances in the previous 12 months; the release asks for cautious interpretation of period-to-period changes and says nothing about the number of distinct firms, openings or closures. 3. The levels are the sector at digits 1 and 2, the subsector at digit 3, the industry group at digit 4, the industry at digit 5 and the Canadian industry at digit 6, with a zero meaning no further national detail; the version matters because classifications are revised, and a code read under one version and a count published under another are not comparable without saying so. 4. The register shows an enterprise's number, its registration status, the names it uses, the addresses of its establishments, the contact details of its directors and officers and the names of its ultimate beneficial owners; it shows no revenue, no customers and no prices. 5. The six fields are the source and its author, the address or file, the date read, the date or version of the source, the license or terms, and what was taken and what was not; the log exists so that another person can find the same page, read it the same way and arrive at the same count or find the mistake.

Chapter References

Copyright Act, R.S.C. 1985, c. C-42, s. 29 (Canada). Justice Laws Website, current to September 21, 2026 (last amended November 7, 2024) — read October 10, 2026. https://laws-lois.justice.gc.ca/eng/acts/C-42/section-29.html

Gouvernement du Québec. (2025). Rechercher une entreprise au registre [Registraire des entreprises; in French]. (Updated September 23, 2025.) Retrieved October 10, 2026, from https://www.quebec.ca/entreprises-et-travailleurs-autonomes/obtenir-renseignements-entreprise/recherche-registre-entreprises/acceder-registre-entreprises

Gouvernement du Québec. (2026). Accéder au Système électronique d'appel d'offres (SEAO) [In French]. (Updated August 31, 2026.) Retrieved October 10, 2026, from https://www.quebec.ca/gouvernement/faire-affaire-gouvernement/outils/systeme-electronique-appel-offre

Government of Canada. (n.d.). Open Government Licence – Canada (Version 2.0). (Page modified December 2, 2022.) Retrieved October 10, 2026, from https://open.canada.ca/en/open-government-licence-canada

Innovation, Science and Economic Development Canada. (n.d.). Other professional, scientific and technical services – 5419: Businesses [Canadian Industry Statistics]. Retrieved October 10, 2026, from https://ised-isde.canada.ca/app/ixb/cis/businesses-entreprises/5419

Institut de la statistique du Québec. (n.d.). Statistics at the service of society [Home page]. Retrieved October 10, 2026, from https://statistique.quebec.ca/en

International Chamber of Commerce & ESOMAR. (2025). ICC/ESOMAR international code on market, opinion and social research and data analytics. ESOMAR. Retrieved October 10, 2026, from https://community.esomar.org/uploads/public/knowledge-and-standards/codes-and-guidelines/ICCESOMAR-International-Code_English.pdf

La Machine. (2026). Documentation — route reference inventory. Internal product documentation, read October 10, 2026, together with the code of the pages this book describes: business/marketing/companies, business/marketing/business-explorer, /liens and /registre.

Ministère des Affaires municipales et de l'Habitation. (2026). Rôle d'évaluation foncière du Québec [Data set; in French]. Données Québec. (Metadata modified August 24, 2026; 2026 files created March 3 and April 1, 2026; licence CC BY 4.0.) Retrieved October 10, 2026, from https://www.donneesquebec.ca/recherche/dataset/roles-d-evaluation-fonciere-du-quebec

Statistics Canada. (2017). North American Industry Classification System (NAICS) Canada 2017 Version 3.0. Retrieved October 10, 2026, from https://www23.statcan.gc.ca/imdb/p3VD.pl?Function=getVD&TVD=1181553

Statistics Canada. (2022). North American Industry Classification System (NAICS) Canada 2022 Version 1.0. Retrieved October 10, 2026, from https://www.statcan.gc.ca/en/subjects/standard/naics/2022/v1/introduction

Statistics Canada. (2026, February 13). Canadian business counts, December 2025. The Daily. Retrieved October 10, 2026, from https://www150.statcan.gc.ca/n1/daily-quotidien/260213/dq260213c-eng.htm

Chapter 4 — Interviews and Focus Groups

A business that wants to know why people choose it, stay with it or leave it has to talk to them, and this chapter shows what a conversation can and cannot establish, how to choose between an individual interview and a group discussion, how to recruit the right people including those who said no, how to write a guide and ask about what people did rather than what they intend, how to run a session and take notes, how many conversations are enough, how to code them into themes, and what consent and recording owe the participants.

What You'll Be Able to Do

• Decide whether a decision needs individual interviews, a group discussion or neither, and say in one sentence what each of them can and cannot establish.
• Recruit participants by group, record the refusals with their group and their reason, and keep every interviewee out of the prospect list.
• Write an interview guide of four to six topics, turn each topic into open questions, and cut every question that suggests its own answer.
• Ask about the last time something happened rather than about what a person would do, and recognize the moment a stated intention is being read as a plan.
• Run a session that opens with the purpose and the consent, take notes during it, and expand those notes within the hour.
• Code the notes into short labels and themes, count how many conversations hold each theme, and say where the new codes stopped arriving.

Why It Matters

A count tells a business what happened. It does not tell the business why. A business that knows it lost three buildings this year knows the size of a problem and nothing about its cause, and a business that knows why those three buildings left has something it can act on. The why lives in the words people use, and the only instrument that reaches those words is a conversation. That is the whole reason this chapter exists, and it is also the reason the chapter spends most of its length on discipline rather than on technique: a conversation is easy to hold and hard to hold honestly, and a business that treats a friendly chat as evidence will build a decision on a story it told itself.

The cost of not talking is paid in assumptions that feel like knowledge. A business that loses a customer to a competitor may have an explanation ready, and the explanation can be about price, which is comfortable because it requires nothing to change except a number. When Nordec Entretien inc., the fictional business followed through this book, sat down with fourteen people in the posed interviews of this chapter, the themes that came back were not about cutting the price: they were about proof of response time, about wanting one contact rather than a bundle, about planning equipment replacement, and about price mattering only after proof. A business that had acted on its own explanation would have cut a price and learned nothing.

The third reason is that conversations and counts are not rivals. They answer different questions and a design needs both. A conversation shows the reasons and supplies the words; a survey counts how many people hold those reasons and gives a margin that a conversation can never give. The conversations come first, because they supply the words and the options that the survey will later put in front of a sample.

In This Chapter

Section 4.1 sets out what a conversation can and cannot establish and draws the boundary with the books beside it. Section 4.2 compares individual interviews with group discussions and explains why neither is a sample of a market. Section 4.3 covers recruiting by group, recording the refusals and keeping interviewees out of the prospect list. Section 4.4 covers the interview guide, its topics and its open questions. Section 4.5 covers the difference between asking about past behavior and asking about future intentions. Section 4.6 covers the conduct of the session and the taking of notes. Section 4.7 asks how many conversations are enough and reads the case's saturation curve. Section 4.8 covers coding, themes and the analysis grid. Section 4.9 covers consent, recording and the respect owed to participants. Section 4.10 sets out the case's fourteen conversations and what they cost.

4.1 What Conversations Can and Cannot Establish

A conversation is the only research instrument that can tell a business why a person did something. A count of departures tells a business how many customers left; it does not tell the business what the customer was thinking on the day the decision was made. A conversation reaches that, and it reaches the words the person uses, which are the words a message will later have to use if it is to be understood. What a conversation cannot do is tell the business how many people think the same thing. Fourteen conversations, or forty, are a set of people who agreed to talk, and the proportion of a market that shares their reasons is a question for a survey, not for a conversation.

This distinction is the whole of the chapter. A conversation establishes what people say they did, why they say they did it, and the vocabulary they use to describe it. It does not establish how many people think so, and no margin of error can be printed beside it, because there is no probability sample behind it. A business that reads fourteen conversations as if they were a poll will make a decision on a number that does not exist. A business that reads them for what they are — a set of reasons, in the customers' own words, from people who agreed to speak — has evidence that no count can supply.

The questions a conversation can answer are the questions that begin with why and how. Why did the last provider change happen? How was the decision made, and who made it? What was looked at before the change, and what was ruled out? What would have had to be true for the decision to go the other way? These are questions about a history, and a history is what a person can describe. The questions a conversation cannot answer are the questions that begin with how many, how much and how often, and those belong to Chapter 5, where a frame, a sample and a margin are built for them.

There is a second limit, and it is about the person rather than the question. People describe their own past imperfectly. A decision made two years ago is remembered through its outcome, and a customer who is satisfied with the current provider will describe the change as obvious, while a customer who regrets it will describe the same change as rushed. The interview does not remove this; it records it. A business that reads a set of conversations reads a set of accounts, and the accounts are the evidence, not the truth behind them. A business that wants to know how a decision was actually made has to ask about the sequence of events rather than about the reasons the person now gives for them, and section 4.5 returns to this.

The boundary between this chapter and the books beside it is worth drawing before the method begins, because a business that confuses the kinds of conversation will hold the wrong one. The table below sets out five neighbors, what each of them holds, and what this chapter holds instead. Read it as a map of the conversations a business can have: the discovery interview of a venture with no customers, the interview with the business's own customers, the use of the customers' words in a message, and the rules that govern the personal information collected along the way.

Table 4.1 — Where this chapter stops and its neighbors begin

The first row matters most in practice, because the two kinds of interview look identical and are not. A venture with no customers interviews to find out whether a need exists at all, and Book A1 (From Idea to Business Model) holds that work in section 3.2. A business with customers interviews to find out why a need is being met the way it is, by whom, and at what moment the choice was made. Book B10 (Customer Experience, Retention and Loyalty) holds the interviews with the business's own loyal, new and lost customers in section 4.6, and this chapter holds the interviews that reach beyond them, to people who have never bought and to people who buy from someone else.

The literature on this stage of research is thinner than a reader might expect, and one paper is worth naming for what it asks rather than for what it found. The abstract of a study of the customer-needs stage of quality function deployment reports that the paper asks how many customers to interview, how many analysts should read the transcripts, which needs analysts miss, and whether focus groups or one-on-one interviews are better (Griffin & Hauser, 1993). The abstract states the questions and not the answers, so this book prints no number of interviews and no verdict on focus groups from it. That is the honest position: the questions are the right questions, and the answers are not in the part of the paper that was read.

DEFINITION The words this book uses for an individual interview, a group discussion, an interview guide, saturation and a code

An individual interview is a conversation between one researcher and one person. A group discussion is a conversation led by a moderator with several people at once. An interview guide is the list of topics and open questions that a conversation follows. Saturation is the point at which new conversations stop bringing new information, as the studies of section 4.7 use the word. A code is a short label given to an idea in the notes. These are this book's own working words, chosen so that a reader can tell the five things apart in a single sentence, and they are not definitions taken from a standard or a law. A business that uses other words for the same five things is doing the same work, provided it keeps the distinctions: one person or several, a plan or a script, a point at which the new stops arriving, and a label that stands for an idea.

4.2 Choosing Between Individual Interviews and Group Discussions

The choice between an individual interview and a group discussion is a choice about what the decision needs, and it is made before any participant is recruited. An individual interview gives one person the whole of the time and the whole of the attention, and it is the instrument that reaches a history in sequence: what happened first, what was looked at next, who was consulted, and what settled it. A group discussion gives people the chance to react to one another, and it is the instrument that reaches a topic rather than a history: what a group of people think about a service, a price or a category, and how they talk about it when somebody else is in the room. Neither is better in general, and this book does not rank them.

The reason the two are not ranked is that the question of which is better has been asked and the part of the literature read does not answer it. The abstract of the study of the customer-needs stage reports that the paper asks whether focus groups or one-on-one interviews are better (Griffin & Hauser, 1993), and the abstract states the question rather than the answer. A business that wants a verdict will not find one here, and it should not look for one, because the answer depends on what the decision needs. A decision about why a specific customer left is a decision about a history, and a history is best reached one person at a time. A decision about how a category of buyers talks about a service is a decision about a topic, and a topic can be reached with several people in a room.

The group discussion has a property that the individual interview does not have and a defect that follows from it. The property is interaction: one participant's account prompts another's, and a business learns what people say to each other rather than only what they say to an interviewer. The defect is that the room has a shape. The loudest voice shapes it, the most confident account becomes the reference point, and a participant who disagrees with the direction of the conversation may simply stop contributing. A moderator can reduce this and cannot remove it, and a business that reads a group discussion as a set of independent opinions will overstate the agreement in the room.

The individual interview has the opposite profile. It reaches depth and it loses interaction. One person can describe a decision over forty-five minutes without being interrupted or corrected, and the account that comes out is longer and more specific than anything a group would produce. What it loses is the corrective that other people supply: in a group, a participant who misremembers a date or a price may be corrected by somebody else, and in an individual interview nobody is there to correct anything. A business that reads a set of individual interviews reads a set of accounts that have not been checked against each other.

Neither instrument is a sample of a market, and this is a point at which a business can go wrong. A group discussion of eight people is not a survey of eight people, and an individual interview is not a survey of one. No margin of error is printed for either, because neither rests on a probability sample, and a business that wants a number has to build one in Chapter 5. What the two instruments supply is the material a survey needs: the reasons, the sequence and the words. A survey built without them asks questions the business invented, and the answers come back in a vocabulary the business chose.

The table below sets out the two instruments side by side, with five questions and one column for each instrument. Read it as a choice about what the decision needs rather than as a ranking, because this book does not rank the two, and read the last row as the case's own choice rather than as a general preference.

Table 4.2 — Individual interviews and group discussions compared

The case, posed by this book, chooses individual interviews. The decision the research serves is where next year's growth should go, and the reasons behind a building's maintenance decision are made by one person for one building: a property manager deciding for a portfolio, or an owner deciding for a building that belongs to that owner. A group discussion would put several of those people in one room and would produce a conversation about the category, which is useful and is not what the decision needs. The case therefore holds fourteen individual conversations, and the counts and the groups are set out in section 4.3 and section 4.10.

KEY CONCEPT Conversations show the reasons and the words, a survey counts them, and a design needs both

The central idea of this chapter is that conversations and counts answer different questions and that a research design needs both. Individual interviews and group discussions are chosen for what the decision needs, and neither is a sample of a market: no margin of error is printed for either, because neither rests on a probability sample. The abstract of the study of the customer-needs stage asks which of the two is better and does not answer the question in the part read (Griffin & Hauser, 1993), so this book offers no verdict and no ranking. What a conversation supplies is the reason, the sequence and the vocabulary; what a survey supplies is the count and the margin. The order matters, because the conversations come first and give the survey its options and its words. This reading is this book's own and is never presented as a finding of a source.

4.3 Recruiting the Right People, Including Those Who Said No

Recruiting begins with a decision about groups, not with a list of names. A business that recruits whoever is willing will end up with a set of people who are easy to reach, and the easy-to-reach may be the satisfied and the nearby. The case, posed by this book, recruits by four groups and holds fourteen conversations: six property managers, four owners who run their own building, two former customers and two customers who renewed. The groups are chosen because the decision needs them: the property managers and the owner-operators are the two kinds of buyer the business sells to, and the former and renewed customers are the two ends of the retention question.

The people who said no are part of the record. A business that counts only the conversations it held has a record of the willing, and the people who declined are evidence about the people the business does not reach. The case, posed by this book, records each refusal with its group and its reason in the research file, and the record is kept beside the notes rather than in a separate place. A refusal is not a failure of the recruiting; it is a fact about the population. No share of refusals is printed here as a norm, because the case counts its own and no source in this book gives a typical figure.

The people who are recruited are volunteers, and a set of volunteers is not a sample of a market. This is the sentence that keeps the chapter honest. A volunteer is somebody who agreed to give forty-five minutes to a business, and the reasons for agreeing are not the reasons of the market: some people agree because they are satisfied, some because they are angry, some because they know the person who asked. A business that reads fourteen volunteers as a cross-section of its market will overstate whatever the volunteers had in common. What the volunteers supply is the range of reasons and the vocabulary, and the range is what the survey will later test against a sample.

The people who are recruited are never added to the prospect list. Their answers stay in the research file, and nobody is put on a mailing list or called back with an offer because of an answer. This is not a courtesy; it is the line between research and selling, and the international code read for this book keeps the two apart. The code requires researchers to keep a clear distinction between research and non-research activities and to inform data subjects of any non-research use, whether promotional, commercial or customer follow-up, before collection and to obtain separate consent for it (International Chamber of Commerce & ESOMAR, 2025, art. 1(d)-(e)). A business that interviews a property manager and then adds that person to a campaign list has turned a research conversation into a sales approach without saying so, and the code read here forbids it. The rules on personal information and consent are Book H2 (Privacy and Personal Information)'s, and the rules on electronic messages and calls are Book H3 (Anti-Spam and Telemarketing Compliance)'s.

The people who recruit say who is calling and for whom. A call that begins with a question and does not say who is asking is a call the recipient cannot place, and a recipient who cannot place the call is being asked to give time to an unknown purpose. The case, posed by this book, has the administrator keep the recruiting log, and the log records the group, the date of the approach, the outcome and, where the person declined, the reason given. The log is what makes the refusals readable and what makes the recruiting repeatable a year later.

WHO DOES THIS The owner and the lead technician talk to the field, and the administrator keeps the recruiting log

Who recruits and who talks depends on the size of the business. A solo owner recruits and talks, with a note-taker where one can be found, because a conversation that is both held and noted loses part of both. A small team has one person recruit and two take part in the session, one asking and one noting, which is the arrangement the case uses. An established business adds a recruiter or a fieldwork service, and the fieldwork service says who is calling and for whom. In the case, Sophie Nadeau and the lead technician talk to the field, and the administrator keeps the recruiting log with its groups, its dates, its outcomes and its reasons for refusal. No hour and no dollar is printed here, because the case counts its own time in section 4.10 and the rates belong to Book B5 (Content Marketing and the Content Factory).

4.4 Writing an Interview Guide

An interview guide is a short list of topics and open questions, and it is not a script. The difference matters more than it sounds. A script is read aloud, and a person who is being read to answers the questions that were written rather than the questions that matter. A guide is held in the hand and glanced at, and the interviewer is free to follow an account where it goes and to return to the next topic when the account has finished. A guide of four to six topics is enough for a forty-five-minute conversation, and a guide of twenty questions is a questionnaire being read aloud, which is a different instrument and belongs to Chapter 5.

The guide opens with the purpose and the consent, and it moves from the recent past to the present. That order is deliberate. A person who is asked to describe a decision made two years ago will reconstruct it, and the reconstruction will be shaped by the outcome. A person who is asked first about the last month and then about the last change describes the recent in detail and the older in outline, and the detail is where the evidence is. The guide closes by asking what the interviewer has not asked, which is a question that can produce an account the guide did not anticipate.

The guide is also where the customers' and competitors' words are collected for later use. Book B6 (Copywriting and Persuasive Communication) holds the use of those words in a message, and this chapter holds how to hear them. The guide therefore asks people to say, in their own words, what they looked for, what they tried and what decided them, and it does not offer the words for them. A guide that asks whether a service was "responsive" has supplied the word and will get it back; a guide that asks what happened after the call was made will get the account, and the word for it will come from the person.

The case's guide, posed by this book, has four topics. The first is how the building's maintenance is decided today: who decides, what is looked at, and what the current arrangement covers. The second is the last time the provider changed: what prompted it, what was looked at, who was consulted and what settled it. The third is what proof of response time looked like: how the person knew a call had been answered, and what record existed. The fourth is how a price was weighed: what was compared, in what order, and what the price was weighed against. The four topics are the four themes the coding later produced, which is not a coincidence: a guide built from the decision's questions tends to produce the themes the decision needs.

No question in this chapter is printed as a quotation of a participant, and none is invented as one: the participants' words are the participants', and the case gives themes, codes and counts rather than words. The moment a quotation is invented to illustrate a theme, the theme stops being evidence.

METHOD Writing an interview guide in five steps

First, write the purpose in one sentence and the decision the answers serve, because a guide written without a decision behind it collects interesting material that nothing will use. Second, list four to six topics, ordered from the recent past to the present, so that the detail is collected before the reconstruction begins. Third, turn each topic into open questions that start with how, when or what, and cut every question that can be answered with a yes or a no. Fourth, test the guide with one colleague and cut every question that suggests an answer, because a question that offers a word will get the word back. Fifth, have the owner approve the guide and rewrite it after the first two conversations if a topic does not work. The sequence is this book's own and no source tests it; the case's four topics are the example, and the guide is approved before the first call is made.

4.5 Asking About Past Behavior Rather Than Future Intentions

People describe what they did better than they predict what they will do, and a guide built on the second will collect promises. The question "tell me about the last time" produces an account with a date, a sequence and a decision in it. The question "would you switch provider if the price were lower" produces an answer that costs the person nothing and commits the person to nothing, and a business that reads that answer as a plan has built a forecast on a courtesy. The whole of this section is the difference between those two questions, and the difference is not a matter of politeness or of technique; it is a matter of what the answer can carry.

The idea behind the preference is stated at the level that Book A6 (Designing a Value-Added Offer) prints it. Customers hire a product to make progress in a given circumstance, and the job, not the customer's attributes, explains the choice (Christensen et al., 2016). The source is reused from Book A6 and was not re-read by this book, and it is printed here only at that level. What the idea implies for a guide is that the circumstance is the thing to ask about: not what kind of buyer the person is, but what was happening when the decision was made. A guide that asks about the circumstance collects the job; a guide that asks about the person collects a category.

The practical form of the rule is a small set of question openings. "Tell me about the last time" is the first. "What happened next" is the second, and it is what turns an account into a sequence. "What did you look at before that" is the third, and it reaches the alternatives that were considered and rejected. "Who else was involved" is the fourth, and it reaches the decision that was not made by the person in the room. Each of these asks about something that has already happened, and each of them can be answered with a fact rather than with an intention.

The trap has a recognizable shape. A question that starts with "would you" invites a promise, and an answer to that question is a statement about a future that has not happened. The signs that a business has fallen into the trap are easy to list: the guide contains questions that begin with "would you", "do you think you might" or "how likely are you to", and the notes from the session are being read as a plan. The remedy is to replace the question with one about the last time the thing happened and what came of it. A person who says the provider would be switched if the price were lower has said nothing about the future; a person who describes the last switch, what prompted it and what it cost has described a decision, and a decision that happened once can happen again.

Stated intentions overstate, and Chapter 8 returns to this when it builds a forecast. The case, posed by this book, keeps a posed fraction of the stated switchers rather than treating them as sales, and the discount is posed by this book rather than taken from a source. The reason for the discount is the one this section gives: an intention is cheap to state and a switch is expensive to make, and the gap between the two is where a forecast built on intentions goes wrong. A business that wants a number has to build a survey and read it honestly, and a business that wants a reason has to ask about the last time.

WATCH OUT A question that starts with would you invites a promise, and a promise is not a plan

The trap of this section is asking an interviewee whether that person would switch provider or pay more, and then reading the answer as a plan. The signs are a question that starts with "would you" and an answer that is written into the notes as a decision. The remedy is to ask about the last time a provider changed and what happened, which is the idea of the circumstance and the job at the level Book A6 (Designing a Value-Added Offer) prints it (Christensen et al., 2016). Stated intentions overstate, and Chapter 8 returns to the point when it builds a forecast; the case keeps a posed fraction of the stated switchers rather than treating them as sales. The reasoning in this callout is this book's own and is not a finding of any source, and no figure is printed beside it.

4.6 Conducting the Session and Taking Notes

The session opens by saying who is asking, for whom, for what, that taking part is voluntary and that the person may stop at any time. Those five things take under a minute and they change the conversation. A person who knows who is asking and why answers differently from a person who has been invited to "a quick chat about the service". A person who knows that stopping is allowed is free to decline a question, and a declined question is information: it says the topic is sensitive, which is worth knowing. The opening is also where the recording is agreed or not agreed, and section 4.9 sets out the rule that governs it.

The session records nothing without a consent given before the recording starts and noted. That sentence is the whole of the recording rule at this level, and it is applied before the first question rather than after it. A business that starts a recorder and asks afterwards has recorded without consent, and the recording is not evidence the business can use. The consent is noted, with its date, in the research file, and the note says what was consented to: a recording, a set of notes, or both. The rules on personal information and consent are Book H2 (Privacy and Personal Information)'s, and the international code read for this book states the same requirement as self-regulation, as section 4.9 sets out.

Notes are taken during the session by a second person where possible, and they are expanded within the hour. The second person matters because the interviewer cannot both follow an account and write it down, and an interviewer who tries will do both badly. The expansion matters because memory decays quickly: the phrases a person used are available for an hour or so and are replaced afterwards by the interviewer's paraphrase. The case, posed by this book, allows forty-five minutes of conversation and fifteen minutes of notes for each session, and the fifteen minutes are the expansion rather than the writing during the call.

A transcription tool produces a draft, and a person checks every quotation against the notes. This is the rule that keeps a tool in its place. A draft produced by a transcription tool contains errors, and the errors are concentrated in exactly the places that matter: names, numbers, technical terms and the words a person used for a thing the business cares about. A person reads the draft beside the notes and corrects it, and no quotation leaves the research file until it has been checked. The same rule applies to any assistant that drafts a summary or a first set of labels: the draft is a draft, and a person reads it before anything is reported.

The session ends with the closing question and with a word about what happens next. The closing question is the one the guide reserves for the end: what has not been asked. The word about what happens next is short and honest: the answers go into a research file, they are read with the other conversations, and they are not used to sell anything to the person who gave them. That last sentence is not a courtesy; it is the line between research and selling, and the code read for this book requires the two to be kept apart, as section 4.3 sets out.

TOOLS OF THE TRADE What a business needs to run and keep interviews: a way to schedule, a consented recorder, a checked draft and a locked folder

The functions a business needs for this chapter are five, and none of them is a specialist product. The first is a way to schedule a meeting and keep its date and its notes, so that the session has a place in the calendar and a record beside it. The second is a recorder used only with consent, and the consent is noted before the recording starts. The third is a transcription tool whose draft a person checks against the notes, quotation by quotation, before anything is used. The fourth is a shared sheet for the codes, where the labels and the themes are kept beside the interviews that hold them. The fifth is a locked folder for the research file, holding the notes, the consents, the recruiting log and the refusals. Chapter 12 gives the functions in full, and no brand name and no price belongs on this list.

4.7 How Many Conversations Are Enough

There is no number of conversations that is always enough, and this book prints no typical number. The question is answered by the material rather than by a target: a business stops when new conversations stop bringing new information, which is what the studies read for this chapter call saturation. That point is different for every study, because it depends on the question, the population and how much the business already knows. A business that has run the study before will reach it sooner than one asking the question for the first time.

Two studies read at their abstracts give the shape of the answer without giving a rule. The abstract of one study reports that guidelines for non-probabilistic sample sizes are virtually nonexistent, that purposive samples rely on saturation, the point at which no new information or themes appear, and that in sixty in-depth interviews with women in two West African countries, saturation occurred within the first twelve interviews, with basic elements of metathemes present as early as six (Guest et al., 2006). That is one study of one population, and it is printed here as one study of one population. The abstract of a second study reports that in twenty-five in-depth interviews in health research, code saturation, when the range of thematic issues has been identified, was reached at nine interviews, but that sixteen to twenty-four interviews were needed for meaning saturation, a richly textured understanding (Hennink et al., 2017). The years printed are the years of the volumes, 2006 and 2017.

The distinction the second study draws is easy for a business to miss. Code saturation means having heard it all: every label that the material will produce has appeared. Meaning saturation means understanding it all: the labels have been explored far enough that the business knows what each of them means in the lives of the people who hold it. A business that stops at code saturation has a list of themes and no understanding of any of them, and a list of themes is not yet a finding. The case, posed by this book, reaches code saturation at interview nine and does not reach meaning saturation, which is why the chapter says plainly that hearing every code is not understanding it.

The case's own curve is set out in Figure 4.1 and in the table of section 4.10. Figure 4.1 draws the case's saturation curve: fourteen bars falling from nine new codes in the first interview to zero, a line rising to a plateau at thirty codes, and two dashed marks at nine and twelve interviews with a note in amber. The two marks are the findings of two studies and not rules for the case: the nine is where one study found code saturation in health research, and the twelve is where another found saturation in interviews with women in two West African countries. The case's last new code appears at interview nine, which is a coincidence of the case and not a confirmation of either study.

Figure 4.1 — New codes by interview. Over 14 interviews the new codes fall from 9 to 0 and the cumulative count ends at 30 codes; the last new code appears at interview 9. One study found code saturation at nine interviews and another at twelve, and hearing every code is not understanding it, which took 16 to 24 interviews in that study.

The question of how many customers to interview is also the question the abstract of the customer-needs study asks, and the abstract does not answer it (Griffin & Hauser, 1993). The answer for a given study is found by running the study and watching the codes, and the watching is what section 4.8 describes. A business that plans fourteen conversations and finds that new codes are still arriving at the fourteenth has learned something useful, and the useful thing is that the study is not finished.

4.8 Coding, Themes and the Analysis Grid

Coding is the step that turns a stack of notes into something a decision can use, and it is done in words rather than in software. The method read for this chapter is thematic analysis, and the abstract of the paper that describes it reports that thematic analysis is a flexible and accessible method for qualitative data that was poorly demarcated, that the paper explains what it is and how it differs from other approaches, that it gives practical guidance for doing it rigorously, and that it names pitfalls and weighs advantages and disadvantages (Braun & Clarke, 2006). The abstract does not give the number or the names of the paper's phases, and this book therefore prints none of them and describes the work in its own words.

The work has five movements, and a business can do all five at a kitchen table. The first is to read the notes of all the conversations once, without marking anything, to get the range of what was said. The second is to read them again and give each idea a short label, which is a code: "response time", "one contact", "equipment age", "price after proof". The third is to group the labels into themes, which are the ideas that several labels belong to. The fourth is to count how many interviews hold each theme, because a theme held by eleven of fourteen conversations is a different finding from a theme held by two. The fifth is to keep the evidence beside the theme, so that the theme can be traced back to the notes that produced it.

A person codes, and an assistant may suggest a first set of labels that a person checks. The distinction is the one this book keeps everywhere: an assistant drafts, and a person reads and decides. An assistant that proposes twenty labels for a set of notes has done a useful first pass, and the pass is not a code book until a person has read the labels against the notes, merged the ones that say the same thing, split the ones that say two things, and discarded the ones that describe the interviewer rather than the participant. Book K4 (Artificial Intelligence for Business) treats the use of assistants in depth, and this chapter treats only the rule that a person reads what an assistant drafts.

The distinction between code and meaning returns here, and it is the reason the counting step is not the end of the analysis. The abstract of the second study reports that code saturation was reached at nine interviews while meaning saturation took sixteen to twenty-four (Hennink et al., 2017). A business that has counted its themes has reached the first and not the second. The second is reached by going back to the notes and asking what each theme means for the person who holds it: not that eleven people mentioned response time, but what response time means to a property manager who is accountable to a building owner, and what it would take for that person to believe a promise about it.

The case's coding, posed by this book, produced thirty codes and four themes. The table below sets out the four themes, what each of them says, and what the business can do with it. Read it as the output of the coding rather than as a finding of the study: the themes are what the fourteen conversations held, and nothing in the table is credited to the research as a result. The fourth theme says that price matters only after proof.

Table 4.3 — The case's four themes

4.9 Consent, Recording and the Respect Owed to Participants

Consent is the first thing a session owes a participant, and it is given before anything is collected. The international code read for this book states the requirement as self-regulation: researchers identify themselves promptly, participation is voluntary and informed at the start, re-contact needs agreement, retention is stated, and data subjects may withdraw at any time (International Chamber of Commerce & ESOMAR, 2025, art. 4). The code is self-regulation recognized by over sixty associations in more than fifty countries, and it binds its members and the associations that adopt it, never a business by law. A business that follows it is following a professional standard, and a business that wants to know what the law requires of its own recording is asking a question for Book H2 (Privacy and Personal Information) and for a lawyer.

The federal privacy statute read for this book states the same requirements in law for the organizations it covers. Under the Personal Information Protection and Electronic Documents Act, Schedule 1, clauses 4.2 to 4.5, current to September 21, 2026, purposes are identified at or before collection, documented and explained; knowledge and consent are required, consent cannot be obtained through deception, and consent can be withdrawn subject to legal or contractual restrictions and reasonable notice; collection is limited to what the identified purposes need and is made by fair and lawful means; and use and disclosure are limited to those purposes, with information kept only as long as necessary and then destroyed, erased or made anonymous. Whether a given interview needs express consent is Book H2's question, and this chapter does not answer it.

In Québec, the private-sector statute read at the level reused from Books C9 and D2 provides that personal information may not be communicated to a third person without consent, and that consent must be clear, free, informed and specific (Act respecting the protection of personal information in the private sector, ss. 13 and 14, up to date to August 12, 2026). The statute is printed here only at that level, and nothing of it beyond sections 13 and 14 is treated in this book; Book H2 (Privacy and Personal Information) is where the statute is treated. The two statutes together are the reason a business keeps its consents in the research file with their dates, rather than relying on a recollection of what was agreed at the start of a call.

The respect owed to participants extends past the session. The code read for this book requires that no personal data be transferred to a client without consent for that purpose, that a clear privacy notice be given, and that data be protected against re-identification by deduction (International Chamber of Commerce & ESOMAR, 2025, art. 6). For a small business the practical form of that requirement is a folder that only the research team can open, a set of notes that carries roles rather than names where a name is not needed, and a rule that a participant's account is never repeated in a way that lets a listener work out who gave it. The answers also stay out of the prospect list, which is the rule of section 4.3 and the reason research data are kept outside the prospecting file.

The table below sets out the five moments of a session and what is said or done at each of them, with the rule it follows and the book that owns the rule. Read it as a checklist for the day of the conversation: the five rows are the five points at which a business either keeps the participant's trust or spends it, and the last row is easy to forget, because storage happens after the interesting part is over.

Table 4.4 — Consent and recording, step by step

THE RULE Researchers identify themselves, participation is voluntary, and nothing is recorded without consent given first (International Chamber of Commerce & ESOMAR, 2025, art. 4)

The international code read for this book, in its 2025 revision read on October 10, 2026, states at article 4 that researchers identify themselves promptly, that participation is voluntary and informed at the start, that re-contact needs agreement, that retention is stated, and that data subjects may withdraw at any time (International Chamber of Commerce & ESOMAR, 2025, art. 4). The rule this chapter draws from it is operational: a session opens by saying who is asking, for whom and for what, and nothing is recorded before a consent that is given first and noted. The code is self-regulation, binding its members and the associations that adopt it, never a business by law. The case's opening words of a session are the example, and the rules on personal information and consent are Book H2 (Privacy and Personal Information)'s.

QUÉBEC & CANADA NOTE This book describes published rules on research, personal information, competition and copyright and is not legal advice

This book describes published rules on research, personal information, competition and copyright and is not legal advice. Under the Personal Information Protection and Electronic Documents Act, Schedule 1, clauses 4.2 to 4.5, current to September 21, 2026, purposes are identified at or before collection, knowledge and consent are required and cannot be obtained through deception, collection is limited to what the purposes need, use and disclosure are limited to those purposes, information is kept only as long as necessary, and consent can be withdrawn subject to legal or contractual restrictions and reasonable notice. In Québec, the Act respecting the protection of personal information in the private sector, ss. 13 and 14, up to date to August 12, 2026, provides at the level reused here that personal information may not be communicated to a third person without consent, and that consent must be clear, free, informed and specific. Both are called to Book H2 (Privacy and Personal Information). This book explains where the rules stand, recommends no vendor, and a lawyer answers the reader's own case.

WHAT CHANGED The ICC/ESOMAR code was revised in 2025 after 2016, and the federal privacy act is current to September 21, 2026

The international code read for this book was revised in 2025 after its 2016 edition, and the 2025 revision was read on October 10, 2026 (International Chamber of Commerce & ESOMAR, 2025). The federal privacy statute read for this book is current to September 21, 2026 and was last amended on March 4, 2025 (Personal Information Protection and Electronic Documents Act, Schedule 1, clauses 4.2 to 4.5), and it is called to Book H2 (Privacy and Personal Information). The Québec private-sector statute is up to date to August 12, 2026 at the level reused from Books C9 and D2 (Act respecting the protection of personal information in the private sector, ss. 13 and 14), and it is called to Book H2 as well. Codes and statutes change, and a business that keeps a research practice has to date its reading: this book dates its own as of October 10, 2026, and a reader who needs the current text reads the current text. No amount of a penalty is printed here, because no text read for this book prints one.

4.10 The Case's Interviews

The case's interviews are posed by this book and are illustrative. Nordec Entretien inc. holds fourteen conversations, and the numbers that follow are the case's own rather than a figure taken from any source. The fourteen conversations are divided into four groups: six property managers, four owners who run their own building, two former customers and two customers who renewed. Each conversation lasts forty-five minutes and takes fifteen minutes of notes, so each costs an hour and the fourteen together cost fourteen hours. The recruiting log records the groups, the approaches and the refusals, and the refusals are kept with their group and their reason in the research file.

The coding of the fourteen conversations produced thirty codes and four themes. The new codes by interview are 9, 6, 4, 3, 3, 2, 1, 1, 1, 0, 0, 0, 0, 0, and the last new code appears at interview nine. The table below sets out the fourteen interviews, the new codes each one produced and the cumulative count, and it is the table from which Figure 4.1 is drawn. Read it as the case's own curve: the bars fall to nothing by the tenth interview, and the cumulative line stops at thirty. The table says that the codes were heard by the ninth conversation, and it does not say that the meaning of the codes was understood, which is the distinction section 4.8 draws.

Table 4.5 — New codes by interview, posed by this book

The four themes are the ones set out in Table 4.3: proof of response time, one contact rather than a bundle, planning equipment replacement, and price only after proof. The themes are what the fourteen conversations held, and the business reads them as reasons and words rather than as counts of a market. The count of how many conversations hold each theme is kept beside the theme in the analysis grid, and the grid is what the survey of Chapter 5 will use when it turns the themes into questions with options. Nothing in this section is credited to the study: the decision rule written before the fieldwork decides, and no contract, sale or share is attributed to the research.

The fourteen conversations are coded by gate 2 of the plan, on day 20 of this book. The plan of Chapter 13 runs thirty days in three slices of ten days, with gates on days 10, 20 and 30, and a gate that is not met is tried again three days later. Gate 2 requires the fourteen interviews coded, the survey closed, the search-interest index and the public reviews read, and the watch log opened on the three competitors. The interviews are therefore not an open-ended activity: they have a date by which the coding is finished, and the date is what keeps the conversations from becoming a permanent habit that never produces a finding.

DO THE MATH The case's fourteen conversations cost 14 hours, and the new codes stop at interview 9

Illustrative figures; the case is fictional. Step 1 — The case holds 14 conversations. Step 2 — Each conversation lasts 45 minutes. Step 3 — Each conversation takes 15 minutes of notes, so each conversation costs 60 minutes. Step 4 — 14 times 60 minutes is 840 minutes, which is 14 hours. Step 5 — The new codes add up to 30 by interview 9 and none appears after that, which says the codes were heard and does not say that the meaning was understood. The hours are the case's own and are posed by this book, and nothing is credited to the study: the decision rule decides.

CASE IN POINT Nordec Entretien inc. holds fourteen conversations and hears the last new code at the ninth

Nordec Entretien inc. is a fictional business created for this encyclopedia; its owner, its employees, its customers, its competitors, its figures and the events described are invented, and any resemblance to a real business is coincidental. Sophie Nadeau and the lead technician talk to the field, and the administrator keeps the recruiting log. The fourteen conversations are posed by this book and divide into four groups: six property managers, four owners who run their own building, two former customers and two customers who renewed. The coding produced thirty codes and four themes, and the last new code appears at interview nine. The themes are proof of response time, one contact rather than a bundle, planning equipment replacement, and price only after proof. The figures are posed by this book and illustrative, and no quotation, no dialogue and no verdict is printed here.

IN LA MACHINE La Machine schedules the meeting of an interview and records nothing, and respondents stay apart from prospects

CRM → Meetings (productivity/meetings) holds the meeting of an interview, its date and its notes, and it records and transcribes nothing. CRM → Contacts (business/marketing/contacts) holds every person who submits a form, and nothing separates a respondent from a prospect automatically, so this book keeps research data outside the prospecting file. The copy read for this book has no recording, transcription or coding of interviews, and the case does not use La Machine for the recording of the interviews. Check it in your installation. La Machine is described and never sold, and the survey of the case is run by the fieldwork service outside the application.

Costly Mistakes

• Reading fourteen conversations as if they were a poll. A set of volunteers is not a sample of a market, and no margin of error can be printed beside a conversation, because there is no probability sample behind it.
• Asking a question that starts with "would you". An intention is cheap to state and a switch is expensive to make, and a business that writes an intention into its notes as a decision has built a forecast on a courtesy.
• Recording a session without a consent given before the recording starts. A recording made first and consented to afterwards is not evidence the business can use, and the consent has to be noted with its date in the research file.
• Adding an interviewee to the prospect list. The answers stay in the research file, and a business that calls a participant back with an offer has turned a research conversation into a sales approach without saying so.
• Stopping at code saturation and calling it a finding. Hearing every code is not understanding it, and a list of themes that has not been explored for what each theme means is a list rather than an analysis.
• Letting a transcription draft stand as the record. A draft produced by a transcription tool carries errors in exactly the places that matter, and a person checks every quotation against the notes before anything is used.
• Recruiting only the people who are easy to reach. A study drawn from the satisfied and the nearby answers a narrower question than the decision asked, and the refusals are part of the record rather than a gap in it.

Key Takeaways

• A conversation establishes what people say they did, why they say they did it and the words they use, and it does not establish how many people think so; that question belongs to a survey with a frame and a margin.
• Individual interviews reach a history in sequence and group discussions reach a topic with interaction, and neither is better in general: the abstract of the customer-needs study asks which is better and does not answer the question in the part read.
• Participants are recruited by group, the refusals are recorded with their group and their reason, and every interviewee stays out of the prospect list because research and non-research activities are kept apart.
• A guide is four to six topics turned into open questions that start with how, when or what, and it asks about the last time something happened rather than about what a person would do.
• The case's fourteen conversations produced thirty codes and four themes, the last new code appeared at interview nine, and hearing every code is not understanding it, which took sixteen to twenty-four interviews in one study of health research.
• Nothing is recorded without a consent given before the recording starts and noted, the consent can be withdrawn subject to legal or contractual restrictions and reasonable notice, and the rules on personal information are Book H2 (Privacy and Personal Information)'s.

Your Action Plan

1. Write the decision the conversations serve before recruiting anybody. Put the decision in one sentence, say what the answers will change, and name the groups the decision needs. Deliverable: one sentence of purpose and a list of groups. Verification: every group on the list is a group the decision can act on, and no group is on the list because it is easy to reach.
2. Build the recruiting log and record the refusals. Record the group, the date of the approach, the outcome and, where the person declined, the reason given, and keep the log in the research file. Deliverable: a recruiting log with its refusals. Verification: every refusal carries a group and a reason, and no participant appears on a prospect list.
3. Write the guide in five steps and have the owner approve it. Write the purpose, list four to six topics from the recent past to the present, turn each into open questions, test the guide with one colleague, and cut every question that suggests an answer. Deliverable: an approved guide of four to six topics. Verification: no question in the guide starts with "would you", and every question can be answered with a fact.
4. Run each session with an opening, a note-taker and an expansion within the hour. Say who is asking, for whom and for what, ask for consent before recording anything, take notes during the session, and expand them within the hour. Deliverable: notes expanded within the hour and a noted consent. Verification: every session has a consent noted with its date, and every quotation has been checked against the notes.
5. Code the notes, count the themes and say where the codes stopped. Read the notes once, label each idea, group the labels into themes, count how many conversations hold each theme, and keep the evidence beside the theme. Deliverable: an analysis grid of codes, themes and counts. Verification: the grid says at which interview the last new code appeared and which themes are held by how many conversations.

In La Machine

The pages named in this chapter carry the scheduling and the contact record, and they carry nothing of the conversation itself (La Machine, 2026). CRM → Meetings (productivity/meetings) holds the meeting of an interview, its date and its notes, which is where a session is placed in the calendar and where a short record of it sits beside the date. CRM → Contacts (business/marketing/contacts) holds every person who submits a form, and the case keeps its research data outside that file, because nothing in the copy read for this book separates a respondent from a prospect automatically. The neutral sentence that describes the platform in this book is worth printing once: La Machine can host a simple questionnaire as a form whose answers become contacts, keep competitors as company records with dated notes, remind a team of its weekly watch through recurring alerts and tasks, and stage public business listings for prospecting; it draws no sample, computes no margin of error, offers no rating scale, records or transcribes no interview, listens to no social network, monitors no competitor and holds no market data. Book N4 (CRM, Contacts and Prospect Data) holds the rules of contacts.

What the platform does not do belongs in the same breath, because a business that expects the interview work to be built into the software will not build it. The copy read for this book has no recording, transcription or coding of interviews, no consent record distinct from marketing consent, no recruiting log and no analysis grid. The recordings, the notes, the consents, the recruiting log, the codes and the research file are kept outside the product, in a locked folder and a shared sheet, and the platform carries the meeting and the contact. Check it in your installation. La Machine is described and never sold, and the survey of the case is run by the fieldwork service outside the application.

Self-Check

1. What does a conversation establish, and what does it not establish?
2. When does a business choose individual interviews, and when does it choose a group discussion?
3. Why are the people who said no part of the record, and what happens to the people who said yes?
4. What is the difference between asking about the last time something happened and asking what a person would do?
5. What is the difference between code saturation and meaning saturation, and what did the case reach?

Answers. 1. A conversation establishes what people say they did, why they say they did it and the words they use, and it does not establish how many people think so, because a set of volunteers is not a sample of a market and no margin of error can be printed beside it. 2. An individual interview is chosen when the decision needs a history in sequence, which is what the case's decision needs because a building's maintenance decision is made by one person for one building; a group discussion is chosen when the decision needs a topic and the interaction of several people in a room, and neither is better in general. 3. The refusals are recorded with their group and their reason because the people who said no are evidence about the people the business does not reach, and the people who said yes stay out of the prospect list because their answers belong to the research file and research and non-research activities are kept apart. 4. A question about the last time produces an account with a date, a sequence and a decision in it, while a question that starts with "would you" produces a promise that costs the person nothing, and stated intentions overstate. 5. Code saturation means having heard every label the material will produce, and meaning saturation means understanding what each label means for the people who hold it; the case reached code saturation at interview nine and did not reach meaning saturation, and one study of health research needed sixteen to twenty-four interviews for the second.

Chapter References

Act respecting the protection of personal information in the private sector, CQLR c P-39.1, ss. 13 and 14 (Québec). LégisQuébec, up to date to August 12, 2026 — read October 9, 2026. https://www.legisquebec.gouv.qc.ca/en/document/cs/P-39.1

Braun, V., & Clarke, V. (2006). Using thematic analysis in psychology. Qualitative Research in Psychology, 3(2), 77–101. https://doi.org/10.1191/1478088706qp063oa

Christensen, C. M., Hall, T., Dillon, K., & Duncan, D. S. (2016, September). Know your customers' "jobs to be done." Harvard Business Review. https://hbr.org/2016/09/know-your-customers-jobs-to-be-done

Griffin, A., & Hauser, J. R. (1993). The voice of the customer. Marketing Science, 12(1), 1–27. https://doi.org/10.1287/mksc.12.1.1

Guest, G., Bunce, A., & Johnson, L. (2006). How many interviews are enough? Field Methods, 18(1), 59–82. https://doi.org/10.1177/1525822X05279903

Hennink, M. M., Kaiser, B. N., & Marconi, V. C. (2017). Code saturation versus meaning saturation. Qualitative Health Research, 27(4), 591–608. https://doi.org/10.1177/1049732316665344

International Chamber of Commerce & ESOMAR. (2025). ICC/ESOMAR international code on market, opinion and social research and data analytics. ESOMAR. Retrieved October 10, 2026, from https://community.esomar.org/uploads/public/knowledge-and-standards/codes-and-guidelines/ICCESOMAR-International-Code_English.pdf

La Machine. (2026). Documentation — route reference inventory. Internal product documentation, read October 10, 2026, together with the code of the pages this book describes: productivity/meetings and business/marketing/contacts.

Personal Information Protection and Electronic Documents Act, S.C. 2000, c. 5, Schedule 1, clauses 4.2 to 4.5 (Canada). Justice Laws Website, current to September 21, 2026 (last amended March 4, 2025) — read October 10, 2026. https://laws-lois.justice.gc.ca/eng/acts/P-8.6/page-7.html

Chapter 5 — Surveys

A business that needs to know how many, and how sure it can be, runs a survey whose frame and sample it can defend.

What You'll Be Able to Do

• Decide whether a question calls for a survey at all, by asking whether the question is one of how many or how much, whether the population can be listed or reached, and whether the people in it can answer from knowledge or experience.
• Draw a sampling frame and a sample from it, and say in one sentence what the frame leaves out and how wide that gap is.
• Distinguish a probability sample from a non-probability sample, and print a margin of error only for the first of the two.
• Compute how many completed answers a survey needs for a target margin, using the formula the national statistics agency's manual gives, and round the result up.
• Write questions that measure one thing, with answer choices that cover every case, and freeze the questionnaire before the first call is made.
• Report a result with its question wording, its frame, its draw, its dates, its completes, its response and its margin, and say plainly when a difference is not shown.

Why It Matters

A survey is the only tool in this book that can tell a business how many. Interviews tell a business why a property manager changed providers, and a desk study tells a business how many buildings stand in a size band, but neither can say what share of a market has heard of the business or what share is under a bundled contract. Those are questions of proportion, and a proportion is what a survey measures. The catch is that a proportion is only as good as the list it was drawn from and the arithmetic behind it, and a business that skips either step ends up with a number that looks like knowledge and behaves like a guess. This chapter is about the two steps that make the difference: a frame a business can describe, and a margin a business can defend.

The temptation to skip the arithmetic is strong, because the arithmetic is unfamiliar and the number is not. A business that asks two hundred of its own contacts a question and gets one hundred and twenty answers has a result that is easy to state and impossible to interpret, because the one hundred and twenty who answered are not a smaller version of the market. They are the people who answered. A survey knows who was eligible, who was drawn, who was reached and who answered, and can therefore say how far the answer might be from the truth. Everything in this chapter serves that single capacity.

There is a second reason the arithmetic matters, and it is the one that a business discovers when it repeats a measurement. Nordec Entretien inc., the fictional business followed through this book, measured its prompted awareness at 19 of 60 in Book B3 (Branding and Visual Identity), section 9.7, which is 31.7 percent with a margin of about plus or minus 11.8 points. The business later measures again and finds 35 of 92, which is 38 percent with a margin of plus or minus 9.5 points. The second number is higher than the first, and the difference is not shown, because the two intervals overlap. A business that had not printed the margins would have announced a rise in awareness that its own data does not support, and would have built the following year's plan on it. The margin is not a decoration on a result. It is the part of the result that says whether the result means anything.

In This Chapter

Section 5.1 sets out when a survey is the right tool and where this chapter stops and its neighbors begin. Section 5.2 separates the population, the sampling frame and the sample, and shows the frame the case draws. Section 5.3 distinguishes probability from non-probability samples and explains why only the first can carry a margin of error and why the second never does. Section 5.4 computes how many completed answers a survey needs. Section 5.5 covers the writing of questions that measure one thing. Section 5.6 covers answer scales, order effects and sensitive questions. Section 5.7 compares survey modes, invitations and response rates. Section 5.8 covers nonresponse, satisficing and the other biases that shape a result. Section 5.9 covers the analysis and the honest reporting of a survey. Section 5.10 gives the case's survey, from the frame to the interval that overlaps.

5.1 When a Survey Is the Right Tool

A survey is the right tool when the question is one of how many or how much, when the population can be listed or reached, and when the people in it can answer from knowledge or experience. Those three conditions are not a formality. Each one rules out a family of questions that a questionnaire answers badly, and each has a cheaper tool beside it. A question of how many buildings in a territory are managed by a property management firm is a survey question. A question of why a property manager left a provider is not, because the answer is a story rather than a count, and a story is what an interview collects. A question of how many buildings stand in a size band is not a survey question either, because the answer is already written in a public register and the register is more complete than any sample.

The first condition, that the question be one of how many or how much, is the one that decides many cases. A survey produces a share: a share of a population that has heard of a business, a share that is under a bundled contract, a share that plans to change providers. A share is a number with a denominator, and the denominator has to be something a business can name. If the question is "what do our customers think of us", the denominator is the business's own clientele, and that question belongs to Book B10 (Customer Experience, Retention and Loyalty), chapter 4, which owns the surveys of the business's own clientele. If the question is "what share of the market has heard of us", the denominator is a market, and that is this chapter.

The second condition, that the population can be listed or reached, is the one that decides whether a survey is possible at all. A population that cannot be listed can sometimes still be reached, and a population that can be neither listed nor reached cannot be surveyed, whatever the budget. The case's population is the commercial buildings in a size band within forty kilometers of the business's base, and the list from which the sample is drawn is the property assessment roll, which is a public data set. That is a favorable case, because a public register exists; a business selling to a population with no register has to build a frame by other means, and that frame is a limit on what its survey can say.

The third condition, that people can answer from knowledge or experience, rules out questions about the future and questions about motives. A question about what a respondent will do next year collects a stated intention, and stated intentions overstate, as section 8.7 of this book sets out. A question about why a respondent chose a provider collects a reconstruction, and a reconstruction is what an interview is for. A survey can ask what a respondent has done, what a respondent knows, and what a respondent currently has; those three are the territory of the tool.

A survey is also the wrong tool when the decision does not need a proportion. A business deciding whether to open a third territory needs a count of buildings, which comes from a register, and a count of tenders won by a competitor, which comes from a public tender system. Neither needs a sample. The discipline of this chapter is to reach for a survey only when a share is what the decision turns on.

The table below sets out where this chapter stops and its neighbors begin. Its three columns name the neighboring book, what that book holds, and what this chapter holds instead; read it as a boundary rather than a hierarchy, because each row is a subject that a business will meet and that belongs to one book only.

Table 5.1 — Where this chapter stops and its neighbors begin

WHO DOES THIS The administrator draws the frame, the fieldwork service makes the calls and the owner approves the questions

The work of a survey divides differently at three sizes of business. A solo owner designs the questionnaire, draws the sample, runs the fieldwork and analyzes the answers, and may buy the calling from a fieldwork service when the volume is beyond one person's week. A small team has one person design the questionnaire and draw the frame, and a fieldwork service collect the answers, which is the arrangement the case uses. An established business adds a statistician or an analyst to check the sample and the analysis before the results are reported, because at that size a survey may be repeated and the comparability of two measurements matters as much as either one. In the case, the administrator draws the frame, the fieldwork service makes the calls and offers nothing for sale, and Sophie Nadeau approves the questions before the first call. No hour and no dollar is attached to those roles here, because the case's costs are posed in section 5.10 and in Chapter 13.

5.2 Population, Sampling Frame and Sample

Three words carry the whole of a survey's credibility, and they are easy to confuse because in ordinary speech they are used interchangeably. The population is everyone or everything the question is about. The sampling frame is the list from which the sample is drawn. The sample is the units actually drawn from that list. A business that can state all three, in one sentence each, has done the part of survey design that most determines what the result can mean. A business that cannot state them has a number without a denominator.

The gap between the population and the frame has a name, and the national statistics agency's manual treats it as one of the sources of error that a survey carries. The manual, read at its archived edition of 2010, sets out that coverage errors are omissions, erroneous inclusions, duplications and misclassifications in the frame (Statistics Canada, 2010). An omission is a unit that belongs to the population and is missing from the list. An erroneous inclusion is a unit on the list that does not belong to the population. A duplication is a unit counted twice. A misclassification is a unit placed in the wrong group. Each of the four narrows what a result can say, and each is worth naming in the report rather than discovering after the fact.

The case's frame illustrates the point. The population is the commercial buildings in the size band within forty kilometers of the business's base. The frame is the 1,259 buildings in that band that are not already under contract with Nordec Entretien inc., which is a smaller list than the population because a building already served is not a prospect for the question being asked. Of those 1,259, the administrator establishes that 1,050 have a reachable decision-maker, which is a coverage of 83.4 percent. The 209 buildings whose decision-maker could not be reached are an omission in the frame, and the report says so. The sample is the 360 numbers drawn at random from the frame, and the completes are the 92 questionnaires answered through their key questions.

The order of those three sentences is the order in which the work is done. The frame is drawn first, from a public register or from a list the business builds, and it is described in writing before a single number is drawn from it. The description says what the frame covers, what it leaves out, how the list was assembled and on what date it was current. A frame assembled from a register that is updated yearly is current to the register's date, and the report says which date that is.

The frame is also where the case's conduct rules bite. The property assessment roll redacts owners' names and postal addresses, and its license forbids any attempt to re-identify the redacted owners; the case therefore counts buildings and never owners, and the frame holds buildings rather than people. Where a frame does hold people, the answers those people give stay in the research file and never enter the prospect list, as section 5.10 sets out and as the international code of research requires of the separation between research and non-research activity, in the article that section 4.3 cites.

DEFINITION The words this book uses for a population, a sampling frame, a sample, a margin of error and a completed answer

A population is everyone or everything the question is about, named precisely enough that a reader could say whether a given unit belongs to it. A sampling frame is the list from which the sample is drawn, described with what it covers, what it leaves out and the date it was current. A sample is the units actually drawn from the frame, and the draw is the act that makes the sample a sample rather than a convenience. A margin of error is the half-width of the interval around a share from a probability sample, stated with its confidence level, so that a share of 38 percent with a margin of plus or minus 9.5 points at 95 percent confidence describes an interval from 28.5 percent to 47.5 percent. A completed answer is a questionnaire answered through its key questions, which is the unit a survey counts and the unit a response rate is built on. These five are this book's own working words, and the margin belongs to a probability sample only.

5.3 Probability and Non-Probability Samples

There are two types of sampling, and the difference between them is not a matter of quality but of what can be computed. The national statistics agency's manual sets out the distinction plainly. Probability sampling selects units at random with a calculable probability, which is what lets estimates of sampling error be produced, and a non-probability sample never gives that estimate. Non-probability sampling is fast, easy and inexpensive, but it uses a subjective method of selection, and inference from it requires the risky assumption that the sample is representative (Statistics Canada, 2010). The two sentences are the whole of the matter. A probability sample carries an arithmetic that says how far a result might be from the truth. A non-probability sample carries an assumption, and the assumption is the part that cannot be checked from the data.

The case uses a probability sample, and the reason is that the frame makes one possible. The 1,259 buildings are a list, and 360 numbers are drawn from that list at random, which gives every building in the frame a calculable chance of being drawn. That is what allows the survey to print a margin. A business that instead posts a link on its own site and asks whoever clicks has a non-probability sample, because the people who clicked chose themselves, and no arithmetic can say how far their answers are from the market's. The two designs can produce the same completes and the same headline share, and only one can say what the share is worth.

The national council's page on reporting precision states what a report must do in each case. The results of a probability sample, and of no other kind of sample, must provide an estimate of sampling error, that is a margin of error, with the limits from non-sampling error and potential bias disclosed. The page recalls that, under the standards of the Government of Canada, no other measure of precision may be quoted for a non-probability sample, and no statement may suggest a sampling error equivalent to that of a probability sample of the same size. For a non-probability sample such as an online panel or an opt-in list, the report states that the sample is non-probability, and a measure of precision is allowed only if it is defined and supported by a detailed description of the model, its validated assumptions and its calculation. The page also notes that open-link surveys carry risks of bias, fraudulent and duplicate answers, manipulation and lack of representativeness, that they must disclose their limits, and that they must never be combined with sample data (Canadian Research Insights Council, n.d.-a). The page is self-regulation: it binds the council's members, and it does not bind a business that is not a member.

The research on the two designs points in the same direction, at the level it was read. The abstract of one study reports that, comparing survey results of American adults with benchmarks, probability-sample telephone and internet surveys were consistently more accurate than non-probability internet surveys, even after demographic post-stratification, which helped some non-probability surveys and harmed others; the abstract also reports that higher completion and response rates of the surveys were associated with less accuracy (Yeager et al., 2011). That is one study of American adults, and it is printed here as one study of American adults. A task force of a national association was set up in the fall of 2011 to study the conditions under which survey designs that do not use probability samples can still support inference to a larger population, and its introduction, read at the publisher's page, states that the assumptions needed for valid inference from such samples are not well understood by survey researchers (Baker et al., 2013). The introduction was read; the task force's conclusions were not, and none is printed here.

KEY CONCEPT A margin of error belongs to a probability sample, and a number printed beside a panel or an open link borrows an authority it does not have

The central idea of this section is this book's own reading and never a finding of a source. Probability sampling is what makes a margin calculable, because it gives every unit in the frame a known chance of being drawn and therefore gives the arithmetic something to work with (Statistics Canada, 2010). The national council's page asks that a report on a non-probability sample say that the sample is non-probability, and it allows a measure of precision only where the model behind it is defined and supported (Canadian Research Insights Council, n.d.-a). Put the two together and the practical consequence is plain. An opt-in panel, an open link, a set of reviews and a focus group get no margin, and a number printed beside them in the shape of a margin borrows an authority that the design does not have. The number may still be useful, and the business may still act on it; what it cannot do is claim a precision it did not earn.

THE RULE Report a margin of error only for a probability sample, and say when the sample is not (Canadian Research Insights Council, n.d.-a)

The rule this chapter draws from the national council's page, read on October 10, 2026, is operational. The results of a probability sample, and of no other kind of sample, must provide an estimate of sampling error, that is a margin of error, with the limits from non-sampling error and potential bias disclosed. For a non-probability sample, the report states that the sample is non-probability, and no measure of precision is quoted unless it is defined and supported by a description of the model, its assumptions and its calculation. The page binds the council's members and not a business that is not a member, and it is self-regulation rather than law. The case's frame of 1,259 buildings is a probability frame, so the case prints a margin for its shares and names the confidence level beside it. A business that runs an open link on its own site prints the share, says the sample is non-probability, and prints no margin.

5.4 How Many Answers a Survey Needs

The number of answers a survey needs is a calculation, not a preference, and the calculation is short enough to do on a page. The national statistics agency's manual gives the simple-random-sample size for a margin of plus or minus 5 points at 95 percent confidence when the estimated proportion is one half, which is the proportion that requires the largest sample. The manual's table gives 44 answers for a population of 50, 222 for a population of 500, 286 for a population of 1,000, 370 for a population of 5,000 and about 400 for large populations, and the table is computed with a value of 2 for the confidence multiplier (Statistics Canada, 2010). The case uses 1.96 rather than 2, which is the same confidence level expressed more precisely.

The table below reproduces the manual's values. Its two columns give the population size and the number of answers the manual's table requires for a margin of plus or minus 5 points at 95 percent confidence; read it to see the shape of the arithmetic rather than to memorize the rows, because the shape is what a business needs: the answers needed rise with the population and then flatten, so that a population of 5,000 and a population of several million need almost the same number of answers.

Table 5.2 — Answers for plus or minus 5 points by population, from the national agency's manual

The manual's equation brings the population size in, so that the sample size is the first estimate divided by one plus the first estimate divided by the population (Statistics Canada, 2010). For a large population the second term is negligible and the answer is the first estimate; for a small population the correction matters, and it is what brings the requirement for a population of 1,000 down from about 400 to 286. A market of a few hundred units therefore needs far fewer answers than a national market, and the arithmetic says so before any fieldwork is bought.

Two further adjustments belong in the calculation, and the manual names both. The first is a design effect, which multiplies the size; the manual gives at least 2 for a cluster design without prior knowledge (Statistics Canada, 2010). A cluster design is one in which the units are drawn in groups rather than one by one, and the multiplication accounts for the fact that units inside a group resemble one another. The case draws its numbers one by one from a list, so no design effect is applied, and the report says so. The second adjustment is the expected response, which divides the required number of completes by the share of drawn units expected to answer; the manual's illustration divides 400 by 0.75 to reach 533 (Statistics Canada, 2010). The case's own division is given in section 5.10, where the draw of 360 numbers is set against the 92 completes it produced.

The manual also states the limit of the whole exercise in one sentence that a business should keep beside the formula: increasing the sample does not remove a bias if nonrespondents differ from respondents (Statistics Canada, 2010). A larger sample narrows the sampling error, which is the part the formula measures, and does nothing to the difference between the people who answered and the people who did not. Section 5.8 takes that difference up.

Figure 5.1 shows how many completed answers a survey needs for two margins by population size, and compares two awareness intervals that overlap. On the left, two curves flatten as the population grows, with the case marked in orange. On the right, two intervals, one indigo and one orange, overlap, with a dashed red note that the rise is not shown.

Figure 5.1 — Sample size and two intervals that overlap. For a simple random sample at 95 percent confidence, the answers needed flatten as the population grows; the case's 1,259 buildings need 90 answers for plus or minus 10 points. Book B3 found awareness of 19 of 60, 31.7 percent, and this book 35 of 92, 38.0 percent; the intervals overlap, so the rise is not shown.

Read the left panel as a pair of curves that rise steeply for small populations and then flatten, so that the difference between a frame of 1,259 and a frame of a million is small. Read the right panel as two intervals drawn on the same line, the indigo one running from 19.9 percent to 43.5 percent and the orange one from 28.5 percent to 47.5 percent, overlapping across most of their length. The overlap is the finding. A business that reads only the two central values sees a rise of 6.3 points; a business that reads the intervals sees that the rise is not shown.

DO THE MATH The case needs 90 completed answers for plus or minus 10 points among 1,259 buildings

Illustrative figures; the case is fictional. Step 1 — The target is a margin of plus or minus 10 points at 95 percent confidence, with the estimated proportion at one half, so the confidence multiplier is 1.96. Step 2 — The first estimate is 1.96 times 1.96 times one half times one half, divided by the margin times the margin, with the margin written as a tenth, which is 96.04. Step 3 — With the population at 1,259, the sample size is the first estimate divided by one plus the first estimate divided by the population, which is 89.2. Step 4 — The value 89.2 is rounded up to 90 completed answers. Step 5 — The formula holds for a simple random sample only, and the case draws 360 numbers to reach its completes, as section 5.10 sets out. Nothing is credited to the study, and the decision rule decides.

5.5 Writing Questions That Measure One Thing

A question measures one thing or it measures nothing. The rule sounds obvious and is broken constantly, because many of the questions a business wants to ask turn out to be compound. "How satisfied are you with our response time and our billing" is two questions wearing one coat, and a respondent who is happy with the response time and unhappy with the billing has no way to answer it. The remedy is to write, before the question, one sentence saying what the question must measure, and then to write the question so that it measures that and only that. If the sentence needs the word "and", the question is two questions.

The research on how respondents handle difficult questions explains why the discipline matters. The abstract of one paper reports that when an optimal answer demands much cognitive effort, some respondents give a merely satisfactory answer, a behavior the paper calls satisficing; it can lead to choosing the first reasonable option, agreeing with an assertion, endorsing the status quo, not differentiating among rated objects, saying "don't know" or answering at random, and the paper names factors that encourage it (Krosnick, 1991). No rate of satisficing is printed here, because the abstract does not give one. The practical consequence is that a hard question does not produce a considered answer; it produces a fast one, and the fast answer is indistinguishable from a considered one in the data.

Plain words are the first defense. The national statistics agency's manual names technical jargon among the causes of measurement error, alongside the respondent, the interviewer, the questionnaire, the collection method and the tool (Statistics Canada, 2010). A question written in the vocabulary of the industry will be answered by the respondents who share that vocabulary and guessed at by the rest, and the guesses will not be marked as guesses. The second defense is neutral wording. A question that begins "how much do you value our fast response" has told the respondent what answer is expected, and a respondent who is satisficing will supply it. The third defense is an answer set that covers every case, including the case of no opinion and the case of not knowing, because a respondent who has no option that fits will choose the nearest one. A reference work on the tailored design of surveys across modes treats the writing and the ordering of questions at length (Dillman et al., 2014); this book prints no figure and no procedure from it.

The case's key question is written to measure one thing, and it is asked in two parts. The first part asks whether the person can name a provider of preventive maintenance and on-call technical service without being prompted, and the answer is recorded as given or not given. The second part, asked only when the first produced nothing, asks whether the person recognizes the business when it is named. The two parts together produce the prompted awareness measure, and the report says which part produced each answer. No question is invented here as a quotation of a respondent; the case gives counts and never words.

A question is tested before it is fielded, and the test is small. Three people who resemble the respondents are asked the questions and asked to say what they understood each one to mean, and the differences between what the writer intended and what the reader understood are the corrections. The test is small and catches the questions that would otherwise be discovered in the data as an answer pattern that makes no sense. The questionnaire is then frozen, with its date, and the date is printed in the report, because a questionnaire that changes during fieldwork produces two surveys wearing one name.

METHOD Writing the questions of a survey in five steps

First, write what each question must measure in one sentence, and split the question if the sentence needs the word "and". Second, write the question in plain words with one idea, avoiding the vocabulary of the industry and avoiding wording that suggests an answer. Third, write answer choices that cover every case, including none of the above and don't know, and keep the same scale in the same direction throughout. Fourth, test the questions on three people who resemble the respondents, and listen for what they do not understand rather than for what they think of the survey. Fifth, have the owner approve the final questions and freeze them with their date, so that a later measurement can be compared with this one. The sequence is this book's own and no source tests it; the case's awareness question is the example, asked first without prompting and then with the business named.

TOOLS OF THE TRADE What a business needs to run a survey: a script or a link, a frame sheet, a frozen questionnaire and a call log

Four functions carry a survey, and none of them is a brand. The first is a survey tool that can present the questionnaire either as a script for a caller to read or as a link for a respondent to open, and that records each answer against the drawn unit. The second is a shared sheet holding the frame, the draw and the outcome of each attempt, so that the response can be counted and the nonrespondents can be described. The third is a place for the frozen questionnaire with its date, so that the wording of a measurement is recoverable when the measurement is repeated. The fourth is a log of the calls, holding the date, the time and the outcome of each attempt, so that the fieldwork can be described in the report. Chapter 12 gives the functions in full and the pricing models without amounts.

5.6 Answer Scales, Order Effects and Sensitive Questions

An answer scale is a measuring instrument, and like any instrument it has to be used the same way twice to be worth anything. The first discipline is to keep the same scale in the same direction throughout a questionnaire. A respondent who has just answered a five-point scale running from very dissatisfied to very satisfied, and who then meets a scale running from very satisfied to very dissatisfied, may answer the second one as if it ran the same way as the first, and the error is then invisible in the data. The second discipline is to include a middle point only when a middle exists. A question about whether a building has a maintenance contract has no middle; a question about satisfaction has one, and removing it forces respondents who are genuinely neutral to pick a side.

The third discipline concerns the order of the items, and it is the one that a small survey can act on. When a questionnaire asks about several providers or several attributes in a row, the position of an item can change its score, and the first item on the list may collect more attention than the last. Where the mode allows, the order of the items is shuffled between respondents, so that the position effect averages out rather than accumulating on one item. Where the mode does not allow it, as in a script read aloud, the report says that the order was fixed and names the order, so that a reader knows what the numbers contain.

The fourth discipline concerns the questions that people would rather not answer, and the research on those questions is worth reading at the level it was read. The abstract of a review of research on reporting errors in surveys on sensitive topics reports that misreporting is common and largely situational, depending on whether respondents have something embarrassing to report and on design features; respondents edit their answers to avoid embarrassment, especially when an interviewer is present, or to avoid repercussions from third parties (Tourangeau & Yan, 2007). The review concerns personal topics, and this chapter says so plainly: it does not measure business topics, and no figure from it is applied to the case. What it establishes for a business survey is the shape of the problem, which is that a question a respondent would rather not answer produces an edited answer rather than a refusal.

The practical rules that follow are short. A sensitive question is asked late in the questionnaire, after the respondent has settled into the task, and it is asked with a reason attached, so that the respondent knows why it is there. A question about a competitor relationship, a budget or a dissatisfaction is sensitive in a business survey, and a respondent who is asked it at the very start may answer it defensively. Where the mode allows, the question is asked in a form that the respondent can answer without saying it aloud. Where the question is genuinely necessary and the respondent declines, the refusal is recorded as a refusal, and the report counts it rather than treating it as a missing value to be filled.

The last point in this section belongs to the tools rather than to the method, and it is a limit rather than a feature. The application copy read for this book builds custom forms whose fields are text, email, telephone, number, date, list, long text, checkbox and call to action, and there is no rating scale field; every submission requires an email address or a telephone number and finds or creates a contact, so an anonymous survey is not possible as such (La Machine, 2026). A form can therefore hold a simple questionnaire only where a contact is acceptable, and a questionnaire that needs a scale, an anonymous answer or a shuffled item order is run outside the application.

5.7 Survey Modes, Invitations and Response Rates

The mode of a survey is the channel through which the questions reach the respondent, and the mode shapes who answers before it shapes what they say. A call reaches the person who is at the number and willing to talk. An email invitation reaches the person who reads the inbox and chooses to click. A form on a business's own site reaches the person who is already there. A face-to-face visit reaches the person who is on site and has a minute. Each of the four reaches a different subset of the frame, and the subset is not random, which is why the mode belongs in the report rather than in the method notes.

The research on modes is worth reading for its shape. The abstract of a meta-analysis reports that a 2008 meta-analysis found web surveys 11 percentage points below other modes in response rate over 45 experimental comparisons, and that the update over 114 comparisons finds web surveys still 12 points lower; it also reports that prenotification, recruitment strategy, solicitation mode, the target population, the number of contacts and the country moderate the difference (Daikeler et al., 2020). The abstract gives no response rate for a business-to-business survey, and none is printed here. What the abstract supports is the general point that the mode is one of the things that decides who answers.

The table below sets out four modes a small business can use, what each one reaches, one limit of each and how the case uses it. Read it as a set of trade-offs rather than a ranking, because no mode is better than another in general; each reaches a different part of the frame, and the report names the one used.

Table 5.3 — Four survey modes and what each reaches

The invitation is the part of a survey that touches the law, and the law is called rather than applied here. In Canada, calls made by organizations conducting surveys, public opinion polls or market research are exempt from the National Do Not Call List Rules; survey, poll and market research calls made solely for those purposes are not considered telemarketing, and market research and polling firms are not required to keep internal do-not-call lists (Canadian Radio-television and Telecommunications Commission, 2018). The page does not state calling hours or identification rules for survey calls, and it does not exempt a call that also sells or books an appointment, which is telemarketing and belongs to Book H3 (Anti-Spam and Telemarketing Compliance). The case's fieldwork service offers and sells nothing, says who is calling and for whom, and records the outcome of each attempt.

The email side is a question this book poses and does not decide. The long-titled anti-spam act of 2010 provides that a commercial activity is any transaction, act or conduct of a commercial character, whether or not for profit, and that an electronic message is a message sent by any means of telecommunication; it provides that a commercial electronic message is one of whose purposes, having regard to its content, hyperlinks or contact information, it is reasonable to conclude is to encourage participation in a commercial activity; and it provides that sending one requires consent (An Act to promote the efficiency and adaptability of the Canadian economy by regulating certain activities that discourage reliance on electronic means of carrying out commercial activities, and to amend the Canadian Radio-television and Telecommunications Commission Act, the Competition Act, the Personal Information Protection and Electronic Documents Act and the Telecommunications Act, ss. 1(1), 1(2), 1(3) and 6(1)). No provision of the act names surveys or research. Whether a given survey invitation is a commercial electronic message is therefore Book H3's question, and this book poses it and does not decide it; the practice this book recommends is an invitation with no commercial content, and a lawyer for the borderline cases.

QUÉBEC & CANADA NOTE This book describes published rules on research, personal information, competition and copyright and is not legal advice

This book describes published rules on research, personal information, competition and copyright and is not legal advice. Survey, poll and market research calls made solely for those purposes are not telemarketing under the national exemption page read on October 10, 2026, and a call that also sells or books an appointment is telemarketing, which is called to Book H3 (Anti-Spam and Telemarketing Compliance). Whether an email invitation to a survey is a commercial electronic message is Book H3's question, which this book poses and does not decide. Consent and the personal information of respondents are Book H2's (Privacy and Personal Information). The book explains where the rules stand, recommends no vendor, and a lawyer answers the reader's own case. No amount of a penalty is printed here, because no text read for this book prints one.

5.8 Nonresponse, Satisficing and Other Biases

Nonresponse is the difference between the units drawn and the units that answered, and it is one of the main threats to a survey's result. The research on it is careful, and the care is the lesson. The abstract of one paper reports that refusal rates of household surveys have risen, that nonresponse can but need not induce nonresponse bias, and that recent findings show cases where the link between nonresponse rates and bias is absent; it reports that a synthesis of studies estimating nonresponse bias shows the bias often present, and that whether probability samples keep an advantage despite high nonresponse depends on the design and on the quality of the auxiliary variables used to adjust (Groves, 2006). The abstract gives no threshold of response rate, and none is printed here. A meta-analysis of 59 methodological studies designed to estimate nonresponse bias explores when nonresponse rates and bias are related, examining design features, sample characteristics and the statistics computed as predictors; the strength of the relation is not in the abstract read (Groves & Peytcheva, 2008).

The practical consequence of those two abstracts is that a response rate is a description and not a verdict. What the rate does is tell a reader how much of the frame is missing from the result, and the missing part is where the risk lives. The remedy is not to chase a number but to describe the missing part: who was drawn, who was reached, who refused, and what is known about the difference between the two groups. The case's report names its frame, its draw, its completes and its response, and it says what it does not know about the 268 drawn numbers that produced no completed answer.

Satisficing is the second bias, and it is the one that a questionnaire can cause. Section 5.5 gave the abstract of the paper that named it (Krosnick, 1991), and the defense is the one set out there: short questions, plain words, one idea each, and answer choices that cover the cases. A questionnaire that respects a respondent's attention collects better answers than one that does not, and the difference is a design decision rather than a matter of respondent goodwill.

Two further biases belong in the same list. The first is the effect of measuring an intention at all. The abstract of one study reports that measuring intentions can itself inflate their association with behavior, a phenomenon the authors call self-generated validity, and that across three large field studies the intention-behavior correlation is on average 58 percent greater among surveyed than among non-surveyed consumers (Chandon et al., 2005). That is a study of consumer products, and no discount factor is taken from it for the case; the case's own discount on stated switchers is posed by this book in section 8.7. The second is the family of method biases that a review examines, its likely sources, the cognitive processes through which they shape answers, and the procedural and statistical remedies available (Podsakoff et al., 2003); the abstract does not list the sources or the remedies, and none is printed from it.

WATCH OUT A response rate is not a verdict, and a larger sample does not remove a bias

The trap is to read a response rate as a judgment on a result. A low rate is treated as proof that the answers are unrepresentative, and a high rate is treated as proof that they are sound; both readings are wrong, and the research says so. The abstract of one paper reports that nonresponse can, but need not, induce nonresponse bias, and that cases exist where the link between the rate and the bias is absent (Groves, 2006). The national statistics agency's manual states the other half of the point: increasing the sample does not remove a bias if nonrespondents differ from respondents (Statistics Canada, 2010). The signs of the trap are a response rate quoted as a verdict and a result reported without the limits of its frame. The remedy is to report the frame, the draw, the response and the limits together, in one paragraph, so that a reader can see what the number does and does not cover. The reasoning is this book's own.

5.9 Analyzing and Reporting Survey Results Honestly

The analysis of a survey begins with counting, and the counting is done against the frame rather than against the answers. The first number is the frame: how many units were eligible. The second is the draw: how many were selected. The third is the outcome of each attempt, which separates the units that were never reached from the units that refused and from the units that answered. The fourth is the completes. The fifth is the response, which is the completes divided by the draw. Only then does the analysis turn to the shares, and each share is printed with the number of answers it rests on, because a share of 92 and a share of 11 are not the same kind of statement even when they are the same percentage.

A share from a probability sample is printed with its margin and its confidence level, and the margin is the half-width of the interval around the share. The case prints its shares that way, and section 5.10 gives the table. A difference between two shares is reported only when the intervals leave room for it. The case's two awareness measurements, set out in section 5.10, are the illustration: their intervals overlap across most of their length, so the rise is not shown. A business that reports the two central values and omits the intervals has reported a rise that its own data does not support, and the omission is one way a survey misleads without anyone intending it.

The national association's disclosure standards list what a report should say, and the list is worth reading as a checklist even though it is not a Canadian rule. The standards, revised in April 2021, list what a release should disclose, from the sponsor and the exact wording of the questions to the frame and its gaps, the sample sizes, the margin of sampling error for probability samples and a statement on unmeasured error; the box on what changed gives the list in full (American Association for Public Opinion Research, 2021). The standards belong to a United States association, and they are an object of comparison rather than a rule for a Canadian business.

The national council's page adds the Canadian half of the same discipline: a probability sample reports its margin, and a non-probability sample says that it is non-probability (Canadian Research Insights Council, n.d.-a). The two together give a business a short report template, and the template is short on purpose. A page that holds the question wording, the frame, the draw, the dates, the completes, the response, the margin with its confidence level and the limits of the frame is a report that a reader can check. A twenty-page report that omits the frame is a report that a reader cannot.

The analysis also has to keep its interpretations apart from its findings, which is the discipline that section 9.4 of this book develops for the whole of a study. A finding is what the answers say. An interpretation is what the business thinks the answers mean. A recommendation is what the business proposes to do. The three are written in three places, and the report says which is which. A survey that reports an interpretation as a finding has turned a reading into a measurement, and the reader has no way to tell.

WHAT CHANGED The association's disclosure standards were revised in April 2021 and list what a report must say

The dated change is the revision of a national association's disclosure standards in April 2021, read on October 10, 2026. The standards list what a report should disclose when results are released: the sponsor and the conductor, the exact wording of the questions and their context, the population, whether the sample is probability-based or not with the frame and its gaps, the recruitment and selection, the modes and languages, the dates, the sample sizes, the margin of sampling error for probability samples, the weighting, the data quality checks and a statement on unmeasured error; for non-probability samples, a measure of precision is reported only if it is defined and accompanied by the model and its assumptions (American Association for Public Opinion Research, 2021). The association is a United States body, and its standards are an object of comparison rather than a Canadian rule. Standards change, and this book dates its reading as of October 10, 2026; a business that follows a standard should read the current text rather than this summary of it.

5.10 The Case's Survey

The survey of Nordec Entretien inc. is posed by this book and illustrative, and it is the second measurement of an awareness that Book B3 (Branding and Visual Identity), section 9.7, measured once before. The frame is the 1,259 buildings in the size band that are not under contract, of which 1,050 have a reachable decision-maker, a coverage of 83.4 percent. The administrator draws 360 numbers at random from the frame. The fieldwork service makes the calls, offers and sells nothing, and says who is calling and for whom. The survey closes at 92 completed answers, which is a response of 25.6 percent, at $25 per complete paid to the fieldwork service, $2,300 in all. The respondents' answers stay in the research file, apart from the prospect list, and no respondent is called back with an offer because of an answer.

The table below gives the case's five measures, the number of answers behind each, the share and the margin at 95 percent confidence. Read it as a set of statements with their precision attached, and read the second row against the first, because the two rows are the same measure taken twice and the comparison is the point of the exercise. The four margins of the survey's own rows apply the correction for the size of the population that section 5.4 applies to the sample size, with the 1,050 buildings that have a reachable decision-maker as the population; the margin of Book B3's row is the one Book B3 prints.

Table 5.4 — The case's survey, posed by this book

The first two rows are the measurement that matters most, and they do not show a rise. The prompted awareness of 35 of 92 is 38 percent with a margin of plus or minus 9.5 points, which describes an interval from 28.5 percent to 47.5 percent. Book B3's awareness of 19 of 60 is 31.7 percent with a margin of about plus or minus 11.8 points, which describes an interval from 19.9 percent to 43.5 percent. The two intervals overlap, so the rise is not shown. The case records that finding as it stands, and it does not credit the survey with a change in the market, because nothing in the design supports such a claim.

The third and fourth rows describe the market's structure rather than the business's reputation, and they are the rows that feed the decision. A little over half the respondents say their building is managed by a property management firm, and a little under a third say it is under a bundled facility contract. Both shares carry margins of about nine points, which means the business can act on the direction and not on the decimal. The fifth row is the smallest and the least reliable: 11 respondents, 12 percent, with a margin of plus or minus 6.3 points. A share resting on 11 answers is a weak statement, and the report says so rather than rounding it into a trend.

The case's response rate is its own, read on its day, and no rate is a norm. The report names the frame, the draw, the completes and the response, and it says what it does not know about the drawn numbers that produced no completed answer. Nothing is credited to the study: the decision rule decides, and the survey informs it. The survey closes by gate 2, on day 20 of the thirty-day plan of Chapter 13, and its results feed the synthesis of section 9.2 and the decision brief of section 9.7 of this book.

CASE IN POINT Nordec Entretien inc. draws 360 numbers, closes 92 completed answers and cannot show that awareness rose

Nordec Entretien inc. is a fictional business created for this encyclopedia; its owner, its employees, its customers, its competitors, its figures and the events described are invented, and any resemblance to a real business is coincidental. The administrator draws the frame of 1,259 buildings and the 360 numbers from it, the fieldwork service makes the calls and offers nothing for sale, and Sophie Nadeau reads the result at gate 2. The prompted awareness is 35 of 92, which is 38 percent with a margin of plus or minus 9.5 points, and Book B3's measurement is 19 of 60, which is 31.7 percent with a margin of about plus or minus 11.8 points. The two intervals overlap, so the rise is not shown, and the case records that finding as it stands. Every figure in this box is posed by this book and illustrative. No respondent is named, no answer is quoted, and no contract, sale or share is credited to the survey.

IN LA MACHINE La Machine's forms hold a simple questionnaire with no rating scale, and its responses page draws no sample

La Machine can host a simple questionnaire as a form whose answers become contacts, keep competitors as company records with dated notes, remind a team of its weekly watch through recurring alerts and tasks, and stage public business listings for prospecting; it draws no sample, computes no margin of error, offers no rating scale, records or transcribes no interview, listens to no social network, monitors no competitor and holds no market data. Marketing → Forms business/marketing/forms builds custom forms whose fields are text, email, telephone, number, date, list, long text, checkbox and call to action, with no rating scale field, and every submission requires an email address or a telephone number and finds or creates a contact, so an anonymous survey is not possible as such. Marketing → Forms → Responses business/marketing/forms/responses lists the submissions of a form and weights nothing, computes no margin and draws no sample. The case's survey is run by the fieldwork service outside the application. Check it in your installation; La Machine is described and never sold.

Costly Mistakes

• Asking a question of why with a questionnaire. A survey measures how many and how much, and a question about a motive or a future intention collects a reconstruction or a stated intention rather than a count, which is what an interview is for.
• Drawing a sample without describing the frame. A sample drawn from a list nobody has described produces a share with no denominator, and the report cannot say what the share covers or what it leaves out.
• Printing a margin of error beside a panel, an open link or a set of reviews. A margin belongs to a probability sample, and a number printed beside a non-probability design borrows an authority the design does not have.
• Reading a response rate as a verdict on the result. A low rate does not prove bias and a high one does not prove soundness, and a larger sample does not remove a bias if the people who answered differ from the people who did not.
• Reporting two shares as a rise without their intervals. Two measurements whose intervals overlap do not show a difference, and reporting the central values alone announces a change that the data does not support.
• Writing a compound question and testing it on nobody. A question that measures two things at once cannot be answered by a respondent who feels two ways, and a question that has not been tested on three people is discovered to be ambiguous in the data.
• Letting a respondent's answer become a prospect. The answers of respondents stay in the research file and apart from the prospect list, and a respondent who is called back with an offer because of an answer has been surveyed under false pretenses.

Key Takeaways

• A survey is the right tool when the question is one of how many or how much, when the population can be listed or reached, and when the people in it can answer from knowledge or experience; it is the wrong tool for why.
• The population is what the question is about, the frame is the list the sample is drawn from, and the sample is the units drawn; the gap between the population and the frame is coverage error and belongs in the report.
• A probability sample gives every unit in the frame a calculable chance of being drawn and therefore supports a margin of error; a non-probability sample supports an assumption, and a report on one says that the sample is non-probability and prints no margin.
• The number of answers a survey needs is a calculation that flattens as the population grows, and the case's frame of 1,259 buildings needs 90 completed answers for a margin of plus or minus 10 points at 95 percent confidence.
• A question measures one thing, in plain words, with answer choices that cover every case, and it is tested on three people and frozen with its date before the first call.
• A result is reported with its question wording, its frame, its draw, its dates, its completes, its response and its margin, and a difference is reported only when the intervals leave room for it.

Your Action Plan

1. Decide whether the question needs a survey at all. Write the question in one sentence, and confirm that it is a question of how many or how much, that the population can be listed or reached, and that the people in it can answer from knowledge or experience. Deliverable: one sentence naming the question and the decision it serves. Verification: the sentence names a share and a population, and it does not contain the word "why".
2. Draw the frame and describe it in writing. Assemble the list from the public source or the internal record it comes from, and write what it covers, what it leaves out and the date it was current. Deliverable: a frame sheet with its coverage stated as a number. Verification: a reader can say whether a given unit belongs in the frame, and the coverage figure is printed with the date it was measured.
3. Compute the answers needed and the draw. Apply the formula for the target margin at the chosen confidence level, round the result up, and divide by the expected response to get the draw. Deliverable: one page showing the target margin, the confidence level, the answers needed and the draw. Verification: the page states that the formula holds for a simple random sample only, and the draw is larger than the answers needed.
4. Write, test and freeze the questionnaire. Write what each question must measure, write the question in plain words with one idea, write answer choices that cover every case, test the questions on three people, and have the owner approve the final version. Deliverable: a frozen questionnaire with its date. Verification: every question has a one-sentence statement of what it measures, and no question contains the word "and" between two ideas.
5. Report the result with its limits in one page. Write the question wording, the frame, the draw, the dates, the completes, the response, the margin with its confidence level and the limits of the frame, and keep the findings apart from the interpretations. Deliverable: a one-page report. Verification: every share is printed with the number of answers behind it, and no difference is reported whose intervals overlap.

In La Machine

The application copy read for this book carries a questionnaire in one place and nothing of the survey method anywhere (La Machine, 2026). Marketing → Forms business/marketing/forms builds custom forms whose fields are text, email, telephone, number, date, list, long text, checkbox and call to action, and its page describes them as going beyond simple email opt-ins, naming discovery calls, intake and surveys among their uses. Marketing → Forms → Responses business/marketing/forms/responses lists the submissions of a form, which is where the answers of a simple questionnaire are read. The Consent page business/marketing/consent governs who may receive a message, and there is no research consent distinct from marketing consent, so the rules on electronic messages and calls are Book H3's (Anti-Spam and Telemarketing Compliance) and the rules on personal information are Book H2's (Privacy and Personal Information). Book N4 (CRM, Contacts and Prospect Data) holds the rules of contacts, and Book B1 (Marketing Fundamentals for Business Owners), chapter 3, sends the reader to the prospecting pages.

What the platform does not do belongs in the same breath, because a business that expects a survey to be built into the software will not build one. La Machine draws no sample, weights no answer and computes no margin of error. A form can therefore hold a simple questionnaire only where a contact is acceptable, and the reader draws the sample and computes the margin outside the product, in a shared sheet and a calculator. The case's survey is run by the fieldwork service outside the application, and its answers stay in the research file, apart from the prospect list. Check it in your installation; La Machine is described and never sold.

Self-Check

1. What three conditions make a survey the right tool, and which question does a survey answer badly?
2. What is the difference between the population, the sampling frame and the sample, and what is the gap between the first two called?
3. Why is a margin of error never printed for a non-probability sample, and what does the report on one say instead?
4. How many completed answers does the case need for a margin of plus or minus 10 points among 1,259 buildings, and what does the formula assume?
5. Why does the case not report a rise in awareness between Book B3's measurement and its own?

Answers. 1. A survey is the right tool when the question is one of how many or how much, when the population can be listed or reached, and when the people in it can answer from knowledge or experience; it answers a question of why badly, because a motive is a story rather than a count, and a story is what an interview collects. 2. The population is everyone or everything the question is about, the sampling frame is the list from which the sample is drawn, and the sample is the units actually drawn; the gap between the population and the frame is coverage error, and the national statistics agency's manual names omissions, erroneous inclusions, duplications and misclassifications among its forms. 3. Probability sampling gives every unit in the frame a calculable chance of being drawn, which is what allows an estimate of sampling error to be produced, while non-probability sampling uses a subjective method of selection and requires the assumption that the sample is representative; a report on a non-probability sample states that the sample is non-probability and prints no measure of precision unless it is defined and supported by a description of the model and its assumptions. 4. The case needs 90 completed answers, which is 89.2 rounded up; the formula holds for a simple random sample only, and the case draws 360 numbers to reach its completes. 5. The case's prompted awareness is 35 of 92, which is 38 percent with a margin of plus or minus 9.5 points, and Book B3's is 19 of 60, which is 31.7 percent with a margin of about plus or minus 11.8 points; the two intervals run from 28.5 percent to 47.5 percent and from 19.9 percent to 43.5 percent, they overlap, and the rise is therefore not shown.

Chapter References

American Association for Public Opinion Research. (2021). AAPOR code of professional ethics and practices: Disclosure standards (Revised April 2021). Retrieved October 10, 2026, from https://aapor.org/standards-and-ethics/disclosure-standards/

An Act to promote the efficiency and adaptability of the Canadian economy by regulating certain activities that discourage reliance on electronic means of carrying out commercial activities, and to amend the Canadian Radio-television and Telecommunications Commission Act, the Competition Act, the Personal Information Protection and Electronic Documents Act and the Telecommunications Act, S.C. 2010, c. 23, ss. 1(1), 1(2), 1(3) and 6(1) (Canada). Justice Laws Website, current to September 21, 2026 (last amended January 1, 2026) — read October 10, 2026. https://laws-lois.justice.gc.ca/eng/acts/E-1.6/page-1.html

Baker, R., Brick, J. M., Bates, N. A., Battaglia, M., Couper, M. P., Dever, J. A., Gile, K. J., & Tourangeau, R. (2013). Summary report of the AAPOR task force on non-probability sampling. Journal of Survey Statistics and Methodology, 1(2), 90–143. https://doi.org/10.1093/jssam/smt008

Canadian Radio-television and Telecommunications Commission. (2018). Exemptions [National Do Not Call List, information for organizations]. (Page modified April 9, 2018.) Retrieved October 10, 2026, from https://lnnte-dncl.gc.ca/en/Organization/Exemptions

Canadian Research Insights Council. (n.d.-a). CRIC requirements for reporting survey precision and on the use of open-link surveys. Retrieved October 10, 2026, from https://www.canadianresearchinsightscouncil.ca/cric-requirements-for-reporting-survey-precision-and-on-the-use-of-open-link-surveys/

Chandon, P., Morwitz, V. G., & Reinartz, W. J. (2005). Do intentions really predict behavior? Self-generated validity effects in survey research. Journal of Marketing, 69(2), 1–14. https://doi.org/10.1509/jmkg.69.2.1.60755

Daikeler, J., Bošnjak, M., & Lozar Manfreda, K. (2020). Web versus other survey modes: An updated and extended meta-analysis comparing response rates. Journal of Survey Statistics and Methodology, 8(3), 513–539. https://doi.org/10.1093/jssam/smz008

Dillman, D. A., Smyth, J. D., & Christian, L. M. (2014). Internet, phone, mail, and mixed-mode surveys: The tailored design method (4th ed.). Wiley.

Groves, R. M. (2006). Nonresponse rates and nonresponse bias in household surveys. Public Opinion Quarterly, 70(5), 646–675. https://doi.org/10.1093/poq/nfl033

Groves, R. M., & Peytcheva, E. (2008). The impact of nonresponse rates on nonresponse bias: A meta-analysis. Public Opinion Quarterly, 72(2), 167–189. https://doi.org/10.1093/poq/nfn011

Krosnick, J. A. (1991). Response strategies for coping with the cognitive demands of attitude measures in surveys. Applied Cognitive Psychology, 5(3), 213–236. https://doi.org/10.1002/acp.2350050305

La Machine. (2026). Documentation — route reference inventory. Internal product documentation, read October 10, 2026, together with the code of the pages this book describes: business/marketing/forms, business/marketing/forms/responses and business/marketing/consent.

Podsakoff, P. M., MacKenzie, S. B., Lee, J.-Y., & Podsakoff, N. P. (2003). Common method biases in behavioral research: A critical review of the literature and recommended remedies. Journal of Applied Psychology, 88(5), 879–903. https://doi.org/10.1037/0021-9010.88.5.879

Statistics Canada. (2010). Survey methods and practices (Catalogue no. 12-587-X; originally published October 2003; archived). Retrieved October 10, 2026, from https://www150.statcan.gc.ca/n1/pub/12-587-x/12-587-x2003001-eng.htm

Tourangeau, R., & Yan, T. (2007). Sensitive questions in surveys. Psychological Bulletin, 133(5), 859–883. https://doi.org/10.1037/0033-2909.133.5.859

Yeager, D. S., Krosnick, J. A., Chang, L., Javitz, H. S., Levendusky, M. S., Simpser, A., & Wang, R. (2011). Comparing the accuracy of RDD telephone surveys and internet surveys conducted with probability and non-probability samples. Public Opinion Quarterly, 75(4), 709–747. https://doi.org/10.1093/poq/nfr020

Chapter 6 — Digital Research

A business that reads the traces people leave online reads them for what they are, and this chapter shows what a digital trace can tell a business, what a relative search-interest index measures, why a keyword estimate is a range, how to listen to public conversations, how to read reviews as evidence, how to analyze text at a small scale, what the business's own analytics and small tests add, where digital data mislead, and what Nordec Entretien inc. reads in a posed index over twelve months and in forty public reviews of its three competitors.

What You'll Be Able to Do

• Explain what a digital trace is and why it shows what people did rather than who they are or what a whole market thinks.
• Read a relative search-interest index for direction and timing, and refuse to turn it into a volume, a demand or a market share.
• Treat a keyword estimate as a range that the planning tool refreshes daily from the last 7 to 10 days, and date every estimate that is printed.
• Listen to public conversations at the scale of a small team, coding each item in the language it was written in and keeping its place and its date.
• Code forty public reviews or posts by hand in a shared sheet, counting themes and never reading a share of customers from them.
• Run a small test with a comparison group or a comparison period, and report its result with its size, its period and its limits.

Why It Matters

Digital data are abundant, and abundance is part of what makes them dangerous. A search-interest index, a keyword estimate, a set of public reviews and a page of the business's own analytics are all available in minutes, all free or nearly free, and all produced for a purpose other than the question the business is asking. They arrive without a frame, without a sample and without a margin, and they look like a picture of a market. The business that reads them as a picture of a market will size a territory from an index, count customers from reviews, and decide where to grow from a peak that a privacy filter put there.

The remedy is not to ignore digital traces. It is to read them for what they are, which is a record of what some people did, collected as a by-product, in a place and at a time. A relative search-interest index says where interest rises and falls; it never says how many people searched. A keyword estimate is a range that a planning tool refreshes daily; it is never a count. A set of public reviews is written by the people who chose to write; it is never a sample of a market. A business that holds those three sentences in mind can use digital evidence well, because the sentences do not forbid the evidence, they only name its shape.

There is a second reason this chapter matters, and it is about honesty with the reader of a report. A decision brief that says "search interest for our service doubled" is a brief that has already misled the person reading it, because an index cannot double in the sense of a count. A brief that says "the index for our main term rose from 38 to 100 between month 1 and month 4 of the year, on a relative scale from 0 to 100, and the rise coincides with the spring changeover" is a brief that can be checked, argued with and acted on. The difference between the two sentences is not caution. It is the difference between a finding and a claim, and the whole of this book is built on that difference.

In This Chapter

Section 6.1 defines what a digital trace is, says what it can and cannot tell a business, and sets out where this chapter stops and its neighboring books begin. Section 6.2 explains the relative search-interest index, what it measures and what two studies of search data do and do not support. Section 6.3 treats keyword estimates as ranges, with the provider's own statement about how they are refreshed. Section 6.4 covers listening to public conversations at the scale of a small team. Section 6.5 reads public reviews as evidence, with the self-selection that shapes them. Section 6.6 covers text analysis at a small scale, by hand, on forty items. Section 6.7 says what the business's own analytics add to a study and what they cannot say. Section 6.8 sets out small experiments and tests with a comparison group or a comparison period. Section 6.9 names the limits and the five traps of digital data. Section 6.10 reads the case's posed index over twelve months and its forty public reviews of the three competitors.

6.1 What Digital Traces Can Tell a Business

A digital trace is what people leave when they search, post, review or click. It is collected as a by-product, by a platform or a provider, for a purpose other than the question a business is asking. That single fact governs everything this chapter says. A trace is not a survey, because nobody was asked. It is not a census, because nobody was counted. It is not a sample in the sense Chapter 5 uses the word, because no frame was drawn and no probability of selection was calculated. It is a record of what some people did, in a place and at a time, and it becomes evidence only when the business says plainly what it is and what it cannot show.

The provider's own help page says as much about its search-interest data. The data are a largely unfiltered sample of actual search requests, anonymized, categorized and aggregated; the data are not a scientific poll and reflect searches only (Google, n.d.-a). That sentence is worth reading twice, because it is the provider describing its own product and declining to call it a poll. A business that repeats the provider's own description cannot go far wrong. A business that upgrades the description into a measurement of demand has invented something the provider did not claim.

What a digital trace can do for a business is narrow and useful. It can show that a topic exists and that people write about it in their own words, which is the raw material of a questionnaire and an interview guide. It can show when interest rises and falls, which is a question about timing rather than about size. It can show what buyers complain about, in the words they chose, which is the material Book B6 (Copywriting and Persuasive Communication) uses in a message. It can show what the people who reached the business's own site did once they arrived, which is the business's own behavior rather than the market's. Each of those is a real contribution to a study, and none of them is a count of a market.

What a digital trace cannot do is answer the questions a market count needs. It cannot say how many buildings in a territory need a service, because the people who need the service may never search for it. It cannot say what share of a market holds an opinion, because the people who post are the people who chose to post. It cannot say what a competitor charges, because a price is published only when the competitor chooses to publish it. It cannot say what a buyer will do next year, because a search is a moment and not an intention. A business that keeps this list beside the previous paragraph has the whole discipline of the chapter in two short paragraphs.

The table below sets out where this chapter stops and where its neighboring books begin. Its three columns name the neighbor, what that book holds, and what this chapter holds instead, so that a reader who needs keyword research for a page, or the business's own attribution, or the collection of reviews, is sent to the book that owns the practice rather than given a second, thinner version of it here.

Table 6.1 — Where this chapter stops and its neighbors begin

DEFINITION The words this book uses for a digital trace, a search-interest index, a keyword estimate, a public review and coding

A digital trace is what people leave when they search, post, review or click, collected as a by-product and not for the question the business is asking. A search-interest index is the provider's relative measure, scaled from 0 to 100, of how popular a search is in a place and at a time (Google, n.d.-a). A keyword estimate is the estimate of monthly searches that a planning tool shows for a keyword, as section 6.3 reads it. A public review is a rating or a comment that a person chose to publish, on a page the person does not own. Coding is labeling an item with a theme and counting the labels, so that forty reviews become four themes with four counts rather than forty impressions. These five expressions are this book's own working words, chosen so that a business can say precisely what it read and what it did with it, and none of them is a term of art borrowed from a source.

6.2 Search Interest and What It Measures

The relative search-interest index is a digital signal that a small business meets early, and it is easy to misread. The provider's help page explains the arithmetic in a paragraph. Each point is divided by the total searches of its place and time and scaled from 0 to 100, so the values show relative popularity and not search counts; low-volume terms show as 0; repeated searches by one person in a short time are removed; queries with special characters are filtered; searches made by the provider's own products are excluded; and noise is added for privacy, so low-interest queries can show misleading spikes (Google, n.d.-a). Every one of those clauses removes a little more of the reader's ability to turn the index into a number of people.

What the index does well is direction and timing. A business that reads a twelve-month series can see when interest in a term rises and when it falls, and can compare that shape with its own calendar. If the shape has two peaks and the business's own busy seasons are two, the coincidence is worth a sentence in a brief, and it is worth a sentence and no more, because a coincidence of shape is not a cause. If the shape has a peak the business cannot explain, the peak is a question for the interviews, not a finding for the report. The index is at its best when it is used to ask better questions rather than to answer them.

Two studies of search data are worth naming, at the level at which they were read. The abstract of one reports that search engine data can help forecast near-term values of economic indicators, illustrated with automobile sales, unemployment claims, travel destination planning and consumer confidence (Choi & Varian, 2012). The abstract gives no accuracy figure, and it does not support a use for a small business market, which is a limit worth stating rather than glossing: the study concerns national economic indicators, not the demand facing a maintenance contractor in a few municipalities. The abstract of another reports that large errors in flu prediction were largely avoidable, which offers lessons for the use of big data (Lazer et al., 2014). The size and the period of the errors are not in the abstract and are not printed here. What the two abstracts together support is a posture rather than a method: search data can carry a signal, and a signal read without its limits can carry an error.

The practical use of the index in a small study is therefore modest and specific. The business reads the series for its main terms, notes the peaks and the troughs, writes the values down with the date of reading, and asks what else was happening in the months of the peaks. The series goes into the brief as a shape, with the scale named, and never as a volume. If the business wants a count of a market, it counts buildings on a roll or walks a zone, as Chapter 3 and Chapter 8 describe, because those methods produce counts and an index does not.

KEY CONCEPT An index says where interest rises and falls, never how many people searched

The central idea of this section is that a relative index is a shape and not a quantity. A series that runs from 38 to 100 tells the business that interest was higher in one month than in another, on a scale the provider chose, relative to all searches in that place and time. It supports questions about timing: when does interest peak, does the peak precede or follow the business's own busy season, does a term rise while another falls. It cannot be turned into a volume, a demand or a market share, because the denominator is every search of the place and time and the numerator is a share of a total the reader never sees. The provider's own page says its values show relative popularity and not search counts (Google, n.d.-a). This reading is this book's own, drawn from the provider's description, and it is never a finding of a source.

THE RULE Search values show relative popularity, not search counts (Google, n.d.-a)

The provider's help page, read on October 10, 2026, states that each point is divided by the total searches of its place and time and scaled from 0 to 100; that low-volume terms show as 0; that the data are a largely unfiltered sample of actual search requests, anonymized, categorized and aggregated; that the data are not a scientific poll and reflect searches only; and that noise is added for privacy, so low-interest queries can show misleading spikes. The rule this book draws from that page is operational: a value from the index is printed as an index value, with its scale and its date of reading, and never as a number of searches, a number of searchers, a demand or a share. The case's own series, read as an index, is the example: the value 100 in month 4 means the highest relative popularity of the twelve months read, and it means nothing about how many people searched.

6.3 Search Volume Estimates and Their Ranges

A keyword estimate is the second digital measure a small business meets, and it is easily mistaken for a count. The provider's help page on its planning tool, read on October 10, 2026, in the part the service returned, states that the tool shows estimates of the number of searches a keyword gets each month; that keywords with very low search volume are not discoverable or forecastable; and that forecasts are refreshed daily and based on the last 7 to 10 days, adjusted for seasonality (Google, n.d.-b). Whether volumes are shown as ranges for some accounts was not stated in the part read, and this book therefore says to read the tool's own display on the day of use and prints no volume at all.

Three consequences follow from that page, and each one changes how the estimate is used. The first is that an estimate is a range and never a count, because the page calls it an estimate and because a figure built from the last 7 to 10 days cannot be a measurement of a month. The second is that the estimate is dated, because a figure refreshed daily is a figure that changes, and an estimate printed without its date of reading is an estimate that cannot be checked. The third is that a keyword with very low volume is not discoverable at all, which means that the absence of a term from the tool is not evidence that nobody searches for it; it is evidence that the tool does not show it.

The practical rule for a small study is short. A keyword estimate is used to compare terms with one another, on one day, and to decide which term deserves a page or a question in an interview guide. It is never used to size a market, never added to another estimate to produce a total, and never printed without the date on which it was read. When a business wants to know how many people are looking for a service, the honest answer is that the estimate is a range, that the range moves, and that the count the business actually needs is a count of buildings, of calls or of inquiries, which the business can obtain by other means.

The table below sets out four digital measures side by side and says, for each one, what it is, what it is not, and where the statement comes from. Its logic is to keep the four apart, because they are read together in a brief and they are not the same kind of object: two of them are the provider's own measures, and two of them are counts of what people chose to write, with no source behind them but the person who did the counting.

Table 6.2 — Four digital measures and what each is not

WHAT CHANGED The provider's keyword estimates are refreshed daily from the last 7 to 10 days and adjusted for seasonality

The provider's help page on its planning tool, read on October 10, 2026, in the part the service returned, states that forecasts are refreshed daily and based on the last 7 to 10 days, adjusted for seasonality, and that keywords with very low search volume are not discoverable or forecastable (Google, n.d.-b). The change this book records is about the shelf life of the figure rather than about the figure itself: an estimate that is rebuilt every day from a window of a week or ten days is an estimate that will differ next week, and a business that prints one without its date of reading has printed something that cannot be checked. This book therefore dates every estimate it mentions as of October 10, 2026, and prints no volume. The practical consequence is that an estimate is used to compare terms on one day, never to build a total, and never as a substitute for a count the business can obtain by counting.

6.4 Listening to Public Conversations

Listening, at the scale of a small business, is not a software function. It is a person reading public pages and writing down what is said, in the words the writers used, with the place and the date. The scale matters, because the studies that describe text mining at a market level describe something a small team cannot reproduce. The abstract of one such study reports that user-generated content in forums and similar places can be text-mined, with semantic network analysis, to learn what customers say about a firm's products and competitors' products and to draw market-structure maps; the method is shown on sedans and diabetes drugs without interviewing a consumer, and compared with structures from sales and survey data to establish validity and to highlight differences (Netzer et al., 2012). The abstract does not support the method at a small business's scale, and this book says so rather than borrowing the prestige of the study for a task it did not test.

What a small team can do is narrower and still useful. The person who runs the study picks a small number of public places where buyers of the service talk: a professional association's public forum, a public page where building owners discuss suppliers, a public review page for the trade. The person reads a fixed number of items, say forty, over a fixed period, and writes each one into a sheet with four columns: the place, the date, the language, and the topic in the writer's own words. The topics are then grouped, and the groups are counted. That is the whole method, and it produces a list of the words buyers use and the complaints they raise, which is exactly the material an interview guide needs.

Two disciplines keep the exercise honest. The first is that the reader never treats the items as a sample. The people who post are the people who chose to post, and the study of self-selection in reviews that section 6.5 describes applies here as well. A count of forty items is a count of forty items, and the brief says so. The second discipline is that the reader keeps the language of each item. In Québec, public posts and reviews about a service are written in French and in English, sometimes in the same thread, and a method that reads only one language has read only part of what is there. The reader codes each item in the language it was written in, and a business that wants both languages read says so when the work is assigned.

QUÉBEC & CANADA NOTE Public posts in Québec are written in French and in English, and personal information in them is called to the privacy book

A business reading public conversations in Québec reads them in two languages, and the reader codes each post or review in the language it was written in rather than assuming that one method reads both. A public post is public, and it can still contain personal information about the person who wrote it or about a third person named in it. The rules on personal information are Book H2's (Privacy and Personal Information), and the Québec private-sector act is called at the level reused from Books C9 and D2: the Act respecting the protection of personal information in the private sector, ss. 13 and 14, up to date to August 12, 2026, establishes that personal information is not communicated to a third person without consent, and that consent is clear, free, informed and specific. This book explains where the rules stand, describes the practice and recommends no vendor; a lawyer answers the reader's own case.

6.5 Reading Reviews as Evidence

Public reviews are accessible digital evidence for a small business, and they are easily read as if they were a survey. They are not. A review is written by a person who chose to write, about an experience that person chose to have, at a moment that person chose to record. The abstract of a study of online product reviews reports that the average online rating is biased by two self-selection biases: acquisition bias, where mostly people predisposed to like a product buy it and review it, and underreporting bias, where people with extreme opinions review more, which together produce the J-shaped distribution of reviews; consumers try to correct for the biases but cannot fully, and firms can respond through prices (Hu et al., 2017). The abstract concerns product reviews, not reviews of business services, and this book says so rather than applying the finding to a maintenance contractor as though it had been tested there.

What the abstract supports is a way of reading. A business reads reviews for the topics they raise and the words they use, and never for a share of the market. The count of reviews that mention response time is a count of the reviews read, not a percentage of customers who care about response time. The count of reviews that mention a billing surprise is a count of the reviews read, not a rate of billing errors in the industry. The distinction is not pedantry; it is the difference between a finding a business can defend in a brief and a claim that collapses the first time somebody asks how the number was obtained.

The case applies this reading to the forty public reviews of its three competitors that section 6.10 sets out. The administrator records each review with its place, its date and its language before giving it a theme, so that a reader of the research file can find the review again and see the month it came from. The theme with the largest count, response time with 14 of the 40, is not read as the market's first concern; it is read as the topic that the forty reviewers who chose to write raised most often. It becomes a question in the interview guide of Chapter 4, where the people interviewed say in their own words whether it matters to them. A theme that appears in the reviews and never in the interviews is kept in the file with that note, and a theme that appears in both carries more weight in the brief than a theme that appears in only one.

For a study of competitors, reviews have a second use that is more valuable than any count. They give the business the buyers' own vocabulary. A review that complains that nobody came back with a price is telling the business that the buyer wanted a price and did not get one; a review that praises the same technician on every visit is telling the business that continuity is noticed. Those are themes, and themes are what an interview guide is built from. Book B6 (Copywriting and Persuasive Communication) uses the customers' words in a message, and the words found in public reviews of competitors are the same kind of material, gathered from a public place.

Three rules keep the reading honest. The first is that no reviewer is ever quoted, in a brief or anywhere else, because a quotation of a real person is that person's words and this book prints none. The second is that no review site is ever named, because naming one would be a recommendation and this book makes none. The third is that the collection of reviews for the business's own offer is Book C9's (Reviews, Testimonials and Social Proof), and this chapter does not redo it; what this chapter holds is the reading of reviews as evidence about a market and about competitors, which is a different task with a different discipline.

6.6 Text Analysis at a Small Scale

Text analysis sounds like a specialist's work, and at a market level it is. The abstract of one overview reports that automated text analysis for consumer research is linked to linguistic theory, gives guidance on choosing methods and on sampling and statistical issues specific to text, and notes that the method cannot study every phenomenon but can reveal patterns in text that neither researchers nor consumers detect unaided (Humphreys & Wang, 2018). The abstract supports no tool, and this book names none. The abstract of another study reports that product reviews of 15 firms in five markets over four years are analyzed with an unsupervised topic model to extract the dimensions of quality, their valence, importance and dynamics, and to map competitive brand positions over time, and that a few dimensions capture quality (Tirunillai & Tellis, 2014). This book prints no contrast between markets.

At a small scale, the method is hand coding. The person who runs the study takes the forty items collected in section 6.4 and reads them once without writing anything, to get the range of what is there. Then the person reads them again and labels each one with a short theme, in the writer's own vocabulary where possible. Then the person groups the labels, merging those that mean the same thing, and counts the groups. The count of each theme is the finding. The whole exercise is reproducible in the sense that matters: another person given the same forty items and the same labels would produce a similar grouping, and where the two disagree, the disagreement is worth writing down.

An AI assistant can help at one point in that sequence and at no other. The person who runs the study may ask an AI assistant for a first draft of labels, then read every item again and check every label against the text, rewriting, merging and discarding as the text requires. The assistant proposes; the person decides. No finding, no theme count and no sentence of a brief is presented as the work of a tool, and nothing is reported before a person has read it. Book K4 (Artificial Intelligence for Business) treats the assistants themselves, and this chapter treats only the one narrow use: a first draft of labels that a person then checks.

TOOLS OF THE TRADE What a business needs to read digital traces: a sheet for the codes, a way to save pages, a checked text tool and a calendar

Four functions carry the work of this chapter, and none of them is a specialist product. The first is a shared sheet with a row per item and columns for the place, the date, the language, the theme and the count, so that forty items become four themes with four counts and the arithmetic can be checked by anybody. The second is a way to save a public page with its date, so that a review read in one month can be found again in another and a reader can see what was actually there. The third is a text analysis tool whose output a person checks, used for a first grouping of labels and never for a finding. The fourth is an AI writing assistant used only for a first draft of labels, with a person reading every item and rewriting every label against the text. A calendar for re-reading the index completes the set. Chapter 12 gives the functions in full, with the tool tables of that chapter.

6.7 The Business's Own Analytics as Research

The business's own analytics are the one digital source that belongs to the business, and it is easily read as though it described a market. It does not. The analytics show what the people who came did: which pages they reached, which searches brought them in, which questions they typed into a form, and when in the year they came. They cannot show who did not come, why those people did not come, or what the rest of the market does. A page that received no visits is a page nobody reached, and it is not evidence that nobody is interested, because the page may never have been shown to anybody. Book B7 (Marketing Analytics, Attribution and Dashboards) owns analytics and attribution, and this chapter does not redo that work; what this chapter holds is the narrow question of what the business's own numbers can and cannot say as research.

What they add to a study is real and specific. They show the seasonal rhythm of visits, which can be set beside the search-interest index and beside the business's own busy seasons, and a rhythm that appears in all three is a rhythm worth planning around. They show the questions people typed into a form, which are the questions a questionnaire should ask. They show which pages led to an inquiry, which is a fact about the business's own site rather than about the market. And they show the terms people searched before arriving, which is a list of the words buyers use, gathered from the business's own traffic rather than from a public page.

Three cautions belong with the reading. The first is that the numbers are the business's own, and a brief that prints them says so, because a reader who takes them for market numbers will size a market from a website. The second is that the filters matter: a period that excludes a month, a filter that removes internal traffic and a definition of a visit all change the count, and the person who pulls the numbers writes down the filters used. The third is that a small number is a small number: a page with nine visits in a month has nine visits, and a difference between nine and twelve is not a trend.

WHO DOES THIS The administrator pulls the business's own analytics and the owner reads them with the index

The work of reading the business's own analytics as research is divided differently at each size of business, and the division is what makes the reading happen. A solo owner reads the numbers once a month, beside the other evidence, and writes the reading on the same page as the index. A small team has the person who runs the study pull the numbers before the reading, so that the person who decides is not the person who assembles. An established business adds an analyst who documents the filters, the period and the definitions used, so that the numbers can be reproduced by somebody else. In the case, the administrator pulls the numbers and Sophie Nadeau reads them with the index at the same sitting, which is the arrangement a team of that size can keep. No hour and no dollar is attached to the reading here; the case's costs are posed in section 6.10 and in Chapter 13.

6.8 Small Experiments and Tests as Research

A small test is the only digital method in this chapter that can support a claim about cause, and it can support one only under a condition. The condition is a comparison: a group that did not receive the change, or a period before the change, so that a difference between the two can be attributed to the change rather than to the season, the weather or the month. Without a comparison group or a comparison period, a test produces a number and no attribution, and a business that reports the number as though it were an effect has reported something it cannot defend. The tests of a venture that has no customers are Book A1's (From Idea to Business Model, section 9.5), and this chapter holds the tests a business with customers runs on its own pages, its own messages and its own offers.

The scale of a small test is deliberately modest. A business changes one thing on one page, or sends one message to one part of its list, and compares the result with the same measure over a period when the thing was not changed. The measure is written before the test begins, because a measure chosen after the result is a measure chosen to fit the result. The period is written before the test begins, because a test that runs until it produces a pleasing number is not a test. And the result is reported with its size and its period, because a difference of three inquiries in a fortnight is a difference of three inquiries in a fortnight and nothing more.

The comparison period deserves the same care as the measure, because the seasons of a service business move every number a test reads. The case's index of section 6.10 peaks in months 4 and 10, the spring and fall changeovers, and a business with that rhythm cannot set a test run in a changeover month beside a quiet month and attribute the difference to the change. The fair comparison is the same weeks of a period in the same season, or a part of the list that did not receive the change during the same weeks. When neither is available, the report says that the test has no comparison, and it attributes nothing.

A result of a small test is the result of that test. It is not a general effect, it is not a benchmark, and it is not transferable to another page, another list or another season without being tested again. The case runs no test in this book, and nothing is credited to the study: the decision rule written before the fieldwork decides, and no contract, sale or visit is attributed to a signal, a test or a watch.

METHOD Running a small test in five steps

First, write the one change and the one measure, so that the test has a single thing to attribute and a single number to read. Second, write the comparison group or the comparison period, because without one nothing is attributed. Third, write the result that would reject the idea, before the test begins, so that the test can fail. Fourth, run the test for the period written and no longer, and resist the temptation to extend a test that is not yet pleasing. Fifth, report the result with its size, its period and its limits, and do not generalize it to another page, another list or another season. This sequence is this book's own and no source tests it; it is offered as a discipline rather than as a finding. The case runs no test in this book, and the example is described in words rather than measured.

6.9 The Limits and Traps of Digital Data

The limits of digital data are the limits of what was collected, and they are worth naming one by one because each one produces a different kind of error. The provider's own page says that its search-interest data are not a scientific poll and reflect searches only (Google, n.d.-a). The lesson of big data drawn from one well-known failure is that large errors in prediction were largely avoidable, which offers lessons for the use of big data (Lazer et al., 2014); the size and the period of the errors are not in the abstract and are not printed here. Reviews carry the self-selection that section 6.5 described (Hu et al., 2017). Text carries sampling and statistical issues specific to text, which one overview names as a reason to choose methods with care (Humphreys & Wang, 2018). And a link posted in public gives no sample at all: the page of a national research council, read on October 10, 2026, states that open-link surveys carry risks of bias, fraudulent and duplicate answers, manipulation and lack of representativeness, that they must disclose their limits, and that they must never be combined with sample data (Canadian Research Insights Council, n.d.-a). That page binds the council's members and is self-regulation, never a law, and this book uses it for one purpose: to say that a link posted in public is not a sample.

The traps are where those limits meet a business's wish for a number. The table below names five of them, with a sign that the trap has been entered and the practice that replaces it. Its three columns are the trap, the sign and the remedy, and its logic is that a trap is easier to avoid when the sign is written down in advance, because the sign is what a reader notices in a draft before the draft leaves the building.

Table 6.3 — Five traps of digital data

WATCH OUT A peak in an index, a burst of reviews or a topic in posts looks like a market and is a signal

The trap this section names is the one that looks most like success: a peak in an index, a burst of reviews or a topic that keeps appearing in public posts, read as the demand of a market. The signs are easy to spot in a draft. A peak is written as a count. An anecdote is written as a share. A week of posts is written as a change in the market. The abstracts of the studies cited above say why the reading fails: reviews are shaped by self-selection (Hu et al., 2017), and large errors in prediction were largely avoidable (Lazer et al., 2014). The remedy is one sentence long and belongs in every brief: say what the signal is, who produced it, and what it cannot show. A peak in an index is a peak in an index, produced by searches, and it cannot show how many people searched. The reasoning here is this book's own.

6.10 The Case's Digital Signals

Nordec Entretien inc. is a fictional business created for this encyclopedia; its owner, its employees, its customers, its competitors, its figures and the events described are invented, and any resemblance to a real business is coincidental. The digital signals of the case are posed by this book and illustrative. The administrator reads a relative search-interest index for the business's main service terms over twelve months of the case's year, and codes forty public reviews of the three competitors. Both readings are done by gate 2, on day 20 of the plan of Chapter 13, and both are written into the research file rather than into the prospecting file.

The index runs from 0 to 100 and reads 38, 44, 71, 100, 83, 52, 41, 45, 74, 92, 66 and 40 for months 1 to 12 of the case's year. The peaks fall in months 4 and 10, which are the spring and fall changeovers of Book B4's (The Marketing Plan, Budget and KPIs) plan seasons, cited as that book's. The values are an index relative to all searches of the place and time, and they are not a number of searches. The lowest value is 38 in month 1 and the highest is 100 in month 4, and the gap between them is 62 points on a relative scale. The administrator writes the series into the research file with the date of reading and with one sentence saying what it is.

The forty public reviews of the three competitors are read and coded by theme. Fourteen name response time, nine name billing surprises, eight name the same technician each visit, and nine fall into other themes. The count of forty is a count of the reviews that were read, written by people who chose to write, and it says nothing about the share of customers who hold any of those views. The themes are used for one purpose in the case: they supply the vocabulary for the interview guide of Chapter 4, and they are set beside the themes the interviews produce rather than substituted for them.

Figure 6.1 plots the case's search-interest index over twelve months and sets beside it the themes of forty public reviews of the three competitors. On the left, twelve orange bars with two peaks and an amber note; on the right, four horizontal bars for the review themes.

Figure 6.1 — Search interest and review themes. On a relative index from 0 to 100, search interest peaks in months 4 and 10, the spring and fall changeovers, and the values are an index relative to all searches, not a number of searches. Of 40 public reviews of the three competitors, 14 name response time, 9 billing surprises and 8 the same technician each visit.

Read point by point, the figure says this. The first bar is 38, the lowest of the twelve months. The second is 44 and the third is 71, a rise that begins before the first peak. The fourth bar is 100, the highest value of the year, and it falls in the month of the spring changeover. The fifth is 83 and the sixth is 52, a decline across two months. The seventh is 41 and the eighth is 45, the quietest stretch of the year. The ninth is 74 and the tenth is 92, the second peak, in the month of the fall changeover. The eleventh is 66 and the twelfth is 40, a return toward the level of the first month. Beside the twelve bars, four horizontal bars carry the review themes: response time 14, billing surprises 9, the same technician each visit 8, and other 9.

The table below gives the same four themes as counts of the forty reviews read, and it is the form in which the coding enters the research file. Its two columns are the theme and the count, and what it retains is the fact that the counts add to forty, which is the number of reviews read and not a number of customers.

Table 6.4 — The themes of the case's forty public reviews, posed by this book

DO THE MATH The case's index peaks at 100 in month 4, and a gap of 62 points is not 62 more searches

Illustrative figures; the case is fictional. Step 1 — The index has twelve monthly values, from 38 to 100, read on a relative scale from 0 to 100. Step 2 — The highest value is 100 in month 4 and the second highest is 92 in month 10, which are the two changeover months of the case's year. Step 3 — The lowest value is 38 in month 1. Step 4 — The gap between the highest and the lowest is 62 points, which is 100 minus 38, on a relative scale, and it is not 62 more searches, because the index is relative and not a count. Step 5 — The 40 reviews split into 14, 9, 8 and 9 across four themes, which counts what 40 reviewers chose to write and says nothing of the share of customers. Nothing is credited to the study, and the decision rule written before the fieldwork decides.

CASE IN POINT Nordec Entretien inc. reads a twelve-month index and forty public reviews of its three competitors

Nordec Entretien inc. is a fictional business created for this encyclopedia; its owner, its employees, its customers, its competitors, its figures and the events described are invented, and any resemblance to a real business is coincidental. The administrator reads the index and codes the reviews, and Sophie Nadeau reads the result at gate 2, on day 20 of the plan of Chapter 13. The index peaks in months 4 and 10, which are the spring and fall changeovers of Book B4's plan seasons, and the forty reviews code to four themes: response time 14, billing surprises 9, the same technician each visit 8, and other 9. The three competitors are the regional integrator, Competitor B and Competitor C, and none of them is named. The index values, the forty reviews and the four counts are posed by this book and illustrative. No reviewer is quoted, no verdict is given on any competitor, and nothing is credited to the study.

IN LA MACHINE La Machine listens to no social network, reads no review and holds no market data

In the application copy read on October 10, 2026 (La Machine, 2026), there is no listening to social networks, no reading of competitors' reviews, no monitoring of competitors' pages and no market data. The reader keeps the index and the review codes in a sheet outside the product, and the case does not use La Machine for its digital signals. Check it in your installation, because what an installation shows is a fact about that installation. La Machine is described here and never sold, and this book is valid with any tool.

Costly Mistakes

• Reading a relative index as a volume. A series that runs from 38 to 100 says that interest was higher in one month than in another on a scale the provider chose, and a brief that turns it into a number of searches has printed a figure nobody can check.
• Printing a keyword estimate without its date of reading. An estimate refreshed daily from the last 7 to 10 days is a figure that changes, and a figure printed without its date cannot be compared with anything.
• Reading reviews as a sample of a market. The people who write reviews are the people who chose to write, and a count of forty reviews is a count of forty reviews rather than a share of customers.
• Quoting a reviewer or naming a review site. A quotation of a real person is that person's words, and naming a site would be a recommendation; the case gives themes, codes and counts.
• Combining answers from an open link with answers from a drawn sample. A link posted in public gives no sample at all, and the page of a national research council states that such answers must never be combined with sample data (Canadian Research Insights Council, n.d.-a).
• Reporting a small test without a comparison group or a comparison period. Without one, nothing is attributed, and a number reported as an effect is a number the business cannot defend.
• Crediting a sale, a contract or a visit to a digital signal. A peak in an index, a theme in reviews and a result of a test are signals, and the case credits nothing to the study; the decision rule written before the fieldwork decides.

Key Takeaways

• A digital trace is what people leave when they search, post, review or click, collected as a by-product, and it shows what people did rather than who they are or what a whole market thinks.
• A relative search-interest index is scaled from 0 to 100 and shows relative popularity, not search counts; the provider's own page says the data are not a scientific poll and reflect searches only (Google, n.d.-a).
• A keyword estimate is a range that a planning tool refreshes daily from the last 7 to 10 days and adjusts for seasonality, and a keyword with very low volume is not discoverable or forecastable (Google, n.d.-b).
• Public reviews and public posts are written by those who chose to write, so they are read for their themes and their words and never for a share of a market, and no reviewer is ever quoted.
• Text analysis at a small scale is hand coding: read forty items, label each with a theme, group the labels and count the groups, with a person checking every label against the text.
• The business's own analytics show what the people who came did, and they cannot show who did not come, why, or what the rest of the market does; a small test attributes a difference only when it has a comparison group or a comparison period.

Your Action Plan

1. Write down what each digital source is before reading it. For every source the study will use, write one sentence naming what it holds, who produced it and what it cannot show, and keep the sentences with the research file. Deliverable: one page of source descriptions. Verification: every sentence names a limit, and no sentence turns an index into a volume or a review into a share.
2. Read the search-interest series and write it as a shape. Pull the twelve-month series for the main service terms, note the peaks and the troughs, and write the values with the scale and the date of reading. Deliverable: one dated page holding the series and one sentence saying what it is. Verification: the page prints index values and no volume, and the sentence names the scale.
3. Collect and code forty public items by hand. Choose the public places, read a fixed number of items over a fixed period, and record the place, the date, the language and the theme of each in a shared sheet. Deliverable: a sheet of forty rows and a count of themes. Verification: every row has a language and a date, the counts add to forty, and no reviewer is quoted.
4. Pull the business's own analytics with the filters written down. Take the visits, the searches that led visitors in and the questions typed into a form, and record the period, the filters and the definitions used beside the numbers. Deliverable: one dated page of the business's own numbers with its filters. Verification: the page says that the numbers are the business's own, and the filters can be reproduced by another person.
5. Write the small test before running it, or run none. If a test is run, write the one change, the one measure, the comparison group or period, and the result that would reject the idea, and report the result with its size, its period and its limits. Deliverable: a one-page test plan and a one-page result. Verification: the result is not generalized beyond the test, and no contract or sale is credited to it.

In La Machine

The application copy read for this book (La Machine, 2026) carries no digital research function, and the honest description of that state is short. La Machine can host a simple questionnaire as a form whose answers become contacts, keep competitors as company records with dated notes, remind a team of its weekly watch through recurring alerts and tasks, and stage public business listings for prospecting; it draws no sample, computes no margin of error, offers no rating scale, records or transcribes no interview, listens to no social network, monitors no competitor and holds no market data. For the subject of this chapter, the last part of that sentence is the whole of it: there is no listening to social networks, no reading of competitors' reviews, no monitoring of competitors' pages and no market data in the copy read on October 10, 2026. The reader keeps the index and the review codes in a sheet outside the product, and the case does not use La Machine for its digital signals.

What the platform does carry is the work around the reading, and that is worth naming so that a business does not expect the reading itself to be built in. A competitor is a company record with dated notes, which is where a note about a public page can sit beside the date it was read. A watch is a recurring reminder and a task, which is where the re-reading of an index can be placed on a calendar. A form can hold a simple questionnaire whose answers become contacts, which is a different task from the survey of the case, run by the fieldwork service outside the application. Check it in your installation, because what an installation shows is a fact about that installation. La Machine is described, never sold, and this book is valid with any tool.

Self-Check

1. What is a digital trace, and what are the three things it cannot tell a business?
2. What does a relative search-interest index measure, and why can it never be turned into a volume?
3. Why is a keyword estimate a range rather than a count, and what does the provider's page say about how it is refreshed?
4. Why are public reviews read for their themes rather than for a share of a market, and what does the study of self-selection report?
5. What makes a small test able to support a claim about cause, and what does a result of a small test never become?

Answers. 1. A digital trace is what people leave when they search, post, review or click, collected as a by-product and not for the question the business is asking. It cannot say how many buildings in a territory need a service, because the people who need it may never search for it; it cannot say what share of a market holds an opinion, because the people who post are the people who chose to post; and it cannot say what a buyer will do next year, because a search is a moment and not an intention. 2. A relative search-interest index measures how popular a search is in a place and a time, scaled from 0 to 100, with each point divided by the total searches of that place and time (Google, n.d.-a). It cannot be turned into a volume because the denominator is every search of the place and time and the numerator is a share of a total the reader never sees, and the provider's own page says its values show relative popularity and not search counts. 3. A keyword estimate is a range because the provider's page calls it an estimate and because it is built from the last 7 to 10 days; the page states that forecasts are refreshed daily and based on the last 7 to 10 days, adjusted for seasonality, and that keywords with very low search volume are not discoverable or forecastable (Google, n.d.-b). 4. Reviews are read for their themes because the people who write them chose to write, so a count of themes is a count of the reviews read and not a share of customers; the abstract of the study of self-selection reports that the average online rating is biased by acquisition bias, where mostly people predisposed to like a product buy it and review it, and by underreporting bias, where people with extreme opinions review more, which together produce the J-shaped distribution of reviews (Hu et al., 2017). 5. A small test supports a claim about cause only when it has a comparison group that did not receive the change or a comparison period before the change, so that a difference can be attributed to the change rather than to the season or the month; a result of a small test is the result of that test, with its size and its period, and it never becomes a general effect or a benchmark.

Chapter References

Act respecting the protection of personal information in the private sector, CQLR c P-39.1, ss. 13 and 14 (Québec). LégisQuébec, up to date to August 12, 2026 — read October 9, 2026. https://www.legisquebec.gouv.qc.ca/en/document/cs/P-39.1

Canadian Research Insights Council. (n.d.-a). CRIC requirements for reporting survey precision and on the use of open-link surveys. Retrieved October 10, 2026, from https://www.canadianresearchinsightscouncil.ca/cric-requirements-for-reporting-survey-precision-and-on-the-use-of-open-link-surveys/

Choi, H., & Varian, H. (2012). Predicting the present with Google Trends. Economic Record, 88(s1), 2–9. https://doi.org/10.1111/j.1475-4932.2012.00809.x

Google. (n.d.-a). FAQ about Google Trends data. Trends Help. Retrieved October 10, 2026, from https://support.google.com/trends/answer/4365533

Google. (n.d.-b). Use Keyword Planner. Google Ads Help. Retrieved October 10, 2026, from https://support.google.com/google-ads/answer/7337243

Hu, N., Pavlou, P. A., & Zhang, J. (2017). On self-selection biases in online product reviews. MIS Quarterly, 41(2), 449–471. https://doi.org/10.25300/MISQ/2017/41.2.06

Humphreys, A., & Wang, R. J.-H. (2018). Automated text analysis for consumer research. Journal of Consumer Research, 44(6), 1274–1306. https://doi.org/10.1093/jcr/ucx104

La Machine. (2026). Documentation — route reference inventory. Internal product documentation, read October 10, 2026, together with the code of the pages this book describes: no route is named in this chapter.

Lazer, D., Kennedy, R., King, G., & Vespignani, A. (2014). The parable of Google Flu: Traps in big data analysis. Science, 343(6176), 1203–1205. https://doi.org/10.1126/science.1248506

Netzer, O., Feldman, R., Goldenberg, J., & Fresko, M. (2012). Mine your own business: Market-structure surveillance through text mining. Marketing Science, 31(3), 521–543. https://doi.org/10.1287/mksc.1120.0713

Tirunillai, S., & Tellis, G. J. (2014). Mining marketing meaning from online chatter: Strategic brand analysis of big data using latent Dirichlet allocation. Journal of Marketing Research, 51(4), 463–479. https://doi.org/10.1509/jmr.12.0106

Chapter 7 — Competitive Intelligence and Monitoring

A competitor is watched through what it publishes, and this chapter says what competitive intelligence is and is not, how a business picks the competitors that matter, what a competitor profile holds, which signals are worth watching, which public sources carry them, how talking to customers, suppliers and former employees stays honest, where the legal and ethical line falls, how a small team builds a watch it can keep, and how a signal becomes an alert and then a decision.

What You'll Be Able to Do

• Distinguish competitive intelligence from spying, from the one-day strategic analysis of Book A3 (Strategy and Competitive Positioning), from the business's own figures and from the voice of the customer, and say which of those four a given question belongs to.
• Choose the competitors worth watching by two tests — whether a competitor meets the same customer, and whether its move would change a decision the business has already framed — and keep the list short.
• Build a competitor profile as a page of dated notes that holds what the competitor publishes, where each fact was read and how sure the reader is, and that never holds a guess presented as a fact.
• Log a signal as a dated fact from a public source, keep the reading apart from the fact, and flag the signals that meet a threshold written in advance.
• Keep the watch inside the legal and ethical line by reading only what competitors publish, by identifying yourself in any conversation, and by never asking anyone for a confidential price, bid, customer list or plan.
• Run a watch a small team can keep — a weekly round, a monthly review, a dated log and a named owner — and turn a signal into a written decision or into nothing at all.

Why It Matters

A business that has customers already knows a great deal about its competitors without ever having written any of it down. The owner hears a name at a trade show, a technician reports what a rival quoted for a rooftop unit, a customer mentions that another firm called twice last month, and all of it lives in memory, undated and unverified. Memory holds the vivid and drops the ordinary: the quote that surprised somebody is remembered; the tenders a competitor quietly won are not. A watch moves that knowledge into a dated list, where it can be compared with a decision rule that was written before the watching started.

The second reason a watch matters is that the alternative to watching is guessing. A business that does not know what its competitors publish fills the gap with a story, usually about the competitor's prices being lower, its trucks newer or its owner knowing somebody, and none of those stories can be checked. A watch replaces the story with a fact and its date: a public price page that changed, a job posting for a second dispatcher, a new establishment entered in the enterprise register, a tender notice published by a public body. A fact can be wrong, but it can be corrected, and a story cannot.

The third reason is the one that decides whether the watch survives its first quarter: a watch that tries to see everything sees nothing, because the person who runs it stops running it. The watch this chapter describes is deliberately small. Nordec Entretien inc., the fictional business followed through this book, keeps a watch of that size on the regional integrator, Competitor B and Competitor C, and the figures of that watch (eleven signals in ninety days, thirty minutes a week, sixty minutes a month, about nine and a half hours a quarter) are posed by this book and illustrative, never a norm. A watch that is kept produces a dated record, and a dated record is what turns a signal into a decision.

In This Chapter

Section 7.1 defines competitive intelligence as this book uses the words and sets the chapter's boundary against the books around it. Section 7.2 shows how a business identifies the competitors that matter and how many to watch. Section 7.3 gives the competitor profile and the six fields it holds. Section 7.4 names the signals worth watching and what each can and cannot establish. Section 7.5 lists the public sources of competitor information and what each one says. Section 7.6 covers conversations with customers, suppliers and former employees and the honesty they require. Section 7.7 draws the legal and ethical line as published rules called to the books that own them. Section 7.8 builds a watch system a small team can keep. Section 7.9 follows a signal from the log to an alert to a written decision. Section 7.10 applies all of it to the watch Nordec Entretien inc. keeps on three competitors.

7.1 What Competitive Intelligence Is and What It Is Not

Competitive intelligence, as this book uses the words, is the ordered gathering and reading of public facts about competitors, customers and the environment so that a decision the business has already framed can be taken with better evidence. The field is read here through one source. A bibliometric assessment of the competitive intelligence literature, read at abstract only, supports the view that competitive intelligence is mainly an activity of environmental scanning and strategic management, and that practitioners need a keen understanding of all business functions, especially marketing and strategic planning (Calof & Wright, 2008). That abstract reports no prevalence of any practice, and this book prints none: there is no source in this book for a share of businesses that run competitive intelligence, and no size, growth or count of the intelligence industry is printed anywhere in it.

The words carry a weight they should not. Competitive intelligence is not spying. Spying obtains something that its owner has not published, usually by deceit, and the line between the two is not a matter of degree but of kind. A watch that reads a competitor's public price page, its job postings, its press releases and the tender notices a public body publishes is reading what somebody chose to make public. A watch does not telephone a competitor's supplier under a false name and does not ask a former employee for a confidential document; a watch that does has crossed into something else entirely, and section 7.7 sets out the published rules that draw that line and calls them to the books that own them.

Competitive intelligence is also not the one-day strategic analysis of Book A3 (Strategy and Competitive Positioning), which holds the five forces in section 2.3 and the annual strategy refresh in section 10.4. That analysis asks where the business stands in its industry and is performed once, or once a year, in a room. The watch this chapter describes is continuous and small: it reads a short list of public sources every week and reviews what it found once a month. The two are complements: Book A3 names the regional integrator, expected to bid a maintenance component at about $560 a month, in chapter 1 of that book, and this chapter makes continuous the watch that Book A3 only begins. The figure belongs to Book A3 and is cited as that book's.

Competitive intelligence is not the business's own figures either: what a business sold, what it spent and what its pipeline holds belong to Book B7 (Marketing Analytics, Attribution and Dashboards). Nor is it the voice of the customer, which Book B10 (Customer Experience, Retention and Loyalty) holds in chapter 4. A watch may read a customer's remark as a signal, and when it does, the remark is a dated fact about what one person said, not a measurement of a market.

The boundary is worth a table, because a business that confuses these four activities will do one of them badly while believing it is doing another. The table below sets six neighbors beside what each holds and what this chapter holds, and it opens the chapter's work by saying where the chapter stops. The last of the six rows names two books, with their titles and nothing about their content.

Table 7.1 — Where this chapter stops and the books around it start

This book describes published rules and practices, is not legal, tax or accounting advice, and recommends no vendor, no platform, no service and no tool. Where a rule is stated in this chapter, the same section sends the reader to the book that holds it: Book H4 for agreements between competitors and the Bureau's guidelines, Book H5 for trade secrets and fair dealing, Book H2 (Privacy and Personal Information) for personal information and consent, Book H3 (Anti-Spam and Telemarketing Compliance) for electronic messages and calls, and Book H1 (Contracts and Commercial Agreements) for a contract. A question about a particular exchange, call or use of material is a lawyer's question, and this book never answers it.

DEFINITION Competitive intelligence, signal, alert, watch log and watch system, as this book uses the words

Competitive intelligence is the ordered gathering and reading of public facts about competitors, customers and the environment so that a decision already framed can be taken with better evidence. A signal is a dated fact from a public source that bears on a decision. An alert is a signal that meets a threshold written in advance. The watch log is the dated list of signals, each with its source and its reading kept apart from the fact. The watch system is the short list of sources, the log, the rhythm and the owner. These are this book's working words, and no source read defines them in this sense. The field is read here as mainly an activity of environmental scanning and strategic management (Calof & Wright, 2008, abstract only), and that abstract reports no prevalence of any practice.

7.2 Identifying the Competitors That Matter

A competitor, for the purpose of a watch, is whoever the customer would call instead. That definition is wider than the list a business usually keeps. It includes the firm that does the same work in the same territory, and the substitute: the general contractor who offers a maintenance component inside a larger contract, the equipment supplier who sells a service agreement with the unit, the national firm whose local branch answers a call. It includes the customer's own staff, because a building with two technicians on payroll will not sign a maintenance contract this year, and a watch that does not count that possibility will misread a quiet quarter as a lost one.

A business finds its competitors in three ways, and each is a method rather than a finding. The first is to read who appears in the same tender notices. A public body that publishes a call for tenders publishes it to everybody, and the firms that answer it are, by that act, competitors for that work. The second is to ask the customers, in the interviews of Chapter 4, who else they called before they called the business, and to write the answers down as themes rather than as a ranking. The third is to read the losses: when a proposal is not signed, the reason recorded in the sales file may name the firm that won, and a year of those reasons is a list of competitors assembled from the business's own experience rather than from anybody's opinion.

How many competitors to watch is a question with a short answer, and the answer comes from the decision rather than from the market. A business watches a few, chosen because a move by one of them would change something the business has already decided to do. The first test is whether the candidate meets the same customer: a firm that works in another city, on another kind of building, or on another kind of work is not a competitor for the decision at hand, however large it is. The second is whether its move would change the decision: if the business is deciding where next year's growth should go, then a competitor that wins public maintenance tenders in the territory under consideration matters, and a competitor that does not is background.

The two tests are what keep a watch small enough to be kept. A list of twelve competitors produces a log nobody reads, and a log nobody reads is worse than no log, because it looks like evidence; a list of three can be held in mind and read in a monthly review. The number is not a rule and no source read gives one; it is a consequence of the decision, and a business that changes its decision may change its list.

Nordec Entretien inc. watches three competitors: the regional integrator, Competitor B and Competitor C, as section 7.10 sets out. The list is short because the decision is narrow, and the chapter does not pretend that three is the right number for every business.

Whether a competitor's name may be used in a comparison, an advertisement or a sales conversation is a question of advertising and competition law, and it is Book H4's; the watch reads what competitors publish and answers nothing about what the business may then say about them.

KEY CONCEPT A signal is a dated fact from a public source, and it matters only when a threshold written in advance turns it into a decision

The central idea of this chapter is that a watch is a chain of three links, and that a chain missing any one of them is not a watch. A fact without a date cannot be compared with anything, because a business cannot tell whether a price page changed last week or three years ago. A fact without a source cannot be checked, and a fact that cannot be checked will be repeated until it becomes a belief. A fact without a threshold cannot call for a decision, because nothing in the business has said in advance what would count as a reason to act. The watch reads what competitors publish and obtains nothing by deceit; that is the line, and it is a line of kind rather than of degree. No source read gives a hit rate for any signal, and the counts of the case in this chapter are posed by this book and illustrative.

7.3 The Competitor Profile

A competitor profile is the page that holds what is known about one competitor, with the date of every item. It is not a dossier and it is not a report. It is a working page, kept current, read before a decision and amended when something new is published. Its purpose is to answer, in a minute, the questions a business actually asks: what does this firm offer publicly, whom does it serve, where does it work, what has it announced, and how sure am I of each of those things.

The profile holds six kinds of content, and the discipline is in the last two. It holds who the competitor is, as the enterprise register gives it; what the competitor offers publicly, on its own pages; whom it serves and where it works, as its published material claims, which is a statement about its marketing and not about its client list; what it has announced, such as a new service, a new establishment, a hiring push or a contract award that a public body has published; and the signals logged, each with its date and its source.

What the profile must never hold is a guess presented as a fact. A page that says a competitor is "aggressive on price" is not a profile; it is an opinion with no date and no source, and it will be read a year later as though it were evidence. A page that says the competitor's public price page listed a monthly rate for a defined service on a named date is a fact, and it can be compared with the same page read six months later. The difference is not pedantry. It is the difference between a record that can be checked and a story that cannot.

The profile is also not a verdict on the competitor. It is the competitor's published picture, incomplete by construction: a competitor that never publishes a price is a competitor whose price is not public, and a competitor with numerous reviews is not necessarily one with many customers, because reviews are written by those who chose to write.

In La Machine, a competitor is kept as dated notes on a company record (CRM → Companies business/marketing/companies), which has no field for a competitor's price, position or strengths, so the discipline of dating every item is the team's rather than the page's; Book N4 (CRM, Contacts and Prospect Data) describes the company records themselves.

The table below sets six fields of a profile beside what each holds, where it is read, when it was read, how sure the reader is and what the field must never contain. The six columns keep the table readable aloud, and the table is the book's method rather than a finding about any competitor.

Table 7.2 — The competitor profile in six columns

WATCH OUT A profile that becomes a story about the competitor

One way a competitor profile stops being useful is that it fills with sentences nobody can check. "They are cutting prices." "They are going after our customers." "They have a new truck." The trap is easy to fall into because the sentences are short and the facts are long: writing that a competitor's public price page listed a defined monthly rate for a defined service on a named date takes a whole line, and writing that the competitor is cheap takes four words. The remedy is a rule the profile keeps: every line carries a source and a date, and a line that cannot carry both is moved to a separate page of questions. The questions page is never read as evidence.

7.4 Signals Worth Watching

A signal is a dated fact from a public source that bears on a decision. The definition has three parts and all three matter. It is a fact, so it is something that was published and can be pointed to. It is dated, so it can be placed in time and compared with the same fact read later. And it bears on a decision, so it is not merely interesting; it is connected to something the business has already decided to do or to decide. A fact that fails the third part is not a signal, and a watch that logs everything it finds will drown in its own log.

The case logs five kinds of signal, and each kind is worth reading for what it can suggest and for what it cannot establish. Public tenders and awards are the first kind: a public body publishes a call for tenders, and a public body publishes an award. A tender notice shows that work is being bought in a territory and on a schedule; it does not show who will win it. An award notice shows that a contract was awarded; it does not, on the page read for this book, show the amount or the losing bidders. Job postings are the second kind: a posting for a second dispatcher suggests that a competitor expects more calls, and it proves nothing about how many calls there are. A change to a public price or offer page is the third kind: it is a fact about what the competitor published on the day it was read, and it is not a statement about what the competitor charges its customers. A new establishment entered in the enterprise register is the fourth kind: it is a fact about a registration, and it does not establish that the establishment is operating or what it does. A burst of public reviews is the fifth kind: it is a fact about how many reviews appeared in a period, and it is not a measure of how many customers the competitor has, because reviews are written by those who chose to write.

Two studies read at abstract only show what can be done with text at a scale far larger than a small business's, and they are printed here as studies of their own settings rather than as methods to copy. One text-mines user-generated content to draw market-structure maps of sedans and diabetes drugs (Netzer et al., 2012); the other analyzes product reviews of fifteen firms in five markets over four years with an unsupervised topic model to map competitive brand positions (Tirunillai & Tellis, 2014). Neither abstract supports the method at a small business's scale, and Chapter 6 reads both more closely.

A signal is logged with its date, its source and its reading, and the reading is kept apart from the fact, as Chapter 9 develops for findings and interpretations. The fact is that a job posting appeared on a named date; the reading is that the competitor may be preparing for more work. A reader who disagrees with the reading can still use the fact, and a log that merges the two cannot be audited.

The case's log holds eleven signals in ninety days, and the figure is posed by this book and illustrative. The eleven split into four public tenders and awards, three job postings, two changes to a public price or offer page, one new establishment in the enterprise register and one burst of public reviews. The split is not a distribution to expect and not a benchmark; it is what the case's log holds on the day it is read, and a business that keeps its own log will find its own split.

METHOD The weekly watch round in five steps

First, open the list of sources and read each page for the changes of the day, in the same order every week, so that nothing is skipped by accident. Second, log each change as a signal with its date, its competitor, its kind and its source, and nothing else. Third, write the reading apart from the fact, in its own column, and mark it as a reading. Fourth, flag any signal that meets a threshold written in the decision rule, and put it on the agenda of the monthly review. Fifth, close the round against its verification criterion, which is that every source on the list was read and every change found was logged on the day it was found. This is the book's reasoning and not a tested procedure; in the case the round takes thirty minutes a week, posed by this book, and the monthly review takes sixty minutes.

7.5 Public Sources of Competitor Information

The register of enterprises of Québec is the first public source and the one that answers the identity question. It can be consulted free of charge, by Québec enterprise number, by name or by a related word, and a natural person can be searched by name to see the enterprises that person is linked to. An enterprise's public file shows its number, its registration status, the names it uses, the addresses of its establishments, the contact details of its directors and officers and the names of its ultimate beneficial owners (Gouvernement du Québec, 2025). The page read carries no rule on reusing that information for solicitation, and this book supplies none: the reuse of personal information is Book H2's question, the rules on electronic messages and calls are Book H3's, and a business that intends to contact the people named in an entry takes the question to those books and to a lawyer.

The electronic tendering system of Québec is the second source and the one that answers the question of what public bodies are buying. It carries public calls for tenders, notices of intent for a contract by mutual agreement and notices of interest from ministries and public bodies, the health and social services network, the education network and the municipalities, with tender documents such as specifications and addenda (Gouvernement du Québec, 2026). The page read does not support contracts awarded, bidders' names or amounts: a business that wants to know who won a tender looks for the award where the public body publishes it, and verifies what the system shows on the day of use, because the system changes. Book E9 (Tenders, RFPs and Public Procurement) treats where public buyers announce what they buy and what a provincial tendering system publishes, in its chapter 2.

The competitor's own pages are the third source, and they are the richest and the least reliable, because they say what the competitor chose to say. A price page is what the competitor published on the day it was read, not a statement of what it charges, because published and negotiated prices differ and a page can be changed the next morning. A press release is a fact about what was announced rather than about what happened.

Public reviews are the fourth source, and they require the most care. A review is written by somebody who chose to write it, so the set of reviews is not a sample of a competitor's customers. A study read at abstract only reports that the average online rating of products is biased by two self-selection biases, an acquisition bias and an underreporting bias, which together produce the characteristic shape of review distributions (Hu et al., 2017); it concerns product reviews, supports no claim about reviews of business services, and this book prints no figure from it. A burst of reviews is a fact about how many reviews appeared, and nothing more. The collection of a business's own reviews is Book C9's (Reviews, Testimonials and Social Proof).

A purchased industry report is the fifth source, and it raises a copyright question this book does not answer: whether a given copy of a report or of a competitor's published material is fair dealing is a lawyer's question, and the license terms of a purchased report govern what may be done with it. Both are Book H5's, and section 7.7 states the rule of the Copyright Act.

The table below sets six public sources beside what each shows and does not show, when each is read and which book owns the rule that governs it. Every source named is a page read on a date, and the table is a map of what a watch can see rather than a list of what it should buy.

Table 7.3 — Public sources of competitor information and what each says

7.6 Talking to Customers, Suppliers and Former Employees

A conversation is a legitimate source of competitor information when three conditions hold: the person speaks freely, the person knows who is asking and why, and the person is never asked for something that is not public. The first two conditions are not courtesies; they are the practice of a research code that binds its members and the associations that adopt it, and never a business by law. The code requires that researchers identify themselves promptly, that participation be voluntary and informed at the start, and that a participant may withdraw at any time (International Chamber of Commerce & ESOMAR, 2025, art. 1 and 4). A conversation held under those conditions is honest research. A conversation held without them is something else, and section 7.7 draws the line.

What a customer may be asked is a question about the past. A customer can be asked which firms were called before the business was called, what was said in those conversations, what was quoted and what made the difference. That is a question about a decision the customer already made, and the answer is a fact about the customer's experience. What a customer should not be asked is to report a competitor's confidential price, bid or plan, because the customer may not know it and, if the customer does know it, the customer may be bound by something that makes repeating it a problem. The interviews of Chapter 4 are the place where these conversations belong, and their recruiting, their guides and their coding are that chapter's method.

A supplier may be asked about what is public and about the supplier's own experience. A supplier who sells to several firms in a territory knows what is generally known in the trade, and a question about a published change, a new establishment or a hiring push is a question about something the supplier can answer without disclosing anybody's confidence. A supplier should not be asked to hand over a competitor's order book, its prices or its plans, and a business that asks has put the supplier in a position it should not be in. The same holds for a former employee of a competitor. A former employee may be asked about the former employee's own experience and about what is public; a former employee should not be asked for a document, a customer list, a price list or a plan, and the reason is not only ethical. Section 7.7 sets out the published rule.

The Bureau's guidelines on competitor collaboration, dated May 6, 2021, say that exchanges of current or future information between competitors are riskier than exchanges of historical information (Competition Bureau, 2021); section 7.7 returns to them, and a verdict on a given exchange is a lawyer's question and Book H4's.

The safeguard this chapter adds is a rule about the notes. Every item in the watch log records where it came from, and an item whose origin is doubtful is not used. That rule sounds obvious and it is easy to break, because a useful fact with a vague origin is tempting. A business that writes "heard from a supplier" beside a fact has written a fact nobody can check, and a fact nobody can check will be repeated in a meeting as though it were established. The rule costs one line per item and it is what makes the log worth keeping.

Nordec Entretien inc. talks to its own customers and former customers in the interviews of Chapter 4 and asks about their own past choices; it never asks a former employee of a competitor for anything. The outside lawyer reviews the script of any conversation of this kind before it is held, and a business that is unsure whether a question is on the right side of the line takes it to a lawyer rather than to the conversation.

7.7 The Legal and Ethical Line

The line a watch must not cross is drawn by published rules, and this section states them as they are written and sends each to the book that owns it. The first is the Competition Act, R.S.C. 1985, c. C-34, s. 45(1) and (2), in Canada, current to September 21, 2026. Under s. 45(1) it is an offense for a person, with a competitor in respect of a product, to conspire, agree or arrange to fix, maintain, increase or control the price of its supply, to allocate sales, territories, customers or markets, or to fix, maintain, control, prevent, lessen or eliminate its production or supply. Under s. 45(2) the offense is indictable, and the fine is left to the court; no amount of a penalty is printed in this book, because no text read prints one. The Bureau's guidelines, dated May 6, 2021, say how exchanges of information are read, and they note that under section 45 becoming a party to an agreement is sufficient, and that activities that help competitors monitor one another's prices can be enough to prove an agreement (Competition Bureau, 2021). The mechanics of all of this are Book H4's, and a verdict on a given exchange is a lawyer's question.

The second is the Criminal Code, R.S.C. 1985, c. C-46, s. 391, in Canada, current to September 21, 2026. Under s. 391(1) it is an offense to obtain, communicate or make available a trade secret knowingly by deceit, falsehood or other fraudulent means, and no one may do so, and under s. 391(2) the same holds for a person who knowingly does so knowing that the trade secret was obtained by such an offense. Section 391(4) provides that there is no offense where the trade secret was obtained by independent development or only by reverse engineering, and s. 391(5) defines a trade secret as information that is not generally known in the trade, that has economic value because it is not generally known, and that is subject to reasonable efforts to keep it secret. The civil remedies and the contract clauses around a trade secret are Book H5's.

The third is the Copyright Act, R.S.C. 1985, c. C-42, s. 29, in Canada, current to September 21, 2026, under which fair dealing for the purpose of research, private study, education, parody or satire does not infringe copyright. Whether a given copy is fair dealing, and what the license terms of a purchased report allow, are lawyer's questions and Book H5's.

The fourth is a code of self-regulation, and it is not a law. The ICC/ESOMAR international code is recognized by more than sixty associations in more than fifty countries, it applies to everyone involved in research including users of self-service platforms, and it binds its members and the associations that adopt it rather than a business by law (International Chamber of Commerce & ESOMAR, 2025). Its article 4 requires researchers to identify themselves promptly, to make participation voluntary and informed at the start, and to state retention; its article 7 requires that research be fit for purpose, that limitations be disclosed, and that findings be distinguished from interpretation and recommendations. A business that adopts the code's practice does so because it is good practice, not because a statute compels it.

The ethical line this chapter draws is stricter than the legal one, and it is drawn deliberately. No one poses as a customer, a candidate or a buyer in order to obtain a fact. No one asks a former employee of a competitor for a document, a price list, a customer list or a plan. No one re-identifies an owner whose name is redacted in the property assessment roll, because the roll's license forbids any attempt to do so. No one discusses a price, a bid, a territory or a customer with a competitor. Each of those is a refusal rather than a permission, and a refusal is easier to keep than a judgment call made in the moment.

QUÉBEC & CANADA NOTE Two Canadian statutes draw the line that a watch must not cross

The Competition Act, R.S.C. 1985, c. C-34, s. 45(1) and (2), in Canada, current to September 21, 2026, makes it an offense for a person, with a competitor in respect of a product, to conspire, agree or arrange to fix, maintain, increase or control the price of its supply, to allocate sales, territories, customers or markets, or to fix, maintain, control, prevent, lessen or eliminate its production or supply. The Criminal Code, R.S.C. 1985, c. C-46, s. 391, in Canada, current to September 21, 2026, makes it an offense to obtain, communicate or make available a trade secret by deceit, falsehood or other fraudulent means, and independent development and reverse engineering are exceptions of that section. No amount of a penalty is printed here, because no text read prints one. This book describes published rules on research, personal information, competition and copyright and is not legal advice. This book explains the law and recommends no vendor. Book H4 owns the agreements between competitors and the Bureau's guidelines, Book H5 owns trade secrets and fair dealing for research, and a question about a particular exchange, call or use of material belongs to a lawyer.

THE RULE No price, bid, territory or customer is discussed with a competitor, and nobody poses as someone else to obtain a fact

The rule is the line of the published texts called to Books H4 and H5. The Competition Act, s. 45(1) and (2), and the Bureau's guidelines dated May 6, 2021, treat the exchange of pricing information, costs, trading terms, strategic plans or marketing strategies between competitors as a matter that can raise concerns (Competition Bureau, 2021), and the Criminal Code, s. 391, treats a trade secret obtained by deceit, falsehood or other fraudulent means as an offense. What this chapter draws from those texts is a rule of conduct and not a restatement of the law: a watch reads only what competitors publish, and no one poses as a customer, a candidate or a buyer. The rule is neither the text of the law nor a verdict on any exchange, and no amount of a penalty is printed. Book H4 and Book H5 own the mechanics, and a lawyer answers the particular case.

7.8 Building a Watch System a Small Team Can Keep

A watch system is four things and nothing more: a short list of sources, a dated log, a rhythm and an owner. The list is written down in the order it is read, so that a round can be performed without deciding what to read. The log is dated, so that a signal can be compared with the same signal read later. The rhythm is fixed, so that the round happens on a day rather than when somebody remembers. The owner is named, so that a missed round is not nobody's business. A system missing any of the four is an intention.

A study of the competitor analysis systems of three large companies, read at abstract only, examines how members and users assess the formal system, what competitor analysis is used for, and how the function tries to improve its contribution and its role (Ghoshal & Westney, 1991). It concerns three large companies and gives no finding for a small business; it supports only the observation that such a function is assessed by those who use it, which is a reason to keep a watch small enough that its users are also its keepers.

Figure 7.1 draws the case's watch system in five steps, from left to right: public sources; the signal log, which holds eleven signals in ninety days, each dated; the monthly review of sixty minutes; the alert, which is a signal that meets a written threshold; and the decision, which is to answer, to watch on, or to do nothing, in writing. The watch takes thirty minutes a week. Below the five boxes, drawn dashed in red, are the three things the case refuses: exchanging prices or plans with a competitor, calling under a false pretext, and asking a former employee for a secret.

Figure 7.1 — A watch a small team can keep. Public sources feed a dated signal log of 11 signals in ninety days, a monthly review of 60 minutes, an alert when a signal meets a written threshold, and a written decision; the watch takes 30 minutes a week. The case refuses to exchange prices or plans with a competitor, to call under a false pretext or to ask a former employee for a secret.

The rhythm is the part a small team is most likely to lose, and the remedy is to make the rhythm smaller rather than to make the effort greater. The case's rhythm is a weekly round of thirty minutes and a monthly review of sixty minutes, about nine and a half hours a quarter, every figure posed by this book and illustrative rather than a norm. When a week is busy, the round is shortened rather than skipped, because a round that reads three sources instead of five still produces a dated log. What is never dropped is the monthly review, because it is where the log is read and where a signal either becomes an alert or is archived.

The log is kept as a task per round, with a verification criterion, so that the round has a definition of done: every source on the list was read and every change found was logged on the day it was found. That criterion is checkable in a minute and it prevents a failure of a watch, which is a round that was performed in the head and never written down.

The table below sets the columns of a signal log beside an example row of each kind of signal. The rows are posed by this book and illustrative, and the table's logic is that a reading is kept apart from a fact, in its own column, so that a reader who disagrees with the reading can still use the fact.

Table 7.4 — The signal log in six columns

WHO DOES THIS The administrator runs the round, the owner reads the log, and the outside lawyer reviews the line

In the case, posed by this book, the administrator holds the weekly round of thirty minutes and keeps the log, and Sophie Nadeau, the owner, joins the monthly review of sixty minutes, reads the log and decides. The outside lawyer reviews the script of any conversation with a customer, a supplier or a former employee and answers whether a step is lawful, and those questions belong to Book H4 and Book H5. In a solo business the owner performs the round and the review in one sitting, and in a larger business one person holds the log for the team and reports to whoever decides; both are this book's suggestions, not findings. No hour is promised to any reader, and the hours of the case are posed by this book and illustrative.

TOOLS OF THE TRADE What any watch tool must offer, whatever it is called

A watch is carried by functions rather than by a product, and the functions are these. The tool must hold a list of sources with the date each was last read. It must hold a log that dates each signal and keeps the reading apart from the fact. It must reach a person on the day, through a reminder or a recurring task, because a rhythm that depends on memory is not a rhythm. It must hold a place for the verification criterion, offer a view by competitor and by kind of signal, and export in a form another tool can read. And it must say what it does to collect, so that the team knows what it has not seen. No brand name belongs on this list and no price.

7.9 From Signal to Alert to Decision

A signal becomes an alert only when it meets a threshold written in advance. That is the whole of the mechanism, and its value lies in the words written in advance. A threshold chosen after the fact is not a threshold; it is a justification, and a business that decides what counts as important only once something has happened will find that everything counts. A threshold written before the watching starts is a decision the business made while it was calm, and it converts a log entry into a question that somebody has to answer.

An alert becomes a decision only when a person with the authority to decide reads it. A signal that reaches a log and stops there has produced nothing, and a watch that produces nothing will be abandoned within a quarter. The person who reads the log is the person who can say yes, no or not yet, and in the case that person is the owner.

There are three answers, and all three are written. The first is to answer: the signal meets the threshold, the business changes something, and the change is recorded with its date. The second is to watch on: the signal is close to the threshold but does not meet it, and the log keeps it with a note that says what would make it meet the threshold. The third is to do nothing: the signal does not bear on any decision the business has framed, and it is archived. The third answer is the most difficult, because a fact that was worth logging feels as though it should be worth acting on. A signal that fits nobody's decision is archived and not chased, and the discipline of archiving is what keeps the log readable.

The case's threshold is written into the decision rule of Chapter 2: the third territory is opened only if its adjusted count is at least four hundred buildings in the size band and the regional integrator won fewer than half of the public maintenance tenders seen there. The second test is a threshold of exactly the kind this section describes, written before the watching started. Four public tenders were seen in the third territory and two were won by the regional integrator, both figures posed by this book and illustrative; two of four is exactly half, and so not fewer than half, and Chapter 9 reads that result against the decision rule.

In La Machine, the alert is a reminder and the task holds the verification criterion (Productivity → Alerts productivity/alerts and Productivity → Tasks productivity/tasks), and nothing evaluates a signal: the judgment about whether a signal meets a threshold is a person's, and a watch without a threshold is a list.

Nothing is credited to the watch: no contract, no sale and no share is attributed to it, and the decision rule decides.

WHAT CHANGED The Bureau's guidelines on competitor collaboration were dated May 6, 2021, and the page was read again on October 10, 2026

The change recorded here is a change in the sources rather than in the case. The Competition Bureau's competitor collaboration guidelines are dated May 6, 2021, and the page was modified on June 27, 2024; they were read again on October 10, 2026, and they say that exchanges of current or future information between competitors are riskier than exchanges of historical information, and that information shared through a third party in aggregated form is generally less concerning than information shared directly (Competition Bureau, 2021). The guidelines are read at the level of the page, and they give no verdict on any particular exchange. The case's own change is counted in days and not in months: on day 90 of the watch the log holds eleven signals, posed by this book and illustrative. Dates of sources are never dates of the case, and Book H4 owns the mechanics of the guidelines.

7.10 The Case's Watch on Three Competitors

Nordec Entretien inc. watches three competitors through public sources only: the regional integrator, Competitor B and Competitor C. The regional integrator and its bid of about $560 a month are Book A3's, cited as that book's and never recomputed here, and Competitor B and Competitor C are posed by this book and never given a name.

The watch is held by the administrator, who performs a weekly round of thirty minutes, and by the owner, Sophie Nadeau, who joins a monthly review of sixty minutes; the two come to about nine and a half hours a quarter, posed by this book and illustrative rather than a norm. The round reads the sources in the same order every week: the tender system, the register of enterprises, the competitors' own pages, their job postings and the public reviews. The review reads the log and produces one of the three written answers of section 7.9.

The log holds eleven signals in ninety days, posed by this book: four public tenders and awards, three job postings, two changes to a public price or offer page, one new establishment in the enterprise register and one burst of public reviews. The split is what the case's log holds on the day it is read, and not a distribution to expect. Each signal carries its date, its competitor, its kind, its source and its reading, kept apart from the fact.

The tender notices of the third territory are the part of the log that feeds the decision rule: four public tenders were seen there and two were won by the regional integrator, both figures posed by this book and illustrative. The case verifies what the tender system shows on the day of use, because the page read for this book does not support contracts awarded, bidders' names or amounts (Gouvernement du Québec, 2026).

What the case refuses is as much a part of the watch as what it reads. Nordec Entretien inc. does not exchange a price, a bid, a territory or a plan with a competitor, does not call under a false pretext, does not ask a former employee of a competitor for a secret, a document or a list, and does not re-identify an owner whose name is redacted in the property assessment roll; it counts buildings rather than owners. Each refusal is a rule of the watch, and each is drawn dashed in red in Figure 7.1.

Nothing is credited to the watch: no contract, no sale and no share is attributed to it, and the decision rule decides.

The watch decides nothing on its own and does not predict what a competitor will do; it produces dated facts, a log, a rhythm and a set of alerts, and the decision belongs to the person who reads the log, with the tender counts feeding the second test that Chapter 9 applies.

DO THE MATH The watch of the case in five steps

Illustrative figures; the case is fictional. Step 1 — The eleven signals in ninety days are four public tenders and awards, three job postings, two changes to a public price or offer page, one new establishment in the enterprise register and one burst of public reviews, which add to eleven. Step 2 — Thirteen weekly rounds of thirty minutes in a quarter make six and a half hours. Step 3 — Three monthly reviews of sixty minutes in a quarter make three hours. Step 4 — Six and a half hours and three hours add to nine and a half hours a quarter. Step 5 — Of the four public tenders seen in the third territory, two were won by the regional integrator, which is exactly half and so not fewer than half. The figures are posed by this book, they are counts and hours rather than a norm, and they credit no contract and no sale to the watch. The sums are exact and no rounding is needed.

CASE IN POINT Nordec Entretien inc. watches three competitors in 30 minutes a week

Nordec Entretien inc. is a fictional business created for this encyclopedia; its owner, its employees, its customers, its competitors, its figures and the events described are invented, and any resemblance to a real business is coincidental. The watch is a weekly round of thirty minutes held by the administrator and a monthly review of sixty minutes held by the owner with the administrator, which comes to about nine and a half hours a quarter. The log holds eleven signals in ninety days: four public tenders and awards, three job postings, two changes to a public price or offer page, one new establishment in the enterprise register and one burst of public reviews. The sources read are the tender system, the register of enterprises, the competitors' own pages, their job postings and the public reviews. Four public tenders were seen in the third territory and two were won by the regional integrator. The case refuses to exchange a price or a plan with a competitor, to call under a false pretext and to ask a former employee for a secret. Every figure here is posed by this book and illustrative, and nothing is credited to the watch.

IN LA MACHINE La Machine keeps a competitor as a company record with dated notes and reminds a team of its weekly round, and it monitors no competitor

La Machine can host a simple questionnaire as a form whose answers become contacts, keep competitors as company records with dated notes, remind a team of its weekly watch through recurring alerts and tasks, and stage public business listings for prospecting; it draws no sample, computes no margin of error, offers no rating scale, records or transcribes no interview, listens to no social network, monitors no competitor and holds no market data. CRM → Companies business/marketing/companies holds one record per company, with notes, so a competitor profile is kept as dated notes on that record. Productivity → Alerts productivity/alerts and Productivity → Tasks productivity/tasks carry the weekly round and the monthly review as recurring reminders and dated tasks, and nothing in them collects a signal, reads a competitor's page or judges a threshold. The case's watch is run by people with a sheet and a calendar, and its log lives in the research file. Check it in your installation, because what a page shows can change. La Machine is described here and never sold, and this book is valid with any tool.

Costly Mistakes

• Treating competitive intelligence as spying. A watch that reads what competitors publish reads what somebody chose to make public; a watch that obtains something by deceit has crossed into an offense under the Criminal Code, s. 391, and the difference is one of kind rather than of degree.
• Filling a competitor profile with sentences nobody can check. A line that says a competitor is aggressive on price cannot be dated or sourced, so it cannot be compared with anything later, and it will be read a year on as though it were evidence.
• Logging a signal without its date or its source. A fact without a date cannot be compared and a fact without a source cannot be checked, and a fact that cannot be checked will be repeated in a meeting until it becomes a belief.
• Reading a burst of reviews as a measure of a competitor's customers. Reviews are written by those who chose to write, and the study read at abstract only reports two self-selection biases in the average online rating of products; a burst of reviews is a fact about how many reviews appeared and nothing more.
• Watching a list too long to keep. A list of twelve competitors produces a log nobody reads, and a log nobody reads is worse than no log, because it looks like evidence; the list is chosen by the decision and kept short.
• Writing the threshold after the signal arrives. A threshold chosen once something has happened is a justification rather than a threshold, and a business that decides what counts as important after the fact will find that everything counts.
• Discussing a price, a bid, a territory or a customer with a competitor. The Competition Act, s. 45(1) and (2), and the Bureau's guidelines dated May 6, 2021, treat such exchanges as a matter that can raise concerns, and Book H4 owns the mechanics; the watch reads what competitors publish and discusses nothing with them.

Key Takeaways

• Competitive intelligence is the ordered gathering and reading of public facts about competitors, customers and the environment so that a decision already framed can be taken with better evidence, and it is not spying, not the one-day strategic analysis of Book A3, not the business's own figures and not the voice of the customer.
• A competitor is whoever the customer would call instead, the list is chosen by the decision rather than by the market, and the two tests of relevance are whether the candidate meets the same customer and whether its move would change the decision.
• A competitor profile is a page of dated notes that holds identity, public offer, customers served, territory, announced moves and signals logged, with a source and a date on every line and no guess presented as a fact.
• A signal is a dated fact from a public source that bears on a decision, the reading is kept apart from the fact, and a signal becomes an alert only when it meets a threshold written in advance.
• The watch reads only what competitors publish, nobody poses as a customer, a candidate or a buyer, and no price, bid, territory or customer is discussed with a competitor; the Competition Act, s. 45(1) and (2), and the Criminal Code, s. 391, are called to Books H4 and H5 and no amount of a penalty is printed.
• The case's watch is a weekly round of thirty minutes and a monthly review of sixty minutes, holding eleven signals in ninety days and four public tenders seen in the third territory with two won by the regional integrator; every figure is posed by this book and illustrative, and nothing is credited to the watch.

Your Action Plan

1. Write the decision the watch will serve, and the threshold that will turn a signal into an alert. Name the decision in one sentence, name the competitor and the source the threshold concerns, and write the condition before any watching starts. Deliverable: one sentence of decision and one written threshold. Verification: the threshold names a competitor, a source and a condition, and it was written before the first round.
2. Choose the competitors by the two tests and keep the list short. Ask of each candidate whether it meets the same customer and whether its move would change the decision, and put the ones that fail both tests on a background list rather than in the watch. Deliverable: a watch list of a few competitors with the reason each is on it. Verification: every competitor on the list passes both tests, and the list fits on one page.
3. Build one profile per competitor as dated notes. Record identity, public offer, customers served, territory, announced moves and signals logged, with a source and a date on every line, and move anything that cannot carry both to a separate page of questions. Deliverable: one profile page per competitor. Verification: every line carries a source and a date, and the questions page is kept apart from the profile.
4. Set the rhythm and name the owner. Put the weekly round and the monthly review on a shared calendar with a duration and a named owner, and give the round a verification criterion that says what a completed round looks like. Deliverable: dated recurring entries and a written criterion. Verification: the criterion is checkable in a minute, and the owner of the round is named.
5. Read the log once a month and answer in writing. Read the log at the monthly review, apply the threshold, and record one of the three answers — answer, watch on, or do nothing — with its date. Deliverable: one dated page per monthly review. Verification: every flagged signal has a written answer, and every archived signal has a note saying which decision it failed to fit.

In La Machine

A competitor can be kept as a company record in the application copy read for this book, and the record carries less than a profile needs (La Machine, 2026). A profile is kept as dated notes on a company record (CRM → Companies business/marketing/companies), and the weekly round and the monthly review live as recurring reminders and dated tasks (Productivity → Alerts productivity/alerts and Productivity → Tasks productivity/tasks). Nordec Entretien inc. keeps the three competitors as company records with dated notes and holds the weekly watch as alerts and tasks. Check it in your installation, because what a page shows can change.

What the platform does not do belongs in the same breath, because a business that expects a watch to be built into the software will not build it. La Machine monitors no competitor, reads no competitor's reviews or pages, has no alert on public tenders and holds no market data, and nothing collects a signal by itself. A watch needs beyond the platform a list of public sources, a signal log that keeps a reading apart from a fact, a threshold written in advance, a person who reads the log and decides, and a lawyer for the line that the published rules draw.

Self-Check

1. What is competitive intelligence as this book uses the words, and how does it differ from the one-day strategic analysis of Book A3 (Strategy and Competitive Positioning)?
2. What are the two tests that decide whether a competitor belongs on the watch list, and why is the list kept short?
3. What does a competitor profile hold, and what must it never hold?
4. What is a signal, what is an alert, and why must the threshold be written in advance?
5. What does the case's watch read, what does it refuse, and what is credited to it?

Answers. 1. Competitive intelligence is the ordered gathering and reading of public facts about competitors, customers and the environment so that a decision already framed can be taken with better evidence (Calof & Wright, 2008, abstract only). Book A3's one-day analysis asks where the business stands in its industry, while the watch is continuous and small. 2. The two tests are whether the candidate meets the same customer and whether its move would change a decision the business has already framed. The list is kept short because a long list produces a log nobody reads. 3. A profile holds identity, public offer, customers served, territory, announced moves and signals logged, each with its source and its date, and it must never hold a guess presented as a fact. 4. A signal is a dated fact from a public source that bears on a decision; an alert is a signal that meets a threshold written in advance. The threshold must be written before the watching starts because a threshold chosen after the fact is a justification rather than a threshold. 5. The watch reads the tender system, the register of enterprises, the competitors' own pages, their job postings and the public reviews, and it holds eleven signals in ninety days, posed by this book. It refuses to exchange a price or a plan with a competitor, to call under a false pretext and to ask a former employee for a secret. Nothing is credited to the watch: no contract, no sale and no share is attributed to it, and the decision rule decides.

Chapter References

Calof, J. L., & Wright, S. (2008). Competitive intelligence. European Journal of Marketing, 42(7/8), 717–730. https://doi.org/10.1108/03090560810877114

Competition Act, R.S.C. 1985, c. C-34, s. 45(1) and (2) (Canada). Justice Laws Website, current to September 21, 2026 (last amended March 26, 2026) — read October 10, 2026. https://laws-lois.justice.gc.ca/eng/acts/C-34/

Competition Bureau. (2021). Competitor collaboration guidelines. Government of Canada. (Enforcement guidelines dated May 6, 2021; page modified June 27, 2024.) Retrieved October 10, 2026, from https://competition-bureau.canada.ca/en/how-we-foster-competition/education-and-outreach/competitor-collaboration-guidelines

Copyright Act, R.S.C. 1985, c. C-42, s. 29 (Canada). Justice Laws Website, current to September 21, 2026 (last amended November 7, 2024) — read October 10, 2026. https://laws-lois.justice.gc.ca/eng/acts/C-42/section-29.html

Criminal Code, R.S.C. 1985, c. C-46, s. 391 (Canada). Justice Laws Website, current to September 21, 2026 (last amended July 18, 2026) — read October 10, 2026. https://laws-lois.justice.gc.ca/eng/acts/C-46/section-391.html

Ghoshal, S., & Westney, D. E. (1991). Organizing competitor analysis systems. Strategic Management Journal, 12(1), 17–31. https://doi.org/10.1002/smj.4250120103

Gouvernement du Québec. (2025). Rechercher une entreprise au registre [Registraire des entreprises; in French]. (Updated September 23, 2025.) Retrieved October 10, 2026, from https://www.quebec.ca/entreprises-et-travailleurs-autonomes/obtenir-renseignements-entreprise/recherche-registre-entreprises/acceder-registre-entreprises

Gouvernement du Québec. (2026). Accéder au Système électronique d'appel d'offres (SEAO) [In French]. (Updated August 31, 2026.) Retrieved October 10, 2026, from https://www.quebec.ca/gouvernement/faire-affaire-gouvernement/outils/systeme-electronique-appel-offre

Hu, N., Pavlou, P. A., & Zhang, J. (2017). On self-selection biases in online product reviews. MIS Quarterly, 41(2), 449–471. https://doi.org/10.25300/MISQ/2017/41.2.06

International Chamber of Commerce & ESOMAR. (2025). ICC/ESOMAR international code on market, opinion and social research and data analytics. ESOMAR. Retrieved October 10, 2026, from https://community.esomar.org/uploads/public/knowledge-and-standards/codes-and-guidelines/ICCESOMAR-International-Code_English.pdf

La Machine. (2026). Documentation — route reference inventory. Internal product documentation, read October 10, 2026, together with the code of the pages this book describes: business/marketing/companies, productivity/alerts and productivity/tasks.

Netzer, O., Feldman, R., Goldenberg, J., & Fresko, M. (2012). Mine your own business: Market-structure surveillance through text mining. Marketing Science, 31(3), 521–543. https://doi.org/10.1287/mksc.1120.0713

Tirunillai, S., & Tellis, G. J. (2014). Mining marketing meaning from online chatter: Strategic brand analysis of big data using latent Dirichlet allocation. Journal of Marketing Research, 51(4), 463–479. https://doi.org/10.1509/jmr.12.0106

Chapter 8 — Market Sizing and Forecasting

A market size is a count with a range and never a single number, so this chapter says why a business sizes a market and which market it sizes, revisits the total, serviceable and reachable markets for a business that already has customers, counts a market from the top down with published counts and from the bottom up with units, rates and prices, triangulates the two into the range that survives, forecasts demand from judgment, trends and stated intentions, says why stated intentions overstate, reads a diffusion curve for what it is, and presents a size or a forecast with its uncertainty, then applies all of it to the reachable market of Nordec Entretien inc. counted by two paths and to its forecast of signings.

What You'll Be Able to Do

• Explain why a business sizes a market and which market a decision actually reaches, and say why a size that no decision consumes is a curiosity.
• Distinguish the total, the serviceable and the reachable market, and say why a business that already has customers sizes the reachable one first.
• Build a top-down count from a published count, narrow it by the filters of the decision, and check a sample of records by hand against the same criteria every time.
• Build a bottom-up count from what can be counted on the ground, multiply it, and disclose how the zones were chosen.
• Triangulate two counts that do not share an error into the range that survives, and print that range with its assumptions and with a sentence on what is not counted.
• Forecast demand as a range with a reason, keep a posed fraction of stated intentions, and present a size or a forecast with its unit, its date, its method and its uncertainty.

Why It Matters

A business sizes a market because a decision needs a denominator, and the denominator is almost never the number the business first reaches for. The question that opens a count is seldom "how big is this market"; it is something narrower: how many buildings could a third territory hold, how many proposals does a year of letters have to produce, how much can a budget of a given size be expected to return. A count built to answer one of those questions is small, cheap and useful. A count built to answer the general question is large, expensive and hard to finish, because nobody can say when it is finished. The discipline of this chapter is to let the decision set the unit, the place and the period, and to refuse every figure that does not serve them.

The second reason the count matters is that a business with customers already has a market and has never counted it. Nordec Entretien inc., the fictional business followed through this book, has 72 buildings under contract and a market of about 1,400 buildings that Book B4 (The Marketing Plan, Budget and KPIs) printed in general terms, and nobody at Nordec could say how many of those buildings were managed by managers the business already knew. A business in that position needs two counts made by two paths that fail differently, and a range that survives the comparison; that is what this chapter builds, in days rather than in months.

The third reason is the forecast, and the forecast is where a business can most easily fool itself. A stated intention to change provider is not a sale, and a business that counts intentions as sales will plan a year it cannot deliver. The honest forecast is a range with a reason attached: a trend extended from the business's own history, a funnel from proposals to signings, and a posed fraction of the intentions that were stated. None of the three is accurate on its own, and no source read for this book gives an accuracy rate for any of them. What the three give together is a range wide enough to be true and narrow enough to plan against, and a range that wide is exactly what a budget needs.

In This Chapter

Section 8.1 explains why a business sizes a market and which market a decision reaches, and sets this chapter beside the books that own the neighboring questions. Section 8.2 revisits the total, the serviceable and the reachable market for a business that already has customers. Section 8.3 builds a top-down count from published counts and checks it on a sample of records. Section 8.4 builds a bottom-up count from units, rates and prices. Section 8.5 triangulates the two into the range that survives. Section 8.6 forecasts demand from judgment, trends and intentions. Section 8.7 examines stated intentions and why they overstate. Section 8.8 reads a diffusion curve for what it is. Section 8.9 presents a size or a forecast with its uncertainty. Section 8.10 applies all of it to the market count and the forecast of Nordec Entretien inc.

8.1 Why Size a Market, and Which Market

A market size is wanted for a decision and never for itself. Nobody needs to know how many commercial buildings stand within forty kilometers of a base unless that number is going to change what the business does next: where to put the next letter wave, how large a budget a territory can carry, whether a third territory is worth opening at all. A size that no decision consumes is a curiosity, and a business that counts a market without a decision waiting for the count will find that the count changes nothing and is never repeated. The decision comes first, and the size is built to fit it.

Three questions come before any count, and a business that answers them in writing saves itself most of the work that follows. The first is which buyers: the count is of the buyers the offer fits, not of everyone who might conceivably buy something. The second is which place: a count is always a count of a place, whether that place is a set of municipalities, a set of postal codes, a set of routes or a set of zones walked on foot. The third is which period: a count is a count at a date, and a count made two years ago describes a market that has since changed. A count that does not name its buyers, its place and its period is not a count; it is a number.

The third question is easily skipped, and it decides whether the count is worth making. A business that counts its market once and never again has a photograph. A business that counts it on a fixed date, with the same unit and the same filters, has a series, and a series is what shows whether a market is growing, holding or shrinking. The case of this book counts its reachable market once, on a day of its study, and the second count is posed on day 365 of this book, which is where a series begins. The date of the count is part of the count, and a count printed without its date is a count that cannot be compared with anything.

A larger market feels like a better market, and a business that has just counted 1,331 buildings where it expected 1,400 may feel that something has gone wrong. Nothing has: a count is an estimate with a range, and what a size feeds is a decision rule written in advance, which is moved only by the condition it names. Chapter 2 owns the question and its decision rule, and this chapter never reopens them.

The same discipline applies to the place. A count across a region the business does not serve is a count of somebody else's market, large, impressive and useless; the place of the count is the place the business can reach with its routes, its vehicles, its technicians and its hours.

The table below sets six neighbors beside what each of them holds and what this chapter holds, so that a reader can see at a glance which question belongs here and which belongs to a book that already owns it. Read it as a boundary rather than a hierarchy: every row names a book, and the third column says what this chapter does with the same territory.

Table 8.1 — Where this chapter stops and the books around it start

This book explains published sources and a method of decision; it is not legal, tax or accounting advice, and it recommends no vendor, no survey tool, no panel, no research firm, no data provider and no platform. Where a rule of law touches the work of this chapter, the section that states it sends the reader to the book that owns the mechanics.

DEFINITION Total, serviceable and reachable market, count, range and forecast, as this book uses the words

These are this book's working words, built on the definitions of Book A1 (chapter 5 of Book A1), which are cited here and never rewritten. The total market is everyone who could buy the kind of thing the business sells, whether or not the offer fits and whether or not the business could ever reach them. The serviceable market is the part of the total market that the offer actually fits, in its size, its price and its terms. The reachable market is the part of the serviceable market that lies within the place the business serves and the means it has to serve it. A count is a number with its unit and its date, and a count without a unit is not a count. A range is the interval that two paths leave standing when they are set side by side. A forecast is a range with a reason attached to it. No source read for this book defines these words in this sense; they are the working vocabulary of this chapter.

8.2 Total, Serviceable and Reachable Markets Revisited

The three nested counts are a long-standing idea in market sizing, and an idea that is easy to recite without using. The total market is everyone who could buy the kind of thing the business sells. The serviceable market is the part of that total that the offer actually fits, in its size, its price and its terms. The reachable market is the part of the serviceable market that lies within the place the business serves and the means it has to serve it. Book A1 (From Idea to Business Model) sets out the three for a venture that has no customers yet, in chapter 5 of Book A1, and that definition is cited here and never rewritten. What this chapter adds is the order in which a business that already has customers should count them.

A business with customers sizes the reachable market first, and the reason is arithmetic rather than preference. The reachable market is the only one of the three that a decision can act on this year: it is the set of buyers the business could actually serve with the routes, the vehicles, the technicians and the hours it has. The serviceable market is larger and the total market is larger still, and both are useful for a strategy that looks several years out, which is Book L1's territory rather than this chapter's. A business that begins with the total market will produce a number in the tens of thousands, discover that it cannot act on any of it, and abandon the exercise. A business that begins with the reachable market will produce a number in the low thousands, discover that it can act on all of it, and count again next year.

The reachable market of the case is the one Book B4 printed: about 1,400 commercial buildings of 2,000 to 9,000 square meters within forty kilometers of the base, in chapter 2 of Book B4. That figure is Book B4's, it is printed in general terms, and this chapter never corrects it. What this chapter does is recount the same reachable market by two independent paths and print the range that survives, which is the work of sections 8.3 to 8.5 and of section 8.10. Book B4 also recorded that nobody knew how many of the 1,400 reachable buildings were managed by managers the business already knew, and that open question is not closed by a count; it is closed by the survey of Chapter 5 and by the watch of Chapter 7.

Every count names its unit, and the unit is the part of a count that is easily lost. A count of buildings is not a count of firms, because one firm can own or occupy several buildings. A count of firms is not a count of locations, because one firm can operate from several locations. A count of locations is not a count of decision-makers, because one location can be managed by a property management firm that decides for a dozen buildings at once. A count of decision-makers is not a count of buyers, because a decision-maker may be bound by a bundled facility contract that leaves no room for a new provider. Each of those units is legitimate, and each answers a different question. What is never legitimate is to count in one unit and read the result in another.

The unit also decides what the count can be compared with. A count of buildings can be compared with a property roll, which lists buildings. A count of businesses can be compared with a business register, which lists enterprises. A count of locations can be compared with the business counts of a statistics agency, which counts locations. A business that keeps its unit straight can put two sources side by side and see whether they agree. A business that mixes its units will find that its sources disagree, and will conclude, wrongly, that one of the sources is wrong.

There is one more distinction worth keeping: a count says how many units there are, and a rate is a count divided by a count. The case's penetration is a rate: 72 buildings under contract divided by a reachable market, which gives 5.1 percent on Book B4's figure of about 1,400 and 5.4 percent on the working count of 1,331, each rounded to the nearest tenth of a percent and each posed by this book. A rate can be compared across years and moves when either of its numbers moves, so a business that reports a penetration rate also reports the count underneath it.

8.3 Top-Down Estimates From Published Counts

A top-down estimate begins with a count that somebody else has already made and narrows it, step by step, until what remains is the market the decision reaches. The appeal of the method is that the first number is free and already published. The danger of the method is that every narrowing step is a filter, and a filter that is wrong by a little produces a count that is wrong by a lot. The discipline is to name each filter, to say where it comes from, and to check the result on a sample of records read by hand.

The published counts of a statistics agency are a usual starting point, and they come with a unit that has to be read carefully. In December 2025 there were 1.37 million employer businesses in Canada and 3.67 million non-employer businesses with annual revenues greater than $30,000, as the release of February 13, 2026 states (Statistics Canada, 2026), and the release says that businesses are usually counted by location, each operating location separately. The count is therefore of locations, not of distinct firms and not of buildings; the box below gives the release's own caveats.

For a market of buildings, the published count that fits is a property roll rather than a business count. The property assessment roll of Québec lists every property of a municipality, assessed on the same basis and at the same date, and it is the central element of the municipal tax system; the data set holds one file per municipal roll, is updated yearly, and its cadastral numbers, owners' names and postal addresses are redacted, with re-identification of the redacted personal information prohibited, under an Attribution license (Ministère des Affaires municipales et de l'Habitation, 2026). The catalog record was read on October 10, 2026, and the files themselves were not opened. Two consequences follow for the case. The first is that the roll is a count of buildings and never a count of owners, and the case counts buildings. The second is that the roll's owners are redacted and re-identifying them is prohibited, which is a question of personal information and belongs to Book H2 (Privacy and Personal Information).

A top-down count overstates when a filter is wrong, and the only way to know is to read records by hand. The case takes the roll's 1,512 buildings in the size band in the municipalities of the two territories, draws 50 records and reads each against the same criteria: is the building inside the size band, is it commercial, is it within the place the business serves. Of the 50 records, 44 are eligible, which is 88 percent, and the adjusted count is 1,512 times 88 percent, which is 1,330.56, that is 1,331. Every figure is posed by this book, illustrative and never a norm.

The table below sets five steps of a top-down count beside what each step can overstate or miss, the check that catches it and the result in the case. Read it as a sequence: each row narrows the count a little further, and the last row is the figure the business prints. The results in the final column are posed by this book.

Table 8.2 — A top-down count and its three checks

QUÉBEC & CANADA NOTE Statistics Canada counts business locations, and a count of locations is not a count of firms

In December 2025, Statistics Canada counted 1.37 million employer businesses in Canada and 3.67 million non-employer businesses with annual revenues greater than $30,000 (Statistics Canada, 2026). The release of February 13, 2026, read on October 10, 2026, says the counts are of businesses active on the Business Register, compiled mainly from Canada Revenue Agency records, and that businesses are usually counted by location, each operating location separately. Employer status depends on payroll remittances in the previous 12 months, and the agency asks for cautious interpretation of changes between periods. The consequence for a business that wants a market count is plain: the count is of locations and not of distinct firms, and it is not a count of buildings either. This book explains published sources and recommends no vendor; the agency's own release is the place to read the count, with its date and its caveats.

IN LA MACHINE The Business Explorer stages listings for prospecting, and no page counts a market

In the application copy read for this book, Reseller → Business Explorer business/marketing/business-explorer, reserved to the administrator and reseller roles, searches public business listings, stages them in a transit table with their industry classification code and their source, and exports them or converts them into companies with the prospect status. It is never a market database and never a count of a market, and Book B1 (Marketing Fundamentals for Business Owners), chapter 3, sends the reader here. La Machine holds no market data, makes no market count and makes no forecast, and no page of it draws a sample or computes a margin of error. Check it in your installation, because what a page offers can differ from one installation to another. La Machine is described here and never sold, and it is one worked example of a business platform rather than the tool of the case's study. The case counts its market in a spreadsheet and does not use La Machine for the count.

8.4 Bottom-Up Estimates From Units, Rates and Prices

A bottom-up estimate begins with something that can be counted on the ground and multiplies it. The appeal of the method is that every factor is visible: a business counts the units it can see, states the rate at which those units stand for the units it cannot see, and multiplies. The danger of the method is the rate, because the rate is an assumption and an assumption is invisible unless it is written down. A bottom-up count is honest when its assumption is printed beside its result, and dishonest when the result is printed alone.

The case walks 6 of 30 commercial zones and counts 262 buildings in them. The count of 262 is a count of what was seen, and the extrapolation to the whole territory is a rate: 262 divided by 6 gives the average per zone, and that average times 30 gives 1,310, exactly. Every one of those figures is posed by this book. The arithmetic is simple and the assumption underneath it is not: the six zones walked are taken to resemble the twenty-four zones not walked. If the six zones were chosen because they are the ones the business already serves, or because they are the easiest to reach on foot, the assumption fails and the count is wrong in a direction nobody can see.

The remedy is not to walk all thirty zones, which would cost more than the count is worth. It is to choose the zones at random or, where the choice cannot be random, to disclose how they were chosen and what the choice is likely to have done to the count: six zones nearest the base give a count that is probably high, and six zones farthest from it a count that is probably low.

A bottom-up count can also be built from units, a rate and a price, and the three factors have different rules. The units are counted, and the count is what was seen. The rate is either taken from a source, in which case the source is named with its year, or it is posed, in which case the word posed is printed beside it. The price is either taken from the business's own records, in which case it is the business's own average, or it is posed in the same way. A count that multiplies three factors without saying which of the three is measured and which is assumed has produced a number that cannot be checked, and a number that cannot be checked will not be used twice.

The case's average contract is a factor that comes from the business's records rather than from a source: Book B4 printed an average contract of about $8,541 a year, in chapter 2 of Book B4, and that figure is never recomputed here. A count multiplied by a share that might buy and then by an average contract is a revenue estimate rather than a market count, and it belongs to the budget.

The bottom-up count also has a role in the case that is not arithmetic. The dispatcher checks the zones walked against the routes the business actually runs, which is 3 hours of the case's time, posed by this book. The check answers a question the arithmetic cannot: whether the zones that were walked are zones the business could serve, and whether the zones that were not walked are zones it would want to. A count of buildings in a zone the business cannot reach is a count of somebody else's market, and the dispatcher is the person who knows which zones those are.

WATCH OUT A count by location, a relative index and a stated intention are three numbers mistaken for a market

Three numbers are easily mistaken for a market, and each has a question that catches it. The first is a count by operating location read as a count of firms or of buildings: the question is what unit the count uses, and the answer is that a business with three locations is one firm and three locations. The second is a relative search-interest index read as a number of searchers: the question is whether the value is relative to all searches in a place and a time, and the answer is that it is, so the value runs from 0 to 100 and is never a volume. The third is a stated intention read as a sale: the question is what share of such statements came true, and who says so, and the honest answer is that no source read for this book gives a share that applies to the case. This book prints no frequency for any of the three traps; it prints the question that catches each one.

8.5 Triangulation and the Range That Survives It

Triangulation is the comparison of two counts that do not share an error. The point of the comparison is not to average the two, because the average of two paths that fail differently is a point that neither path supports. The point is to see where the two agree, where they differ, and what the difference says about the weak point of each. Two counts that agree closely are two counts that have probably made the same mistake, and two counts that differ widely are two counts that have probably each made a different one. The useful output is a range, and the range is what the business prints.

The case's two counts are 1,331 from the top-down path and 1,310 from the bottom-up path, and the range that survives is 1,310 to 1,331. Book B4's figure of about 1,400 sits above the range, and the raw roll of 1,512 sits above that. The arrangement is the ordinary shape of a triangulation: the raw published count is the largest, because no filter has been applied; the adjusted top-down count is smaller, because the sample check removed the ineligible records; and Book B4's figure, printed in general terms before any count was made, sits between the range and the raw roll. Every figure is posed by this book.

A difference between two paths is information and not a flaw to hide. When the two counts differ, the difference points at one of three things. The first is a filter that is wrong: the size band read from the roll may not match the band the offer serves, or the place drawn on a map may not match the place the business actually serves. The second is a zone that is unrepresentative: the six zones walked may be denser or sparser than the twenty-four that were not. The third is a unit that has been confused: the roll may count a building that the walk counted as two, or the walk may have counted a building that the roll lists as part of a larger property. Naming which of the three is at work is the whole value of the comparison, and it takes a paragraph rather than a study.

The range is printed and no point inside it is called the market. A business that prints a single number invites a reader to treat it as exact, and a reader who does will be disappointed by the first month that does not match it. A business that prints a range can say that the market is somewhere between the two counts, made by two paths that fail differently, and that it uses the lower figure for planning and the higher for ambition.

Figure 8.1 lays out the case's triangulation of its reachable market, a top-down count and a bottom-up count, and places Book B4's figure on a scale beside the range that survives.

Figure 8.1 — Two counts and the range that survives. The property roll gives 1,512 buildings and a check of 50 records leaves 1,331; walking 6 of 30 commercial zones counts 262 buildings and extrapolates to 1,310. The range that survives runs from 1,310 to 1,331, and Book B4's figure of 1,400 sits above it and below the raw roll.

The case's working count becomes the range of 1,310 to 1,331, and Book B4's figure of about 1,400 is not corrected. Book B4 counted in general terms, before any roll was read and before any zone was walked, and a general figure printed in a plan is not the same object as a count made by two paths. The working count is this book's, it is posed by this book, and it is used for the penetration rate and for the decision of Chapter 9. A business that keeps both figures and says which is which has done the honest thing; a business that quietly replaces one with the other has lost the record of how it changed its mind.

KEY CONCEPT A market is counted by two paths that do not share an error, and what survives is a range

The central idea of this chapter is that a market is counted twice, by two paths that fail in different ways, and that what survives the comparison is a range rather than a point. A top-down count can overstate when its filter is wrong: a size band read from a published field may not match the band the offer serves, and a place drawn on a map may not match the place the business actually serves. A bottom-up count can miss when the zones walked differ from the zones not walked, and the difference is invisible unless it is disclosed. Two paths that fail differently show the weak point when they differ, and a difference is information rather than a flaw to hide. The range is printed, and no point inside it is called the market. The counts of the case are posed by this book, illustrative and never a norm.

METHOD Triangulating a market count in five steps

First, name the unit and the place of the count, and write them down before any figure is taken. Second, build the top-down count from a published count, narrow it by the filters of the decision, and check a sample of records by hand against the same criteria every time. Third, build the bottom-up count from what is counted on the ground, multiply it, and disclose how the zones were chosen. Fourth, set the two counts side by side with the figure the business used before, and write down why they differ. Fifth, print the range that survives, with the assumptions behind each path and a sentence on what is not counted. This is the book's reasoning and not a tested procedure, and the numbers of the case are posed by this book. The method is deliberately short, because a method that takes a week will not be repeated, and a count that is not repeated is a photograph rather than a series.

8.6 Forecasting Demand: Judgment, Trends and Intentions

A forecast is a range with a reason attached, and the reason is what separates a forecast from a guess. A guess is a number somebody believes. A forecast is a number somebody can defend, because the path from what is known to what is expected is written down and can be checked. Three paths are honest enough to use in a small business, and they are best used together, because each one fails in a different direction.

The first path is judgment, written down with its reasons. A judgment forecast is the owner's own reading of the year ahead, and it is legitimate as long as it is written in a form that can be contradicted: the number, the reasons behind it, and the conditions under which it would be wrong. A judgment forecast that is written down can be compared with what happens, and a judgment forecast that stays in the owner's head cannot. The value of writing it down is not that the number improves; it is that the reasoning becomes visible, and reasoning that is visible can be corrected.

The second path is a trend extended from the business's own history. A business that has signed a certain number of contracts in each of the last several periods has a series, and a series can be extended. The extension is only as good as the stability of the series, and a business whose signings have moved by a factor of two from one period to the next has a series that does not support an extension. The trend is the simplest of the three paths and the one easiest to believe without examining it, because a straight line drawn through past numbers looks like evidence.

The third path is a funnel from proposals to signings. A business that knows how many proposals it issues and what share of them become contracts can forecast the contracts that a planned number of proposals will produce. The case's funnel is 38 proposals at a close rate of 54.3 percent, which gives 20.6 signings, posed by this book. The funnel is the most mechanical of the three paths and the one most sensitive to its inputs: a close rate taken from a year in which the business was unusually lucky will produce a forecast that is too high, and a close rate taken from a year in which a large customer left will produce one that is too low.

Search data belong to a fourth kind of path, and the honest way to describe them is narrow. Search engine data can help forecast near-term values of economic indicators, illustrated with automobile sales, unemployment claims, travel destination planning and consumer confidence (Choi & Varian, 2012, at abstract only). The abstract reports that finding and gives no accuracy figure, and it concerns economic indicators rather than a small business market. A business that reads a relative search-interest index as a demand for its own service has moved from a forecast of an economy to a forecast of a customer, and the move is not supported by anything read for this book. Chapter 6 owns the digital signals and their limits.

The three paths are combined by printing all three and the range they span. The case's trend gives 19 signings, which is the number of Book B4's year; the funnel gives 20.6; and the intentions, which are the subject of section 8.7, give a third figure. The range that the three produce is 19 to 23 signings, with a central value of 21, and every one of those figures is posed by this book. The range is wide, and a wide range is the honest output of three paths that disagree. A business that narrows the range by choosing the path it likes best has produced a number rather than a forecast.

THE RULE A stated intention is never counted as a sale, and a relative index is never counted as a demand

The rule of this chapter is the book's own, drawn from the abstracts read and from no standard and no legal duty. Studies that related stated intention to actual purchase often found substantial variation between them, and research suppliers estimate that 70 percent to 90 percent of their clients regularly use purchase intention scales (Jamieson & Bass, 1989, at abstract only). Measuring intentions can itself inflate their association with behavior, and across three large field studies the intention-behavior correlation was on average 58 percent greater among surveyed than among non-surveyed consumers (Chandon et al., 2005, at abstract only). Neither abstract supports a discount factor, so a forecast keeps a posed fraction of the intentions, says that the fraction is posed, and prints the range. The case keeps a quarter of its 151 stated switchers, which is 38, and the quarter is posed by this book.

8.7 Stated Intentions and Why They Overstate

A stated intention is what a person says about a future choice, and it is not the choice. The gap between the two is the subject of a body of research, read here at abstract level. Studies that related stated intention to actual purchase often found substantial variation between them, and research suppliers estimate that 70 percent to 90 percent of their clients regularly use purchase intention scales (Jamieson & Bass, 1989, at abstract only). The abstract reports the estimate and the variation; it does not report an adjustment factor, and no adjustment factor is printed here.

There is a second effect, and it is the one that catches a business that surveys its own market. Measuring intentions can itself inflate their association with behavior, and across three large field studies the intention-behavior correlation was on average 58 percent greater among surveyed than among non-surveyed consumers (Chandon et al., 2005, at abstract only). The abstract reports the finding and its setting, which is three large field studies of consumers; it does not report a discount factor, and the case's discount is posed by this book. The practical consequence is that a survey which asks people whether they intend to change provider has changed the thing it measured, at least a little, and the change runs in the direction of more stated intention rather than less.

Neither abstract supports a number that a business could apply to its own intentions, and this is the point at which a chapter has to be honest about what it does not know. The case keeps a quarter of its 151 stated switchers, which is 38, and the quarter is posed by this book. The quarter is not derived from either study, it is not a benchmark, and it is not a norm; it is a decision the business made, written down so that it can be compared with what happens. A business that keeps a different fraction is not wrong, as long as the fraction is written down before the year begins and the result is compared with it afterward.

The survey of Chapter 5 gives the case its count of intentions, and the count is the survey's rather than a forecast. Of the 92 completes, 11 said they planned to change provider within twelve months, which is 12 percent, plus or minus 6.3 points at 95 percent confidence, and that figure is the survey's own, computed from its probability sample with its formula, its confidence level and its population. It is a count of what people said, and it is not a number of contracts. Scaled to the frame of 1,259 buildings, those 11 of 92 give the 151 stated switchers of the forecast, a scaling that carries the survey's margin with it and is posed by this book. A stated intention to change provider is a signal to follow up, and following up is a conversation, not a forecast.

The right use of a stated intention is therefore as a list rather than as a number. A business that has a list of names of people who said they might change provider has as many reasons to make a call, and the calls will produce whatever they produce. A business that has only a number and multiplies it by a fraction has a forecast that depends entirely on the fraction, and the fraction is the part nobody can source. The case does both: it keeps the list for the sales work and it keeps the posed fraction for the forecast, and it says which is which.

8.8 New-Product Adoption and Diffusion Curves

The diffusion model of new-product adoption separates innovators from imitators and draws an S-shaped curve of cumulative adoption (Bass, 1969, reused from the book that prints it and not re-read for this book). That is the whole of what is printed here about the model, because its parameters are not available at the level this book read it. Book L1 (Growth Strategy and Scaling) uses the model for growth and scaling, and a reader who wants to apply it should go there.

The idea behind the curve is worth taking, and it can be taken without a parameter. Some buyers move first, before anybody they know has tried the thing, and other buyers move later, after they have seen somebody they know try it. A business that watches its own signings can often see the two groups: the first buildings that signed did so on the strength of a conversation or a referral, and the buildings that signed later did so after seeing a neighbor's building serviced. That observation is a reason to keep a record of how each signing arrived, which is a task of Chapter 7 and of the business's own analytics in Book B7 (Marketing Analytics, Attribution and Dashboards).

The curve fits a maintenance service badly, and the reason is worth stating as the book's reasoning rather than as a finding. A diffusion curve describes a product that spreads among buyers who did not have it before. A maintenance contract is not a new product to a building owner; it is a replacement for a provider the owner already has, or a first contract for an owner who has been managing the building alone. The first kind of buyer is choosing between providers, and the second is choosing whether to buy at all, and the two choices follow different logics. A curve that describes the spread of a novelty does not describe either of them well.

The case does not use a diffusion curve, and it says so: the forecast of section 8.6 rests on a trend, a funnel and a posed fraction of stated intentions. What the case takes from the idea is the practice of recording how each signing arrived, so that first movers and followers can be told apart in the business's own history.

A curve fitted to a history describes that history, and a curve extended forward is a forecast like any other, with a range and a reason; presented as a prediction it is a forecast without its uncertainty, which is the error that section 8.9 exists to prevent.

8.9 Presenting a Size or a Forecast With Its Uncertainty

A size is presented with its unit, its date, its method and its range, and a forecast is presented with its reason and its range. Those elements are not decoration; they are what makes the number usable by somebody who did not make it. A count printed without its unit cannot be compared with anything. A count printed without its date cannot be compared with a later count. A count printed without its method cannot be checked. A count printed without its range invites a reader to treat it as exact, and a reader who treats it as exact will be disappointed.

Uncertainty has two kinds, and they behave differently. Sampling error is the error that comes from measuring a sample rather than the whole population, and it decreases as the sample grows; it exists only in a sample survey, and it is the error that a margin of error describes. Nonsampling errors are everything else: coverage, measurement, nonresponse and processing, and they exist in samples and in censuses alike. The distinction matters because the two behave in opposite ways under pressure. A larger sample reduces sampling error and does nothing to nonsampling error, and a systematic error leads to a bias that a larger sample does not reduce (Statistics Canada, 2010). The manual is a statistical agency's methodology manual, archived, and it binds no private business; it is read here for what it establishes about error.

The practical consequence for this chapter is a rule about margins. A margin of error belongs to a probability sample and to nothing else. The case's survey has a probability sample, and its margin is printed with its formula, its confidence level and its population, which is Chapter 5's work. The count made by walking six zones has no margin of error, because the zones were not drawn as a probability sample and no measure of precision applies to them. A set of reviews has no margin of error, because the people who wrote the reviews chose to write them. An opt-in panel has no margin of error, because the people who joined the panel chose to join it. A business that prints a margin beside any of those has printed a number that means nothing.

The table below sets six elements of a presentation beside what to print, what never to print and an example from the case. Read it as a checklist for the paragraph in which a size or a forecast is delivered: each row is one element, and the third column is the sentence that must not appear. The examples in the final column are posed by this book, and the table is the book's reasoning rather than a finding.

Table 8.3 — How to present a size and a forecast

TOOLS OF THE TRADE What any counting sheet must offer, whatever it is called

A counting sheet is a spreadsheet or a table, and what matters is the functions it offers rather than the name on it. It needs a column for the unit of every count, so that a count of buildings is never read as a count of firms, and a column for the source of every figure. It needs a check sample kept apart from the full list, a place for the filters and the date of each, and a cell that shows the range of two counts rather than only their average. It needs a history of the counts by date, so that the second count can be compared with the first, and an export in a form another tool can read. No brand name and no price belongs on this list.

8.10 The Case's Market Count and Forecast

Nordec Entretien inc. counts its reachable market by two paths and forecasts its signings, and every figure in this section is posed by this book, illustrative and never a norm. The top-down path begins with the property assessment roll, which gives 1,512 buildings in the size band in the municipalities of the two territories; the administrator checks 50 records against the same criteria and finds 44 eligible, which is 88 percent; the adjusted count is 1,512 times 88 percent, which is 1,330.56, rounded to the nearest building, that is 1,331. The roll is read on its public site, its owners are redacted and re-identifying them is prohibited, and the case counts buildings and never owners (Ministère des Affaires municipales et de l'Habitation, 2026). The rules on personal information behind that redaction belong to Book H2 (Privacy and Personal Information).

The bottom-up path begins on the ground. Six of the thirty commercial zones are walked and 262 buildings are counted in them, so each zone averages 262 divided by 6, and the thirty zones give exactly 1,310. The case discloses how the zones were chosen, and the dispatcher checks the zones walked against the routes the business actually runs, which is 3 hours of the case's time, posed by this book.

The range that survives is 1,310 to 1,331, and Book B4's figure of about 1,400, printed in general terms in chapter 2 of Book B4, sits above it and below the raw roll of 1,512; it is never corrected here, and the working count is posed by this book. The penetration of the business is 72 buildings under contract divided by the working count, which gives 5.4 percent on 1,331 and 5.1 percent on Book B4's 1,400, each rounded to the nearest tenth of a percent. The third territory is counted at 470 buildings raw and 414 adjusted, and that adjusted figure is the one the decision rule of Chapter 2 tests against its threshold of 400.

The forecast is built from three paths and printed as a range. The trend gives 19 signings, which is the number of Book B4's year. The funnel gives 38 proposals at a close rate of 54.3 percent, which is 20.6 signings. The stated switchers number 151, which is the survey's 11 of 92 scaled to the frame of 1,259 buildings, and the case keeps a quarter of them, which is 38, a quarter posed by this book. The three paths span 19 to 23 signings, with a central value of 21, and the range is what the budget takes as its input. The forecast feeds the budget and the decision rule decides; nothing is credited to the study, and no contract, sale or share is attributed to the count or to the watch.

Sophie Nadeau, the owner, reads the range rather than a point: the market is between 1,310 and 1,331 buildings, the business holds 5.4 percent of the working count, the third territory clears the threshold of the decision rule at 414 adjusted buildings, and the year ahead is expected to produce between 19 and 23 signings. Chapter 9 turns the evidence into the decision brief, where the count, the forecast and the watch are set beside the rule that was written before any of them was made.

DO THE MATH The market count and the forecast of the case, in five steps

Illustrative figures; the case is fictional. Step 1 — The property roll gives 1,512 buildings in the size band, and 44 of the 50 records checked are eligible, which is 88 percent, so the top-down count is 1,512 times 88 percent, which is 1,330.56, rounded to the nearest building, that is 1,331. Step 2 — The 6 zones walked count 262 buildings, so each zone averages 262 divided by 6, and the 30 zones give exactly 1,310. Step 3 — The range that survives is 1,310 to 1,331, and Book B4's figure of about 1,400 sits above it. Step 4 — The 72 buildings under contract are 5.4 percent of 1,331 and 5.1 percent of 1,400, each rounded to the nearest tenth of a percent. Step 5 — 38 proposals at a close rate of 54.3 percent give 20.6 signings, and a quarter of the 151 stated switchers, rounded up, is 38, so the forecast is a range of 19 to 23 around a central value of 21. The figures are posed by this book, are counts and not a norm, and credit no contract or sale to the study.

CASE IN POINT Nordec Entretien inc. counts its reachable market by two paths and forecasts its signings

Nordec Entretien inc. is a fictional business created for this encyclopedia; its owner, its employees, its customers, its competitors, its figures and the events described are invented, and any resemblance to a real business is coincidental. The property assessment roll gives 1,512 buildings in the size band in the municipalities of the two territories, and a check of 50 records finds 44 eligible, which is 88 percent, so the top-down count is 1,331. Walking 6 of 30 commercial zones counts 262 buildings, which extrapolates to 1,310. The range that survives is 1,310 to 1,331, and Book B4's figure of about 1,400 sits above it. The third territory is counted at 470 buildings raw and 414 adjusted. The forecast is a range of 19 to 23 signings around a central value of 21. Every one of these figures is posed by this book, illustrative and never a forecast of results, and nothing is credited to the study.

WHO DOES THIS The owner sets the unit, the administrator counts, and the dispatcher checks the zones

The roles are the case's, and the hours are posed by this book. Sophie Nadeau, the owner, fixes the unit and the place of the count, writes the filters into the research brief and reads the range when it is printed. The administrator takes the property roll, checks the 50 records by hand against the same criteria every time, and builds the top-down count, as part of the 38 hours of Chapter 11. The dispatcher checks the zones walked against the routes the business actually runs, which is 3 hours. The lead technician walks or checks a zone when asked. In a solo business the owner does all of it, and in a larger business an analyst builds both paths; both are this book's suggestions, not findings. No hour is promised to any reader.

WHAT CHANGED The business counts of December 2025 were released on February 13, 2026

The dated change belongs to the sources and not to the case. Statistics Canada released its business counts for December 2025 on February 13, 2026, as the release read on October 10, 2026 states, and the property assessment roll's 2026 files were created on March 3 and April 1, 2026, with the catalog record modified on August 24, 2026 (Ministère des Affaires municipales et de l'Habitation, 2026). The dates of sources are never dates of the case: the case's counts are made once, on a day of its study, and the second count is posed on day 365 of this book, which is where Chapter 13 places it.

Costly Mistakes

• Counting a market without a decision waiting for the count. A size that no decision consumes is a curiosity, and a business that counts without a decision will find that the count changes nothing and is never repeated.
• Reading a count by location as a count of firms or of buildings. The business counts of a statistics agency are counts of operating locations, each location separately, and a business that reads them as firms or as buildings has read a number in the wrong unit.
• Narrowing a top-down count without checking a sample of records by hand. Every narrowing step is a filter, and a filter that is wrong by a little produces a count that is wrong by a lot; the sample check is what catches it.
• Walking the zones nearest the base and calling the result a market. A bottom-up count rests on the assumption that the zones walked resemble the zones not walked, and a choice made for convenience breaks that assumption in a direction nobody can see.
• Averaging two counts that fail differently. The average of a top-down count and a bottom-up count is a point that neither path supports; what survives the comparison is a range, and the range is what gets printed.
• Counting a stated intention as a sale. Intentions overstate, and measuring them can inflate their association with behavior; the case keeps a posed fraction of its stated switchers and says that the fraction is posed.
• Printing a margin of error beside a count that has no probability sample. A margin belongs to a probability sample with its formula, its confidence level and its population, and a count made by walking zones, a set of reviews or an opt-in panel has no margin at all.

Key Takeaways

• A market size is wanted for a decision, and the count is built to fit the decision: the buyers the offer fits, the place the business serves, and the period the count covers.
• A business that already has customers sizes the reachable market first, because the reachable market is the only one of the three nested counts that a decision can act on this year.
• A top-down count begins with a published count, narrows it by the filters of the decision and checks a sample of records by hand; the case's roll of 1,512 buildings and its check of 50 records leave 1,331, posed by this book.
• A bottom-up count begins with what can be counted on the ground and multiplies it, disclosing how the zones were chosen; the case's 262 buildings in 6 of 30 zones extrapolate to 1,310, posed by this book.
• Two counts that do not share an error leave a range rather than a point, and the case's range of 1,310 to 1,331 sits below Book B4's figure of about 1,400 and below the raw roll of 1,512.
• A forecast is a range with a reason, and a stated intention is never counted as a sale; the case's forecast of 19 to 23 signings, central 21, rests on a trend, a funnel and a posed fraction of 151 stated switchers, and nothing is credited to the study.

Your Action Plan

1. Write the three questions before taking any figure. Name which buyers the count covers, which place it covers and which period it covers, and put the three in the research brief beside the decision the count serves. Deliverable: one paragraph naming the buyers, the place and the period. Verification: the paragraph names a unit, and the unit is the same one the decision needs.
2. Build the top-down count and check a sample by hand. Take a published count, narrow it by the filters of the decision, draw a sample of records and read each one against the same criteria, then multiply the published count by the eligible share and print the raw count beside the adjusted one. Deliverable: one adjusted count with its sample size and its eligible share. Verification: the sample was read against written criteria, and the raw count is printed beside the adjusted one.
3. Build the bottom-up count and disclose how the zones were chosen. Count what can be seen in a set of zones, state the rate that carries the count to the whole territory, and write one sentence on how the zones were chosen and what the choice is likely to have done to the count. Deliverable: one extrapolated count with its assumption printed beside it. Verification: the sentence on the choice of zones is present, and the assumption is stated as an assumption.
4. Set the two counts side by side and print the range. Put the two counts and the figure the business used before on one line, write down why they differ, and print the range that survives with a sentence on what is not counted. Deliverable: one range with its two paths and its assumptions. Verification: no point inside the range is presented as the market, and the earlier figure is kept rather than replaced.
5. Build the forecast from three paths and print it as a range. Extend the trend from the business's own history, build the funnel from proposals to signings, keep a written fraction of the stated intentions and say that the fraction is posed, then print the range and the central value. Deliverable: one forecast range with its three paths. Verification: the fraction kept from the intentions is written down before the year begins, and no margin of error is printed for any path that is not a probability sample.

In La Machine

The application copy read for this book carries one page that touches a market count, and it carries no count of its own (La Machine, 2026). Reseller → Business Explorer business/marketing/business-explorer, reserved to the administrator and reseller roles, searches public business listings, stages them in a transit table with their industry classification code and their source, and exports them or converts them into companies with the prospect status. That is a prospecting page: it stages listings so that somebody can work them, and it is never a market database and never a count of a market. Book B1 (Marketing Fundamentals for Business Owners), chapter 3, sends the reader here, and the page is worth checking in your installation, because what a page offers can differ from one installation to another.

What a business needs for the work of this chapter lives elsewhere, and the list is short. It needs a published count with its unit and its date, read on the site that publishes it, whether that is a statistics agency's release or a property roll. It needs a register read on its public site when the count is of enterprises rather than of buildings. It needs a sample of records checked by hand, with the criteria written down. It needs a spreadsheet that holds both paths side by side, keeps the check sample apart from the full list, and shows a range rather than an average. And it needs a sentence, printed with the figure, that says what is not counted. La Machine holds no market data, counts no market and makes no forecast, and its Business Explorer is never a count of a market; the case counts its market in a spreadsheet and does not use the platform for the count.

Self-Check

1. Why does a business size a market, and what are the three questions that come before any count?
2. Why does a business that already has customers size the reachable market first, and what does Book A1 hold for a venture that has no customers yet?
3. What does a top-down count begin with, what can go wrong at each narrowing step, and how is the result checked?
4. What is the weak point of a bottom-up count, and what does a business do about it?
5. Why is a range printed instead of a point, and why is a stated intention never counted as a sale?

Answers. 1. A business sizes a market because a decision needs a denominator, and a size that no decision consumes is a curiosity that is never repeated; the three questions are which buyers the count covers, which place it covers and which period it covers, and a count that does not answer all three is a number rather than a count. 2. The reachable market is the only one of the three nested counts that a decision can act on this year, because it is the set of buyers the business could serve with the routes, the vehicles, the technicians and the hours it has; Book A1 (From Idea to Business Model) holds the total, serviceable and obtainable markets of a venture that has no customers yet, in chapter 5 of Book A1, and that definition is cited and never rewritten. 3. A top-down count begins with a published count, such as the property roll's 1,512 buildings in the size band; each narrowing step is a filter that can be wrong, and the result is checked by reading a sample of records by hand against the same criteria, as the case's check of 50 records, 44 eligible, gives an adjusted count of 1,331, posed by this book. 4. The weak point is the assumption that the zones walked resemble the zones not walked, and the remedy is to choose the zones at random or, where the choice cannot be random, to disclose how they were chosen and what the choice is likely to have done to the count. 5. A range is printed because two paths that fail differently leave an interval standing and the average of the two is a point neither path supports, and no point inside the range is called the market; a stated intention is never counted as a sale because studies that related stated intention to actual purchase often found substantial variation between them (Jamieson & Bass, 1989, at abstract only) and because measuring intentions can itself inflate their association with behavior (Chandon et al., 2005, at abstract only), so the case keeps a posed fraction of its stated switchers and says that the fraction is posed.

Chapter References

Bass, F. M. (1969). A new product growth for model consumer durables. Management Science, 15(5), 215–227. https://doi.org/10.1287/mnsc.15.5.215

Chandon, P., Morwitz, V. G., & Reinartz, W. J. (2005). Do intentions really predict behavior? Self-generated validity effects in survey research. Journal of Marketing, 69(2), 1–14. https://doi.org/10.1509/jmkg.69.2.1.60755

Choi, H., & Varian, H. (2012). Predicting the present with Google Trends. Economic Record, 88(s1), 2–9. https://doi.org/10.1111/j.1475-4932.2012.00809.x

Jamieson, L. F., & Bass, F. M. (1989). Adjusting stated intention measures to predict trial purchase of new products: A comparison of models and methods. Journal of Marketing Research, 26(3), 336–345. https://doi.org/10.1177/002224378902600307

La Machine. (2026). Documentation — route reference inventory. Internal product documentation, read October 10, 2026, together with the code of the pages this book describes: business/marketing/business-explorer.

Ministère des Affaires municipales et de l'Habitation. (2026). Rôle d'évaluation foncière du Québec [Data set; in French]. Données Québec. (Metadata modified August 24, 2026; 2026 files created March 3 and April 1, 2026; licence CC BY 4.0.) Retrieved October 10, 2026, from https://www.donneesquebec.ca/recherche/dataset/roles-d-evaluation-fonciere-du-quebec

Statistics Canada. (2010). Survey methods and practices (Catalogue no. 12-587-X; originally published October 2003; archived). Retrieved October 10, 2026, from https://www150.statcan.gc.ca/n1/pub/12-587-x/12-587-x2003001-eng.htm

Statistics Canada. (2026, February 13). Canadian business counts, December 2025. The Daily. Retrieved October 10, 2026, from https://www150.statcan.gc.ca/n1/daily-quotidien/260213/dq260213c-eng.htm

Chapter 9 — Turning Research into Decisions

Research changes a decision only when it arrives as a short brief that a person with the authority to decide can read in five minutes, and this chapter is about that last step: turning the evidence of several methods into a decision that is actually taken.

What You'll Be Able to Do

• Name the five ordinary reasons research fails to change a decision, and say what a small business can do about each of them before the fieldwork starts.
• Build an evidence grid that sets one row per condition of the decision rule against one column per method, and report a disagreement between two methods instead of averaging it away.
• Write an insight statement in one sentence that names the evidence, bears on the decision and can be contradicted.
• Keep findings, interpretations and recommendations in three separate paragraphs, and disclose the limits of each method beside its result.
• Draft a decision brief of one page that a person can read in five minutes, and present it to a room that did not ask for the research.
• Keep a research repository in which every file carries its question, its dates, its method, its limits and its retention date, apart from the prospect list.

Why It Matters

Research can fail at the last hundred meters, between the moment the evidence exists and the moment somebody decides. The interviews were conducted, the survey closed, the counts were made, and then the material sat in a folder because nobody had written the one page that says what it means and what to do. A business that has read this far in this book has learned how to ask a question, how to read public sources, how to talk to people, how to survey a sample whose margin it knows, how to read digital traces for what they are, how to watch competitors through what they publish, and how to count a market by two paths. All of that is wasted if the last step is skipped, and the last step is a page.

One reason the last step is skipped is that it requires a judgment, and a judgment can be wrong. A finding is safe: it says what the evidence shows. An interpretation is exposed: it says what the finding may mean. A recommendation is more exposed still: it says what to do, and it puts a name beside a choice. A business that has spent weeks gathering evidence may be reluctant to spend an afternoon deciding, because deciding closes the question and makes the business accountable for an answer.

There is a second reason the last step matters, and it is about honesty rather than effort. When findings, interpretations and recommendations are mixed into one paragraph, the reader cannot tell which part is evidence and which part is the writer's opinion, and the opinion borrows the authority of the evidence. When they are kept apart, a reader who disagrees with the recommendation can still accept the findings, and a reader who doubts a finding can see exactly which method produced it and what that method cannot say. That separation is what a research code asks for, and it is also what makes a brief useful a year later, when the business reads it again and wants to know what it actually knew at the time.

In This Chapter

Section 9.1 explains why research fails to change decisions and sets this chapter's borders against the books around it. Section 9.2 covers synthesis, the evidence grid and the honest treatment of disagreement. Section 9.3 covers insight statements and their three tests. Section 9.4 separates findings, interpretations and recommendations and tests the case's decision rule condition by condition. Section 9.5 builds personas from evidence and shows them only with the number of respondents they summarize. Section 9.6 revisits segments and jobs with new evidence. Section 9.7 writes the decision brief on one page. Section 9.8 presents research to people who did not ask for it. Section 9.9 keeps a research repository with a retention date on every file. Section 9.10 tests the case's rule against the evidence and takes the decision.

9.1 Why Research Fails to Change Decisions

Research can fail to change a decision for five reasons, and none of them is about the quality of the data. The first is that the question was never tied to a decision. A business asks what its customers think about response time, learns something interesting, and finds that no choice in front of it depended on the answer. The study was a curiosity rather than a question, and a curiosity produces a conversation rather than a decision. The remedy is the one Chapter 2 sets out: write the decision first, in one sentence, and write the rule that will follow from the evidence before any fieldwork begins.

The second reason is that the rule was not written first. When the rule is written after the evidence arrives, the evidence is read in the light of what the business already wanted to do, and the rule quietly becomes a description of the answer. A rule written before the fieldwork says what evidence would change the decision, and it makes the answer independent of how impressive the data look. The third reason is that the evidence arrived too late. A decision has a date, and evidence that arrives after the date is not evidence; it is a record. The gates of Chapter 13 exist for this reason, and so does the five-minute brief of section 9.7: a brief that can be read in five minutes can be read before the decision rather than after it.

The fourth reason is that the report was too long. A thirty-page report is read by the person who wrote it and by nobody else, and the person who wrote it already knows what it says. The fifth reason is that nobody had the authority to decide on it. A brief that lands on a desk where the decision cannot be taken is a brief that produces a request for more research, and more research produces another brief on the same desk. The remedy is to name the decider before the study begins, to write the brief for that person, and to keep it to a page.

No source read for this book measures how often research is used, ignored or acted upon, and this chapter prints no such share. The five reasons above are this book's reasoning, not a finding with a frequency, and a business should treat them as a checklist. A brief that is short, carries its sources and its limits, and is addressed to a named person with the authority to decide can be acted upon, and the business controls all three before the first interview is booked.

This book describes a method of decision. It is not legal, tax or accounting advice, it does not replace the advice of a lawyer, an accountant or a research professional, and it recommends no vendor, no survey tool, no panel, no research firm, no fieldwork service, no data provider, no platform and no service to any reader.

The table below sets six neighbors beside what each of them holds and what this chapter holds, so that the reader can see where the work of this chapter stops. Read it as a boundary rather than as a summary: the middle column is another book's subject, the right column is this chapter's, and the two never overlap.

Table 9.1 — Where this chapter stops and the books around it start

DEFINITION Finding, interpretation, recommendation, insight statement and decision brief, as this book uses the words

These are this book's working words, and no source read defines them in this sense. A finding is what the evidence shows, with its source and its limits attached. An interpretation is what the finding may mean, written as a possibility rather than as a result. A recommendation is what to do, and it belongs to the person who decides rather than to the person who gathered the evidence. An insight statement is one sentence that says what was found, for whom and why it matters to the decision. A decision brief is the page a person can read in five minutes and decide from. The distinction is not this book's invention: the international code of research asks that research be fit for purpose, that its limitations be disclosed and that findings be distinguished from interpretation and recommendations (International Chamber of Commerce & ESOMAR, 2025, art. 7). The code is self-regulation; it binds its members and the associations that adopt it, never a business by law.

9.2 Synthesis: Putting the Evidence Side by Side

Synthesis is the reading of several methods against one question, and it starts from the decision rule rather than from the data. This is the discipline that separates a synthesis from a summary. A summary walks through the methods in the order they were run: the desk study found this, the interviews found that, the survey found something else. A synthesis walks through the conditions of the rule and asks, for each condition, which methods bear on it and what they say. The difference is not stylistic. A summary produces a document that is read once; a synthesis produces a table in which a gap is visible, and a visible gap is a reason to gather one more piece of evidence rather than a reason to guess.

Thematic analysis is a way of turning a set of conversations into themes, and it is described in the literature as a flexible and accessible method for qualitative data that was poorly demarcated, with practical guidance available for doing it rigorously, with its pitfalls named and its advantages and disadvantages weighed (Braun & Clarke, 2006, at abstract only). The abstract read for this book gives neither the number nor the names of the method's phases, and this chapter prints none of them.

The questions that a study of the voice of the customer asks are the questions a business asks itself when it reads its own interviews. The abstract reports that the study examines the customer-needs stage of quality function deployment, the identifying, structuring and prioritizing of needs, and it asks how many customers to interview, how many analysts should read the transcripts, which needs analysts miss, and whether focus groups or one-on-one interviews are better; it compares group consensus charts with customer sorts and tests whether frequency of mention can stand for importance (Griffin & Hauser, 1993, at abstract only). The abstract states the questions and not the results, so this chapter prints no number of interviews and no verdict on focus groups from it. What the questions give a business is a set of cautions: the number of conversations is a choice, the reader of a transcript is a filter, and a theme that appears often is not automatically the theme that matters most.

The instrument that holds a synthesis together is an evidence grid. It has one row per condition of the decision rule and one column per method, and each cell holds the finding, its source and its strength. Strength is this book's own word for how many independent methods agree on the same condition: a condition supported by two methods that were run separately is stronger than a condition supported by one, and a condition supported by a method that cannot speak to it at all is not supported by that method. The grid is filled from the evidence and never from the rule, which is why the rule is written at the top of the page rather than in the middle of it.

When two methods disagree, the disagreement is reported and not averaged. Averaging two methods that measure different things produces a number that belongs to neither of them, and it hides the most useful fact the business has: that its interviews and its survey point in different directions. A disagreement is a finding in its own right, and it usually has a cause that can be named: the two methods reached different populations, or asked different questions, or were run at different moments. Naming the cause is what turns a disagreement into a decision rather than into an argument.

The table below sets four findings of the case beside the condition each bears on, the method that produced it, its source, its strength and what it cannot say. The findings are posed by this book and illustrative, and the strength column is this book's own word for how many independent methods agree, not a measure computed from any source.

Table 9.2 — An evidence grid in six columns

WATCH OUT A quotation read as proof, a persona read as a person and a theme read as an importance are three ways research misleads

The first trap is a quotation that stands for a whole group. One vivid sentence from one interview is memorable, and memory is not evidence; the question that catches it is how many respondents said it, and who did not. The case prints no quotation of any respondent, and it gives themes, codes and counts instead. The second trap is a persona that gets a name and a face and is then treated as a real buyer; the question that catches it is how many respondents the persona summarizes, and a persona shown without that number is a story rather than a summary. The third trap is a theme that appears often and is taken to matter most; whether frequency of mention can stand for importance is a question that the study cited above states and does not answer, so a business that reads frequency as importance is answering a question the research left open. None of the three traps has a frequency in any business, and this book prints none.

9.3 Writing Insight Statements

An insight statement is one sentence that says what was found, for whom and why it matters to the decision. It is written from the evidence grid and never from a single quotation, and it is written so that it can be contradicted. That last property is what makes it useful. A sentence that cannot be wrong cannot be tested, and a sentence that cannot be tested cannot change a decision. The statement is not a slogan and not a summary of the study; it is the shortest form of the finding that still carries the evidence behind it.

Three tests decide whether an insight statement is worth writing. The first test asks whether it names the evidence. A statement that says customers care about speed names nothing; a statement that says owner-run buildings are 44.6 percent of the survey's completes and name proof of response time among their themes names two methods, a count and a theme. The second test asks whether it bears on the decision. If the statement is true and the business would decide exactly the same way, the statement is interesting and not useful, and it belongs in the repository rather than in the brief. The third test asks whether it can be wrong. A statement that can be wrong is a statement that a further piece of evidence could overturn, and a business that writes such statements is a business that knows what it would take to change its mind.

The case gives one insight statement, posed by this book and illustrative: owner-run buildings are 44.6 percent of the survey's completes and name proof of response time among their themes. The statement passes the three tests. It names its evidence, which is a count from the probability sample of the survey and a theme from the interviews. It bears on the decision, because the decision is about where growth should go and the statement identifies a reachable base in the two territories the business already serves. And it can be wrong: a second survey with the same frame and questions could find a different share, and a further set of interviews could fail to find the theme again.

An insight statement is not a recommendation, and the difference matters at the moment the brief is written. The statement above says what was found and for whom; it does not say to go after owner-run buildings. The recommendation that follows from it belongs to section 9.10, where it is written as a recommendation and kept apart from the finding it rests on. A recommendation written as an insight statement hides a choice inside a description.

A study should produce few insight statements. Twelve of them make a summary rather than a synthesis, and a brief that carries twelve will not be read in five minutes. This book suggests as many statements as the decision rule has conditions, plus one for anything the evidence showed that the rule did not anticipate, with the rest kept in the repository as findings.

The statements are also the place where a business records what it did not find. A statement that says the survey found no measurable rise in awareness is as useful as one that says it found a rise, and it is more honest, because it reports the result the business did not want. The case's awareness measure is the example: the survey's prompted awareness is 35 of 92, which is 38 percent with a margin of plus or minus 9.5 points at 95 percent confidence, and Book B3 (Branding and Visual Identity) prints an earlier measure of 19 of 60, which is 31.7 percent with a margin of about plus or minus 11.8 points. The two intervals overlap, so the rise is not shown, and the honest statement says so rather than reporting the higher number alone.

KEY CONCEPT Research changes a decision when it answers the question that the decision rule asked, in a page that is read

The central idea of this chapter is that a rule written before the fieldwork is what makes research decisive. The rule says, in advance, what evidence would change the decision, so the brief has only to report that evidence against the rule, and the answer does not depend on how impressive the data look or on who gathered them. A study that answers a question nobody asked, however well it was run, changes nothing; a study that answers the rule's question with modest evidence changes the decision, because the decision was waiting for it. No source read for this book measures how often research is used or acted upon, and this chapter prints no such share. The findings of the case are posed by this book and illustrative, and they are counts rather than norms: the case counts its own completes, its own interviews and its own tenders, and nothing in them is offered as a benchmark for another business.

9.4 Separating Findings, Interpretations and Recommendations

A finding is what the evidence shows. An interpretation is what the finding may mean. A recommendation is what to do. The three are written in three separate paragraphs, in that order, and they are never mixed, because a reader who cannot tell them apart cannot weigh them. The separation is what the international code of research asks for: research must be fit for purpose, its limitations must be disclosed, and findings must be distinguished from interpretation and recommendations, and known or suspected biases must be disclosed as well (International Chamber of Commerce & ESOMAR, 2025, art. 7 and art. 8). The code is self-regulation, revised in 2025, and it binds its members and the associations that adopt it rather than a business by law.

The disclosure standards of a national association of the United States, used here as an object of comparison and never as a Canadian rule, list what is disclosed when results are released, among them the exact wording of the questions, the population, whether the sample is probability-based or not, the sample sizes, the margin of sampling error for probability samples and a statement on unmeasured error; for non-probability samples, measures of precision are reported only if they are defined and accompanied by the model and its assumptions (American Association for Public Opinion Research, 2021, revised April 2021). A one-page brief cannot carry all of it. What the brief carries is the part that bears on the decision, and the rest stays in the repository where the study file holds it.

Figure 9.1 separates findings, interpretations and the recommendation in three columns, and tests the case's decision rule, written before fieldwork, against the evidence.

Figure 9.1 — From evidence to decision. The findings carry their sources, the interpretations are kept apart from them, and the recommendation follows: deepen the two territories, first among owner-run buildings. The rule written before fieldwork passes on the adjusted count of 414, at least 400, and fails on the integrator's 2 tenders won of 4 seen, so the rule says deepen the two territories.

The case gives three findings, three interpretations and one recommendation, all posed by this book and illustrative. The first finding is that the third territory counts 470 buildings raw and 414 adjusted, from the desk study of the property assessment roll with a check of a sample of records. The interpretation is that the third territory is large enough to be worth entering on the count alone. The second finding is that the regional integrator won 2 of the 4 public maintenance tenders seen in that territory, from the watch of published tenders and awards. The interpretation is that bundled contracts already hold much of that territory, which is a reason for caution rather than a reason to enter. The third finding is that 41 of the 92 survey completes, which is 44.6 percent, are buildings not managed by a property management firm. The interpretation is that owner-run buildings are a reachable base in the two territories the business already serves.

The recommendation follows from the rule rather than from the interpretations, and this is the point of writing the rule first. The rule says to open the third territory only if its adjusted count is at least 400 buildings in the size band and the regional integrator won fewer than half of the public maintenance tenders seen there; otherwise the business deepens the two territories. The first condition passes, because 414 is at least 400. The second condition fails, because 2 of 4 is exactly half and not fewer than half. A rule that requires both conditions is not satisfied by one of them, so the rule says to deepen the two territories, and the recommendation adds the first move: among owner-run buildings.

The separation also governs what the brief says about its own limits. The count of the third territory comes from a desk study of a public roll, and it cannot say whether the buildings are reachable or whether their owners buy maintenance. The tender count comes from a watch of what public bodies published, and it cannot say whether the tenders seen are all the tenders there were. The survey share comes from a probability sample with a margin printed for it, and it cannot say why those buildings are not under a management firm. The interview theme comes from a qualitative method, and it cannot say how many buyers hold the theme. Each limit is written beside the finding it belongs to, in the same paragraph, and never in a general note at the end.

THE RULE Findings, interpretations and recommendations are kept apart, and limitations are disclosed with the results

The rule this chapter draws is a rule of self-regulation, not of law. Research must be fit for purpose, its limitations must be disclosed, and findings must be distinguished from interpretation and recommendations (International Chamber of Commerce & ESOMAR, 2025, art. 7); known or suspected biases must be disclosed (art. 8). The disclosure standards of the association cited above ask, when results are released, for the wording of the questions, the population, whether the sample is probability-based or not, the dates, the sample sizes and, for probability samples only, the margin of sampling error, with a statement on unmeasured error (American Association for Public Opinion Research, 2021). A code binds its members and the associations that adopt it, never a business by law, and the association is a body of the United States used here as a comparison. The example is the brief of the case, which prints a margin only for the probability sample of its survey.

9.5 Personas Built From Evidence

A persona is a composite of respondents that makes a pattern easy to hold in mind, and in this book it is shown only with the number of respondents it summarizes. That single rule prevents most of the trouble a persona can cause. A persona without a number is a character, and a character invites a business to design for an imagined buyer rather than for the people who answered. A persona with a number is a summary, and a summary can be checked against the evidence behind it. Personas as a concept belong to Book B1 (Marketing Fundamentals for Business Owners), which treats segments, the ideal customer profile and personas in chapters 3 and 4 of that book; this chapter builds one from the evidence of a study and does not restate the concept.

The literature on data-driven personas is a caution as much as a guide. The abstract of a review of 77 data-driven persona articles published between 2005 and 2020 reports three periods, which the abstract calls quantification, diversification and digitalization, and it names remaining gaps in shared resources, evaluation, standardization, inclusivity and the risk of losing in-depth insight, concluding that organizations should assess their readiness (Salminen et al., 2021, at abstract only). The abstract supports no method step, and this chapter takes none from it. A business adopting the practice should look at its own readiness.

A shorter and sharper caution comes from a proceedings paper on the method itself. The abstract argues that it is hard to know how many users, if any, a persona represents, that personas cannot be adequately verified or falsified and so lack demonstrable validity, and that they may cause political conflicts; until those problems are resolved, the paper concludes, personas are best not used to communicate data (Chapman & Milham, 2006, at abstract only). The paper is a short proceedings paper rather than a study of business buyers, and it supports no claim about any particular market. It is the reason this book requires a persona to carry the number of respondents it summarizes, and the reason a persona is never the only form in which a finding is presented.

The case builds no persona beyond a one-line sketch of the owner-run building, posed by this book and tied to 41 of the 92 respondents of the survey. The sketch is deliberately thin, because the evidence behind it is thin: the survey says that 41 of 92 completes are not managed by a property management firm, which is 44.6 percent, and the interviews say that one contact rather than a bundle is one of four themes. Everything else in the sketch is an interpretation and is labeled as one. A business that fills in the rest of the sketch from imagination has built a character, and the character will be wrong in ways nobody can see.

The table below sets five elements of a persona beside the evidence behind each, the number of respondents summarized and what the element is not. The numbers in the respondents column are posed by this book and illustrative, and they are the case's own counts rather than a benchmark for any other business.

Table 9.3 — A persona and its evidence

9.6 Segments and Jobs Revisited With New Evidence

New evidence can confirm a segment, split it in two or redraw it, and the decision about which of the three has happened is a judgment rather than a calculation. A segment drawn from the business's own records and impressions is a hypothesis about who buys, and a study is the test of that hypothesis. When the evidence confirms the segment, the business keeps it. When the evidence splits it, two groups with different needs were being served with one message. When the evidence redraws it, the line was in the wrong place.

The idea that a buyer hires a product to make progress in a given circumstance, and that the job rather than the buyer's attributes explains the choice, comes from Book A6 (Designing a Value-Added Offer), which prints it, and this chapter reuses it at that level only and does not re-read it (Christensen et al., 2016). The idea is useful here because it moves the question from who the buyer is to what the buyer is trying to get done, and a study can answer the second question more reliably than the first. A business that asks a property manager what a maintenance contract is for will get an answer about the job; a business that asks what kind of person the property manager is will get a description that cannot be acted upon.

The case's survey separates two groups inside the frame. Of the 92 completes, 51 are buildings managed by a property management firm, which is 55.4 percent with a margin of plus or minus 9.7 points at 95 percent confidence, and 41 are not, which is 44.6 percent. The two groups are not merely different in who answers the phone. The interviews suggest that the job is different: the theme of one contact rather than a bundle appears in the conversations, and it points to a buyer who wants a single person accountable for a building rather than a package of services with several hands in it. That reading is an interpretation, posed by this book as an interpretation and never as a finding, and it is the kind of reading that a further round of interviews could confirm or overturn.

The practical consequence of a revisited segment is small and concrete. A segment that has been revisited with research evidence is entered by hand in the platform, because no segment is drawn from a survey: Marketing → Segments (business/marketing/segments) groups contacts by tags and fields, and a business that wants a segment to reflect what a study found enters the tags itself (La Machine, 2026). That is a limitation of the tool rather than a defect of the method, and it is worth stating plainly, because a business that expects a survey to populate its segments will wait for something that does not happen.

The case's segment work ends with a decision about the first move rather than with a new segment definition. The two territories contain both groups, and the rule's answer is to deepen those territories; the recommendation adds that the first move is among owner-run buildings, which are 41 of the 92 respondents. Nothing in that sentence is a claim about the market as a whole, and nothing in it is credited to the study: the rule decides, and the study reports what the rule asked for.

IN LA MACHINE Responses list submissions, segments are entered by hand, and no page keeps research

Two pages of the application copy read for this book touch the work of this chapter, and neither of them does the work. Marketing → Forms → Responses (business/marketing/forms/responses) lists the submissions of a form; it weights nothing, computes no margin and draws no sample (La Machine, 2026). Marketing → Segments (business/marketing/segments) groups contacts by tags and fields, and a segment revisited with research evidence is entered by hand, because no segment is drawn from a survey. Nordec Entretien inc. uses the form for its questionnaire only where a contact is acceptable, keeps the three competitors as company records with dated notes and holds its weekly watch as alerts and tasks; the survey of the case is run by the fieldwork service outside the application, and its answers stay in the research file, apart from the prospect list. La Machine holds no repository of research, builds no persona and writes no brief. Check it in your installation.

9.7 The Decision Brief

The decision brief is one page, and a person can read it in five minutes. That is the whole specification, and everything else in this section is a way of meeting it. The brief is not a summary of the study and not a report of the method; it is the page from which a decision can be taken. A brief that requires the reader to consult another document has failed, and a brief that runs to four pages has failed in a different way, because it will be read in a hurry or not at all. The five-minute test is not a metaphor: a business can time the reading, and the timing is the verification criterion.

The brief has six parts, in a fixed order. The first is the decision and the rule, copied from the one-page research brief that Chapter 2 describes, so that the reader sees at the top what is being decided and what evidence was going to settle it. The second is the evidence, with its sources and its limits, one paragraph per condition of the rule. The third is the options, stated as the choices the rule leaves open rather than as a menu of everything the business could do. The fourth is the recommendation, in one sentence, with the name of the person who will carry it out. The fifth is what would change the recommendation, which is the honest statement of the brief's own fragility. The sixth is the next step, with a date.

The order of the parts matters, and so does the order inside the second part. Findings come first, then interpretations, then the recommendation, in three separate paragraphs, as section 9.4 sets out. A brief that opens with a recommendation and then lists the evidence that supports it is an argument rather than a brief, and a reader who disagrees with the recommendation will stop reading at the first line. A brief that opens with the decision and the rule, then gives the evidence, then the interpretation, then the recommendation, gives the reader the chance to reach a different conclusion from the same evidence, which is what a decision-maker is for.

The brief also says what was not found and what was not asked. A study that did not measure price sensitivity should say so, because a reader who assumes it did will use the brief for a decision it cannot support. The case's brief says that the third territory's count comes from a public roll and not from a walk of every zone, that the tender count is small, that the survey's share carries a margin printed for the probability sample, and that the interviews cannot say how many buyers hold a theme. Each of those sentences costs a line and saves an argument.

The table below sets six parts of the brief beside what each says, how long it should be, what it never holds and the case's line. The lengths are this book's suggestion rather than a finding, and the case's lines are posed by this book and illustrative.

Table 9.4 — The decision brief on one page

METHOD Writing the decision brief in five steps

First, copy the decision and the rule from the one-page research brief of Chapter 2, word for word, so that the reader sees what was going to settle the question before any evidence was gathered. Second, list the findings with their sources and their limits, and nothing else, one paragraph per condition of the rule. Third, write the interpretations in a separate paragraph, each tied to the findings it rests on, and label them as interpretations. Fourth, write the recommendation in one sentence, name the person who will carry it out, and add what would change the recommendation. Fifth, cut the page until a person can read it in five minutes, and check the verification criterion: the decision can be taken from the page alone, without consulting another document. This is this book's reasoning and not a tested procedure, and no source read measures how long a brief should be.

9.8 Presenting Research to People Who Did Not Ask for It

Presenting research to people who did not ask for it is a different task from writing it, and it has its own order. Begin with the decision and the answer, then the evidence, then the limits. A room that is told the answer first can listen to the evidence with a purpose; a room that is walked through the method first will decide, somewhere in the second slide, that the presentation is not for them. The order is not a trick of persuasion. It respects the time of people who did not commission the work and who have their own decisions to make that day.

The room reads a number only if its unit and its source are on the same line. A slide that says 44.6 percent and nothing else invites the question of what was counted, and the question will be asked by the person least persuaded by the answer. A slide that says that 41 of the 92 completes of the survey are buildings not managed by a property management firm, with the margin printed for the probability sample, answers the question before it is asked. The same rule applies to a count: 414 buildings adjusted from a public roll is a sentence, and 414 alone is a number that will be repeated later without its meaning.

A presenter should expect three objections. The first is that the sample is small, and the honest answer is to say what the sample was and what margin it carries, and to say that the interviews are not a sample at all. The second is that the competitor is different, and the honest answer is to say what was read about the competitor and what was not, and to keep the watch to what the competitor publishes. The third is that the market has changed since the study, and the honest answer is to say when each piece of evidence was gathered and what would be done differently now. None of the three objections is answered by repeating the finding more loudly.

A presentation never hides a result that fails the rule. If the second condition of the case's rule fails, the presentation says that it fails, in the same tone as it says that the first condition passes. This is the discipline that makes a rule written in advance worth writing: a rule that is only reported when it agrees with the business's preference is not a rule. The case's presentation reports both conditions, the pass and the fail, and the recommendation follows from the rule rather than from the presenter's view of the third territory.

The meeting in which a decision is taken is Book I7 (Communication, Meetings and Decision-Making), which holds the agenda, the conduct and the record of a decision meeting; this chapter holds the page that is carried into it. The two are different subjects and they meet at one point: the brief is read before the meeting, so that the meeting can decide rather than listen. A business that presents a brief for the first time inside the decision meeting has spent the meeting on reading rather than on deciding.

In the case, Sophie Nadeau, the owner, reads the brief and decides, and nobody else is asked to approve it. That is a fact about the size of the business rather than a principle: a business with a board, a partner or a lender will have more than one reader, and the order of the page and the separation of the three kinds of sentence do not change with the number of readers.

WHO DOES THIS The owner decides, the administrator writes the brief, and the outside lawyer reads what leaves the firm

In the case, posed by this book and illustrative, Sophie Nadeau, the owner, reads the decision brief and decides; the administrator assembles the evidence grid, writes the findings and keeps the repository, within the 38 hours the case counts for the administrator; the lead technician checks that the technical words of the interpretations are right, because a sentence about response time that a technician would not recognize is a sentence that will not survive the first service call; and the outside lawyer reads any text that leaves the firm and any question about how long a file may be kept, which belongs to Book H2 (Privacy and Personal Information). In a business of one, the owner writes the brief and decides in one sitting, which is this book's suggestion and not a figure in the model. In a larger business, an analyst writes the brief and a manager decides, which is also this book's suggestion. No hour is promised to any reader.

9.9 Keeping a Research Repository

A research repository is a folder of the research files, and each file carries its question, its dates, its method, its limits, its raw material and its brief. The purpose is not archiving for its own sake. It is that the work can be redone. A study whose method is not written down cannot be repeated, and a study that cannot be repeated cannot be compared with the next one. The case's awareness measure is the example: the survey's 35 of 92 can be compared with Book B3's 19 of 60 only because both measures were taken with a stated method and a stated margin, and the comparison shows that the rise is not shown.

The repository is also where the separation of section 9.4 is preserved over time. A file that holds findings, interpretations and recommendations in one document will be read a year later as though all of it were evidence, and the recommendation will borrow the authority of the count. A file that keeps the three apart can be read a year later for exactly what it holds. The repository is not a place where the business stores its conclusions; it is a place where it stores its questions and what it did about them.

Research data is kept apart from the prospect list, and this is a rule of the code rather than a preference. The code asks researchers to keep a clear distinction between research and non-research activities, to inform data subjects of any non-research use before collection and to obtain separate consent for it, and it asks that personal data not be transferred to a client without consent for that purpose, with protection against re-identification by deduction and retention no longer than necessary (International Chamber of Commerce & ESOMAR, 2025, art. 1(d)-(e) and art. 6). The code is self-regulation and binds its members and the associations that adopt it, never a business by law. In the case, a respondent who answers the questionnaire is not put on a mailing list and is not called back with an offer because of an answer.

The retention of a research file is written when the file is opened. The Personal Information Protection and Electronic Documents Act, Schedule 1, clause 4.5, asks that personal information be retained only as long as necessary and then destroyed, erased or made anonymous, as the Québec and Canada note below sets out and as Book H2 (Privacy and Personal Information) explains. A business that writes a retention date on each file at the moment the file is created has answered the question in advance, and a business that writes it later has a folder of undated material.

The platform's part in this is small and worth stating exactly. Marketing → Forms → Responses (business/marketing/forms/responses) lists the submissions of a form, and it weights nothing, computes no margin and draws no sample (La Machine, 2026). A respondent who submits a form becomes a contact, so the case runs its survey outside the application and keeps the answers in the research file, apart from the prospect list. The rules of contacts belong to Book N4 (CRM, Contacts and Prospect Data), and the personal information of participants belongs to Book H2.

The case keeps a folder per study and a log of the files, posed by this book and illustrative. The log carries, for each file, the question it answers, the dates it was gathered and read, the method, the limits, the retention date and the location of the brief. The log is one page, and it is the document that makes the repository usable a year later, when the business wants to know what it knew and when it knew it.

QUÉBEC & CANADA NOTE Research files are kept only as long as necessary, then destroyed, erased or made anonymous

In Canada, the Personal Information Protection and Electronic Documents Act, S.C. 2000, c. 5, Schedule 1, clause 4.5, current to September 21, 2026 as its entry gives it, limits the use and disclosure of personal information to the identified purposes unless the individual consents or the law requires, and asks that the information be retained only as long as necessary and then destroyed, erased or made anonymous. A research repository holds respondents' answers, so the retention of each file is written when the file is opened rather than when somebody asks. Whether a given repository raises such a question is Book H2's question, and this book never decides it. This book explains the text and recommends no vendor. A business that wants to know what its own repository requires should read Book H2 (Privacy and Personal Information) and speak to a lawyer.

TOOLS OF THE TRADE What any repository or brief template must offer, whatever it is called

The functions are what matter, and no brand name and no price belongs on this list. A repository needs a page per study holding the question, the rule, the dates, the method and the limits; a place for each finding with its source; a separate place for interpretations and a separate place for recommendations, so that the three cannot be read as one; a log of files with a retention date on each; a view that keeps respondents apart from any prospect list; and an export in a form another tool can read, so that the work is not trapped in one place. A brief template needs the six parts of section 9.7 in a fixed order, a field for the source of every number, and a length that a person can read in five minutes. A business that has those functions has a repository and a brief, whatever the tools are called.

9.10 The Case's Decision

Nordec Entretien inc. tests its decision rule condition by condition, and the test is short because the rule was written before any fieldwork. The rule says to open the third territory only if its adjusted count is at least 400 buildings in the size band and the regional integrator won fewer than half of the public maintenance tenders seen there; otherwise the business deepens the two territories. The first condition asks for a count of at least 400, and the third territory counts 470 buildings raw and 414 adjusted, which is 14 above the threshold, so the first condition passes. The second condition asks that the regional integrator won fewer than half of the tenders seen, and the regional integrator won 2 of the 4 public maintenance tenders seen there, which is exactly half and not fewer than half, so the second condition fails.

The rule requires both conditions, and one of them has failed, so the rule says to deepen the two territories. The recommendation adds the first move: among owner-run buildings, which are 41 of the 92 respondents of the survey, or 44.6 percent. Every one of those figures is posed by this book and illustrative, and none of them is a norm, a benchmark or a forecast. The counts are the case's own, read on the day the case read them, and another business counting its own market would arrive at its own numbers.

The brief that carries the decision has three findings, three interpretations and one recommendation, all posed by this book. The findings are the adjusted count of 414 in the third territory, the 2 of 4 tenders won there by the regional integrator, and the 41 of 92 completes that are buildings not managed by a property management firm. The interpretations are that the third territory is large enough to be worth entering on the count alone, that bundled contracts already hold much of it, and that owner-run buildings are a reachable base in the two territories the business already serves. The recommendation is to deepen the two territories, first among owner-run buildings. The three kinds of sentence sit in three separate paragraphs, in that order, and a reader who disagrees with the recommendation can still accept the findings.

What would change the recommendation is named without a number, because naming it with a number would turn a condition into a forecast. A further public maintenance tender seen in the third territory and won by somebody other than the regional integrator would move the second condition, and a second count of the third territory taken by a different path would test the first. Neither of those is a prediction, and neither is scheduled. The brief says what would change its own conclusion, which is what makes the conclusion usable rather than final.

Nothing is credited to the study. The rule decides, and no contract, sale, signing or share is attributed to the research, the insight statements or the watch. The study cost $3,903 in all, posed by this book, which is the internal time of the administrator, the lead technician and the dispatcher at the rates Book B5 (Content Marketing and the Content Factory) prints, plus the cash paid to the fieldwork service; the owner's 22 hours are counted and not priced, as Book B5 does. The revenue at stake, last year's signings multiplied by the average contract of about $8,541 a year that Book B4 (The Marketing Plan, Budget and KPIs) prints, is $162,279, and the study is 2.4 percent of it. Those figures describe the cost of asking a question, and they say nothing about what the answer will produce.

The decision is taken by the owner, who reads the brief and decides, and the brief is filed in the repository with its question, its dates, its method, its limits and its retention date. Chapter 10 puts the routine on the calendar: the weekly round of the watch, the monthly signal review and the quarterly market brief, so that the evidence that produced this decision is refreshed rather than remembered.

DO THE MATH The decision rule of the case, tested in five steps

Illustrative figures; the case is fictional. Step 1 — the first condition asks for an adjusted count of at least 400 buildings, and the third territory counts 470 raw and 414 adjusted, which is 14 above 400, so the condition passes. Step 2 — the second condition asks that the regional integrator won fewer than half of the public maintenance tenders seen, and it won 2 of the 4 seen, which is exactly half, so the condition fails. Step 3 — the rule needs both conditions, so it says to deepen the two territories. Step 4 — of the 92 completes, 51 are managed by a property management firm and 41 are not, which is 41 divided by 92, or 44.6 percent, rounded to the nearest tenth. Step 5 — the first move is among the 41 owner-run buildings of the survey. The figures are posed by this book, are counts and not a norm, and credit no contract or sale to the study.

CASE IN POINT Nordec Entretien inc. decides to deepen its two territories

Nordec Entretien inc. is a fictional business created for this encyclopedia; its owner, its employees, its customers, its competitors, its figures and the events described are invented, and any resemblance to a real business is coincidental. The rule was written before any fieldwork: open the third territory only if its adjusted count is at least 400 buildings in the size band and the regional integrator won fewer than half of the public maintenance tenders seen there; otherwise deepen the two territories. The third territory counts 470 buildings raw and 414 adjusted, so the first condition passes. The regional integrator won 2 of the 4 public maintenance tenders seen there, which is exactly half, so the second condition fails. The rule therefore says to deepen the two territories, and the recommendation adds the first move among owner-run buildings, which are 41 of the 92 respondents of the survey. The owner reads the brief and decides. Every figure here is posed by this book, illustrative and not a forecast, and no result of the decision is claimed.

WHAT CHANGED The code of research was revised in 2025, and its articles on findings and limitations were read on October 10, 2026

The international code of research was revised in 2025 after its 2016 version, and it applies to everyone involved in research, including users of self-service platforms, with the client responsible for making its contractors comply; it asks that the client be informed when artificial intelligence or synthetic data are used (International Chamber of Commerce & ESOMAR, 2025, at the level read, articles 1 to 8). The articles read on October 10, 2026 ask that research be fit for purpose, that limitations be disclosed, that findings be distinguished from interpretation and recommendations, and that known or suspected biases be disclosed. The code is self-regulation and binds its members and the associations that adopt it, never a business by law. Dates of sources are never dates of the case, which is counted in days: the case's decision brief is read by the owner on day 30 of the plan that Chapter 13 sets out.

Costly Mistakes

• Running a study that no decision was waiting for. A question that is not tied to a choice produces an interesting conversation and no change, and the remedy is to write the decision and its rule before the first interview is booked.
• Writing the decision rule after the evidence arrives. A rule written late is read in the light of what the business already wanted to do, and it quietly becomes a description of the answer rather than a test of it.
• Mixing findings, interpretations and recommendations in one paragraph. A recommendation written as a finding borrows the authority of the evidence, and a reader who disagrees with the choice is forced to disagree with the count as well.
• Averaging two methods that disagree. An average of an interview theme and a survey share belongs to neither method, and it hides the most useful fact the business has, which is that its evidence points in two directions.
• Showing a persona without the number of respondents it summarizes. A persona without a number is a character, and a character invites a business to design for an imagined buyer rather than for the people who answered.
• Writing a brief that cannot be read in five minutes. A brief that requires another document has failed, and a brief that runs to four pages will be read in a hurry or not at all, which is the same as not being read.
• Keeping respondents' answers in the prospect list. An answer given to a research question is not a reason to send an offer, and a repository that is not kept apart from the prospecting file turns a study into a mailing list.

Key Takeaways

• Research changes a decision when the decision and its rule were written first, when the evidence answers that rule, and when the answer arrives as a page a person can read in five minutes.
• A synthesis starts from the conditions of the rule rather than from the methods, and an evidence grid with one row per condition and one column per method makes a gap visible instead of hiding it.
• Findings, interpretations and recommendations are written in three separate paragraphs, in that order, with the limits of each method beside its result, as the international code of research asks of its members.
• An insight statement says what was found, for whom and why it matters to the decision, in one sentence that names its evidence and can be contradicted.
• A persona is a composite shown with the number of respondents it summarizes, and the case's sketch is tied to 41 of the 92 respondents of its survey, posed by this book.
• The decision brief has six parts on one page, and the case's rule passes on the adjusted count of 414 and fails on the regional integrator's 2 tenders won of 4 seen, so the rule says to deepen the two territories, first among owner-run buildings.

Your Action Plan

1. Write the decision and its rule before anything else. State the choice in one sentence and the condition that will settle it, in advance, and keep both at the top of every page that follows. Deliverable: a decision sentence and a rule with named conditions. Verification: the rule names what evidence would change the decision, and it was written before the first interview was booked.
2. Build the evidence grid from the rule. Give the grid one row per condition and one column per method, and fill each cell with the finding, its source and its strength. Deliverable: one grid, with a visible gap wherever a condition has no evidence. Verification: every cell names a method and a source, and no cell holds an interpretation.
3. Write the insight statements and test each one. Write one statement per condition of the rule, plus one for anything the evidence showed that the rule did not anticipate, and put each through the three tests. Deliverable: three or four statements of one sentence each. Verification: each statement names its evidence, bears on the decision and can be contradicted.
4. Draft the brief in six parts and time the reading. Write the decision and the rule, the evidence with its limits, the options, the recommendation, what would change it and the next step, in that order, on one page. Deliverable: a one-page brief with the findings, the interpretations and the recommendation in separate paragraphs. Verification: a person who did not write it can read it in five minutes and take the decision from the page alone.
5. Open the repository file with its retention date. Create one file per study, holding the question, the dates, the method, the limits, the raw material and the brief, and write the retention date on the day the file is opened. Deliverable: a folder per study and a one-page log of the files. Verification: every file carries a retention date, and no respondent's answer appears in the prospect list.

In La Machine

Two pages of the application copy read for this book touch the work of this chapter, and both are narrower than the work. Marketing → Forms → Responses (business/marketing/forms/responses) lists the submissions of a form, where the answers of a questionnaire can be read as they arrive, and Marketing → Segments (business/marketing/segments) groups contacts by tags and fields, where a segment revisited with research evidence is entered by hand (La Machine, 2026). Because a respondent who submits a form becomes a contact, the survey of the case is run by the fieldwork service outside the application. Check it in your installation, because what a given installation shows depends on its own configuration.

What the reader needs elsewhere is the rest of the work, and it is worth naming plainly. La Machine holds no repository of research, builds no persona, writes no brief, weights no answer and draws no segment from a survey. The evidence grid, the insight statements, the decision brief of one page, the repository with a retention date on every file and the person with the authority to decide all live outside the platform, in a document, a folder and a calendar. A business that expects the platform to produce a brief will wait for something that does not happen.

Self-Check

1. What are the five ordinary reasons research fails to change a decision, and what does a business do about each of them?
2. What is an evidence grid, what does each of its cells hold, and what does a business do when two methods disagree?
3. What are the three tests of an insight statement, and why is an insight statement not a recommendation?
4. What are the six parts of the decision brief, in order, and what is the verification criterion of the page?
5. How does the case's decision rule resolve, condition by condition, and what does the rule say to do?

Answers. 1. The question was not tied to a decision, the rule was not written first, the evidence arrived too late, the report was too long, and nobody had the authority to decide on it; the remedies are to write the decision and its rule before the fieldwork, to respect the gates that put the evidence before the decision date, to keep the brief to one page, and to name the decider before the study begins. 2. An evidence grid has one row per condition of the decision rule and one column per method, and each cell holds the finding, its source and its strength, which is this book's word for how many independent methods agree; when two methods disagree, the disagreement is reported and not averaged, and the cause of the disagreement is named. 3. The three tests ask whether the statement names its evidence, whether it bears on the decision and whether it can be wrong; it is not a recommendation because it says what was found and for whom, while a recommendation says what to do and belongs to the person who decides. 4. The six parts are the decision and the rule, the evidence with its sources and limits, the options, the recommendation, what would change the recommendation and the next step; the verification criterion is that a person can read the page in five minutes and take the decision from it alone. 5. The first condition passes, because the third territory's adjusted count of 414 is at least 400; the second fails, because the regional integrator won 2 of the 4 public maintenance tenders seen there, which is exactly half and not fewer than half; the rule requires both conditions, so it says to deepen the two territories, first among owner-run buildings, which are 41 of the 92 respondents.

Chapter References

American Association for Public Opinion Research. (2021). AAPOR code of professional ethics and practices: Disclosure standards (Revised April 2021). Retrieved October 10, 2026, from https://aapor.org/standards-and-ethics/disclosure-standards/

Braun, V., & Clarke, V. (2006). Using thematic analysis in psychology. Qualitative Research in Psychology, 3(2), 77–101. https://doi.org/10.1191/1478088706qp063oa

Chapman, C. N., & Milham, R. P. (2006). The personas' new clothes: Methodological and practical arguments against a popular method. Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 50(5), 634–636. https://doi.org/10.1177/154193120605000503

Christensen, C. M., Hall, T., Dillon, K., & Duncan, D. S. (2016, September). Know your customers' "jobs to be done." Harvard Business Review. https://hbr.org/2016/09/know-your-customers-jobs-to-be-done

Griffin, A., & Hauser, J. R. (1993). The voice of the customer. Marketing Science, 12(1), 1–27. https://doi.org/10.1287/mksc.12.1.1

International Chamber of Commerce & ESOMAR. (2025). ICC/ESOMAR international code on market, opinion and social research and data analytics. ESOMAR. Retrieved October 10, 2026, from https://community.esomar.org/uploads/public/knowledge-and-standards/codes-and-guidelines/ICCESOMAR-International-Code_English.pdf

La Machine. (2026). Documentation — route reference inventory. Internal product documentation, read October 10, 2026, together with the code of the pages this book describes: business/marketing/forms/responses and business/marketing/segments.

Personal Information Protection and Electronic Documents Act, S.C. 2000, c. 5, Schedule 1, clauses 4.2 to 4.5 (Canada). Justice Laws Website, current to September 21, 2026 (last amended March 4, 2025) — read October 10, 2026. https://laws-lois.justice.gc.ca/eng/acts/P-8.6/page-7.html

Salminen, J., Guan, K., Jung, S.-G., & Jansen, B. J. (2021). A survey of 15 years of data-driven persona development. International Journal of Human–Computer Interaction, 37(18), 1685–1708. https://doi.org/10.1080/10447318.2021.1908670

Chapter 10 — Typical Tasks, Cadence and Workflows

A study is a project with an end, but a watch is a routine that has no end, because competitors publish after the study is closed, public sources change on their own dates, and a survey is worth most when it can be repeated the same way — so this chapter puts the decisions of the earlier chapters on a calendar.

What You'll Be Able to Do

• Describe the four rhythms of a lasting research routine — a weekly monitoring round, a monthly signal review, a quarterly market brief and an annual study — and say what each one keeps current.
• Run a weekly monitoring round of thirty minutes that reads only what competitors publish, logs each change as a dated signal and keeps the reading apart from the fact.
• Chair a monthly signal review that reads the month's log against the thresholds of the decision rule and writes down which signals matter and which call for an alert.
• Write a one-page quarterly market brief that keeps findings, interpretations and recommendations apart and that credits no contract or sale to the watch.
• Repeat a survey with the same frame, the same questions and the same mode, and print each wave's own margin rather than reading a difference between two overlapping intervals as a change.
• Set the routine in the format of a task table, with a name, a date, an input, an output, a verification criterion and a role on every line, and name a backup for the person who runs the round.

Why It Matters

A market study is the most expensive thing this book asks a business to do, and it is also the thing that ages fastest. The desk study of Chapter 3 reads a property assessment roll that a municipality updates every year, an enterprise register that changes whenever a business is registered or struck, and a public tender system that publishes a call the day it is issued. The interviews of Chapter 4 describe a market at a moment. The survey of Chapter 5 measures awareness, provider arrangements and switching intentions in a frame that was drawn on a particular day. The digital signals of Chapter 6 are an index and a set of reviews read on a date. Every one of those sources moves on its own schedule, and none of them sends a notice when it does. A business that closes its study and files it has a photograph of a market that was true when the shutter opened.

The remedy is not a second study. It is a routine small enough to survive an ordinary week, and the routine is deliberately modest: thirty minutes a week to read the sources of the watch and log what changed, sixty minutes a month to read the log and decide which signals matter, one page a quarter to say what changed and what to do about it, and one repeat a year of the survey that the business already knows how to run. Nordec Entretien inc., the fictional business followed through this book, holds that routine at about nine and a half hours a quarter for the weekly rounds and the monthly reviews, and the figure is posed by this book rather than measured anywhere. The point of the routine is not that it produces a great deal of new information. The point is that it produces the information at the moment it becomes true, in a form that a decision can use, instead of in a burst once a year when the business finally has time.

There is a second reason the routine matters. The earlier chapters of this book each decided something: Chapter 2 wrote the question and its decision rule, Chapters 3 to 6 chose the methods of the study, Chapter 7 drew the legal line of the watch, Chapter 8 counted the market and gave a forecast, and Chapter 9 turned the whole of it into a decision brief. None of those decisions says when the work is done again. A decision rule is only useful if somebody reads the signals against it while there is still time to act. This chapter decides nothing that the earlier chapters decided; it puts their decisions on a calendar, with a name, a date, an input, an output and a way to check that each task is done.

In This Chapter

Section 10.1 explains why research is a routine rather than a project, sets out the four rhythms and places the routine of the case in a table. Section 10.2 describes the weekly monitoring round, what it reads and what it never does. Section 10.3 covers the monthly signal review, its five steps and the thresholds it reads. Section 10.4 covers the quarterly market brief and the one page it produces. Section 10.5 covers the annual study and the conditions under which a repeat can be compared with the first wave. Section 10.6 sets the whole routine in the format of a task table, in two tables of five and six columns. Section 10.7 divides the same tasks across a solo business, a small firm and an established business. Section 10.8 separates what can be automated from what stays with a person who reads the signal and decides. Section 10.9 names what breaks a research routine and what to do when a task is missed.

10.1 Research as a Routine, Not a Project

A study has a beginning and an end. It is designed, fielded, analyzed and written up, and then it is closed, and the business that ran it has an answer to the question it asked. The trouble is that the answer ages. A competitor opens a new establishment, a public tender system publishes a call that was not there last month, a statistics agency revises a count, a property assessment roll is updated for a new municipal year, and a survey that measured awareness in one month measures something slightly different twelve months later. None of these changes announces itself, and none of them waits for the next study. A business that treats research as a project therefore learns about its market in bursts, once a year or once every two years, and spends the time between the bursts working from a picture that is quietly going out of date.

The alternative is to treat research as two things at once. The study is a project: it has a question, a design, a budget, a close and a brief, and Chapters 2 to 9 of this book are about running it well. The watch is a routine: it has no close, it runs all year, and its job is to keep the study's picture current enough that a decision taken in month nine is not taken on month-one evidence. The two are not rivals. The routine is what tells a business that the moment has come to run the next study, and the study is what resets the routine's baseline. A business that runs only studies is always slightly behind its market. A business that runs only a routine never gets the depth that a survey, a set of interviews or a count by two paths can give.

The routine has four rhythms, and each one exists because the others cannot do its work. The weekly rhythm is a monitoring round: a short look at the sources of the watch, with every change logged as a signal carrying its date, its competitor, its kind and its source. The monthly rhythm is a signal review: Sophie Nadeau, the owner, and the administrator read the month's log against the thresholds of the decision rule and decide which signals matter and which call for an alert. The quarterly rhythm is a market brief: one page that says what changed, what it means and what to do. The annual rhythm is a study: the survey repeated with the same frame and the same questions, so that the two waves can be compared at all. Each rhythm hands its output to the next, and the annual study resets the baseline that the weekly round then watches.

The cadences of this chapter are the book's own method, and they are posed by this book rather than drawn from any source. No source read for this book measures how often a business should read its sources, how long a signal takes to read, how many signals a log should hold or what share of businesses repeat a study. The four rhythms are a reasoned proposal, offered because a business needs some rhythm and because these four are small enough to survive a busy week. A reader who adopts a different rhythm has not departed from a standard. What the book insists on is that some rhythm exists, that each task in it has a name and a date, and that the person who reads a signal is a person and not a reminder.

The routine also has a place in the wider calendar of the business. Book A3 (Strategy and Competitive Positioning) holds the annual strategy day, and the quarterly brief and the annual study are two of its inputs. Book N2 (Piloting Tasks, Executions and Projects) holds the ten questions of the task format that the task table of this chapter uses, and Book J2 (Processes, SOPs and the Definition of Done) holds the definition of done that a verification criterion applies. Neither book is redone here.

Figure 10.1 sets the research calendar in four rows, week, month, quarter and year, with what each produces and who holds it in the case.

Figure 10.1 — The research calendar. Every week a monitoring round of 30 minutes, held by the administrator; every month a signal review of 60 minutes, held by the owner and the administrator; every quarter a one-page market brief, held by the owner; every year the survey repeated with the same frame and questions, held by the owner with help.

The table below sets four lines of the routine beside who holds each and what it produces. It is posed by this book and is never a norm: the hours are the case's own, and no hour is posed for the quarterly brief or for the annual study.

Table 10.1 — The routine of the case, by rhythm

The weekly rounds and the monthly reviews of the case come to about nine and a half hours a quarter, a figure posed by this book and derived later in this chapter. The quarterly brief and the annual study carry no posed hour, because the model does not give one and this book prints none. What the table is meant to show is not the size of the routine but its shape: four rhythms, four outputs, and a named role on every line.

DEFINITION Round, review, brief, annual study and routine, as this book uses the words

These are this book's working words, and no source read defines them in this sense. A round is the weekly reading of the sources of the watch, in which each source is opened, each change is logged as a signal with its date and its source, and the reading is kept apart from the fact. A review is the monthly reading of the log, in which the owner and the administrator decide which signals matter and which call for an alert. A brief is the one-page quarterly statement of what changed, what it means and what to do, written so that a decision can be taken from the page alone. The annual study is the yearly repeat of the survey with the same frame and the same questions, so that the two waves can be compared. The routine is the four of them on a calendar, with a name, a date, an input, an output and a verification criterion on every line. The cadences are the book's own method, posed by this book and never a standard.

10.2 The Weekly Monitoring Round

The weekly round is the smallest task in the routine and the one that keeps everything else honest. It takes thirty minutes, it is held by the administrator, and its whole purpose is to read the sources of the watch and to write down what changed. The sources are the ones Chapter 7 identified as the public sources of a competitor watch: the competitors' own published pages, the public tender system, the enterprise register, the job postings a competitor publishes, and the public reviews a competitor's customers leave. The round does not interpret what it finds. It records it. A change on a competitor's public price page is logged as a change with its date and its source, and the question of what the change means belongs to the monthly review.

The discipline of the round is that the reading and the fact stay apart. A log entry that says "Competitor B raised its price" is a reading, and a log entry that says "Competitor B's public offer page showed a different figure on this date, from this page" is a fact. The distinction matters because the log is read months later, by people who did not do the round, and a log full of readings cannot be re-read. A fact can be checked against the page it came from. A reading cannot, and a reading that turns out to be wrong contaminates every decision built on it. The round therefore writes facts, and the review writes readings, and the brief writes recommendations, and the three are never mixed in the same line.

What the round never does is as important as what it does. It never calls a competitor, never poses as a customer, a candidate or a buyer, and never exchanges a price, a bid, a territory or a plan with a competitor. It reads only what competitors publish, which is the line that Chapter 7 drew and that this chapter keeps. The reason is not only ethical but legal, as the Québec and Canada note below sets out: the Competition Act, s. 45(1) and (2), is called to Book H4 (Consumer Protection, Advertising Law, Competition and Language Requirements), which holds the rule and the Bureau's guidelines on competitor collaboration, and a business that wants to know whether a given step crosses the line asks a lawyer rather than this book.

The round is also where the case's watch lives in the application. The weekly round of La Machine is a recurring alert and a dated task, and nothing collects a signal by itself: the alert rings, a person opens the sources, a person writes the entries, and the task is closed when the criterion is met (La Machine, 2026). That division is worth stating plainly, because a business that expects the software to watch the market will not watch it. The application reminds; the person reads. The reminder is useful precisely because it is dull: it arrives on the same day every week, it does not depend on anybody remembering, and it stops the round from being the task that is always postponed because the week is busy.

The case holds the round at thirty minutes a week, a figure posed by this book, and the round is the first line of Table 10.1. Thirty minutes is enough for the sources of a watch of three competitors, provided the sources are listed in advance and the log is open beside them, and the figure is posed by this book. The round is not a research session and it is not a reading of the market; it is a sweep, and a sweep that takes longer than half an hour is a sweep that will not happen next week. The verification criterion is the one that makes the round checkable: every source on the list has been read, and every change has been logged on the day it was seen.

QUÉBEC & CANADA NOTE A weekly watch reads only what competitors publish, and the Competition Act marks the line it must not cross

In Canada, the Competition Act, R.S.C. 1985, c. C-34, s. 45(1) and (2), current to September 21, 2026, makes it an offense for a person, with a competitor in respect of a product, to conspire, agree or arrange to fix, maintain, increase or control the price of its supply, to allocate sales, territories, customers or markets, or to restrict its production or supply, and the fine is left to the court; no amount of a penalty is printed here. A routine that reads published pages, logs what it finds and talks to no competitor stays on the right side of that line as this book describes it, and whether a given step does is a question for Book H4 (Consumer Protection, Advertising Law, Competition and Language Requirements) and for a lawyer. This book explains the text and recommends no vendor, no tool and no practice to any reader. The rule is called, never rewritten: the Bureau's guidelines on competitor collaboration and the mechanics of the section belong to Book H4, and the sentence of this note is a pointer rather than a substitute for advice.

10.3 The Monthly Signal Review

The monthly review is where the log becomes useful. Once a month, the owner and the administrator sit down for sixty minutes with the month's log, and the review does three things: it reads each signal with its date and its source, it compares the signals that matter against the thresholds written into the decision rule, and it writes down what it decided. The review is the meeting at which the routine produces a judgment, and the judgment is deliberately narrow. The review does not decide what the business should do about a competitor's new offer. It decides whether the signal calls for an alert, a note on a competitor profile, or nothing at all.

The first thing the review reads is the log itself. Each entry carries a date, a competitor, a kind and a source, and the review reads them in order, which is why the round writes facts rather than readings. In the case, a month of entries is short: the watch produced eleven signals in ninety days, a figure posed by this book, which is a little under one a week across three competitors. A short log does not show that nothing is happening; it shows that the sources are being read, and a month with no change on a competitor's public pages is itself a dated reading.

The second thing the review reads is the thresholds. Chapter 2 wrote a decision rule for the case, and the rule contains conditions: open a third territory only if its adjusted count reaches a certain level and if the regional integrator won fewer than half of the public maintenance tenders seen there. The review reads the month's signals against those conditions, and a signal that meets a condition is flagged. A public tender in the third territory that the regional integrator won is a signal that meets a threshold, and it is flagged. A change to a competitor's public page that has no bearing on any condition in the rule is not flagged, and it is archived rather than chased. The distinction is what keeps the routine from becoming a hobby.

The third thing the review writes is the reading of the month. One page, dated, holding the flagged signals, the reading of each one and the decision flag: an alert, a note on a profile, or nothing. The reading is written apart from the fact, in the same way the brief of Chapter 9 keeps findings, interpretations and recommendations apart. A line that says "the regional integrator won two of four tenders seen in the third territory" is a finding. A line that says "the regional integrator is defending the third territory" is an interpretation. A line that says "the third territory should be watched more closely" is a recommendation. The review writes all three, in that order, and never merges them.

The review also decides which competitor profiles to update. Chapter 7 keeps a profile of each competitor that matters, and a profile is a living document rather than a one-time exercise. A signal that changes what the business knows about a competitor's offer, its hiring, its footprint or its public pricing is a signal that belongs on the profile, with its date and its source. The update is a small task, and it is placed in the routine rather than left to memory, because a profile that is updated only when somebody thinks of it is a profile that is out of date within a quarter.

The case holds the review at sixty minutes a month, a figure posed by this book, with the owner and the administrator in the room. Sixty minutes is enough for a month of a three-competitor watch, provided the log is current and the thresholds are written down where the review can see them. The verification criterion is that every flagged signal has a reading and a decision flag, and the review is not closed until that is true. The next review goes on the calendar before the meeting ends, because a review that is scheduled when it becomes urgent is a review that happens after the decision it was meant to inform.

METHOD The monthly signal review in five steps

First, open the month's log and read each signal with its date and its source. Second, read the thresholds of the decision rule and mark each signal that meets one. Third, write the reading of each marked signal apart from the fact, in the order finding, interpretation, recommendation. Fourth, decide for each marked signal whether it calls for an alert, a note on a competitor profile or nothing, and write the decision down. Fifth, close the review with its verification criterion — every flagged signal has a reading and a decision flag — and put the next review on the calendar. This is the book's reasoning and not a tested procedure, and the sixty minutes of the case are posed by this book. The steps are deliberately few, because a review with ten steps is a review that will be shortened in a busy month, and a shortened review is one that reads the log and decides nothing.

10.4 The Quarterly Market Brief

Once a quarter, the owner writes one page. The page says what changed, what it means and what to do, and it is built from the three monthly readings of the quarter rather than from the log. The brief is the routine's contribution to the business's decisions, and its form is the form of the decision brief of Chapter 9: findings first, interpretations second, recommendations third, and the three kept visibly apart. The difference between the two documents is scope. The decision brief of Chapter 9 answers the question the study was built around. The quarterly brief answers a smaller and more frequent question: what has changed since the last brief, and does anything in it call for action now.

The brief's first section is what changed. It reports the quarter's signals with their dates, in counts rather than in adjectives: how many signals were logged, how many were flagged, how many tenders were seen and how many were won by which competitor, how many changes appeared on public offer pages, how many job postings were published. The counts are the log's own, and the brief prints them with the dates they carry. It also says what was not seen, which is a discipline that Chapter 9 established and that is easy to lose in a shorter document. A brief that says "no public tender was published in the third territory this quarter" is telling the reader something, and it is telling the reader something different from a brief that says nothing about tenders at all.

The brief's second section is what it means. This is where the owner writes the interpretation, and the interpretation is written as an interpretation: "the pattern of the last two quarters suggests that the regional integrator is concentrating on the eastern municipalities" is an interpretation, and it is labeled as one. The brief does not present an interpretation as a finding, and it does not present a finding as a recommendation. The reason is not pedantry. A brief that merges the three cannot be re-read a year later, because the reader cannot tell which part was observed and which part was inferred, and the inferred part is exactly the part that may have been wrong.

The brief's third section is what to do. The recommendations are short and each one names a person and a date: deepen the watch on one competitor, ask the fieldwork service for a quote on a repeat survey, add a source to the list, drop a source that has produced nothing in two quarters. The brief does not credit any contract, sale or signing to the watch, and it never says that a decision was caused by a signal. The decision rule decides, and the routine informs the rule. A brief that claims credit for an outcome is a brief that has confused a watch with a cause.

The brief is also where the source list is reviewed. Once a quarter, the administrator confirms that each source on the list still exists and is still read on its public site, and the list is dated. A source that has moved, closed or stopped publishing is removed or replaced, and the change is noted. This is a small task with a large effect, because a source list that is never reviewed slowly fills with pages that no longer say anything, and a round that reads dead pages is a round that produces no signals and that the administrator will eventually stop doing.

The case writes four quarterly briefs by day 365, a figure posed by this book, and no hour is posed for each brief. The brief is one page, and the test of the page is that the owner can decide from it alone, without opening the log. If the page cannot carry the decision, the page is too long or too vague, and either fault is fixed by cutting rather than by adding.

TOOLS OF THE TRADE What any brief template must offer, whatever it is called

The functions are few and none of them is a specialist product. A dated page with three visible sections, one for findings, one for interpretations and one for recommendations, so that the three cannot be merged by accident. A place for the counts of the quarter, each with the date it carries, and a line for what was not seen. A field for the decision the brief calls for, with a person and a date beside it. A link to the log the brief was built from, so that a reader can check a finding against its source. A version history that keeps the previous quarter's brief beside the current one, because the comparison between two briefs is often more useful than either. A reminder that puts the next brief on the calendar. No brand name belongs on this list and no price; a business that has these six functions has a brief template, and a business that has a document with three headings has most of them already.

10.5 The Annual Study

Once a year, the business repeats its survey. The repeat is the routine's largest task and the one that resets the baseline the rest of the routine watches. The case repeats its survey on day 365, with the same frame and the same questions, and no result of the second wave is printed here, because the second wave is posed by this book and its numbers are not given. What the chapter can say is what makes a repeat worth doing and what makes it worthless.

A repeat is comparable only if three things are unchanged: the frame, the questions and the mode. The frame is the list of units the sample is drawn from, and a frame that has been redrawn is a different population. The questions are the wording and the order, and a question that has been reworded is a different question. The mode is the way the answers are collected, and a survey that moves from one mode to another is a survey whose differences may be mode effects rather than changes in the market. A business that changes any of the three has run a second study, not a repeat, and the two cannot be compared.

Even with all three unchanged, a comparison has to respect the margins. Each wave of a probability sample prints its own margin of error, with its confidence level and its population, and two intervals that overlap show no change. The case's first awareness measure was nineteen of sixty (Book B3, section 9.7), and the survey of Chapter 5 gave a prompted awareness of thirty-five of ninety-two; the two intervals overlap, and the rise is not shown. A business that reads the two point estimates as a trend has read a change that the data do not support, and a year is long enough for a reader to want a story.

The annual study is also where the business decides whether to repeat the interviews. Interviews are repeated only when a question calls for it, and the number of interviews is the study's own. The abstract of one study reports that in sixty in-depth interviews with women in two West African countries, saturation occurred within the first twelve interviews, with basic elements of metathemes present as early as six (Guest et al., 2006, at abstract only). That is one study of one population, and it supports no rule of twelve for any other study. The case's own interviews reached their last new code at interview nine, out of fourteen conversations and thirty codes, and that is the case's own count, posed by this book. A business that repeats interviews should expect to find its own saturation point rather than borrow one.

The annual study is also the moment at which the business checks whether its question is still the right question. A survey that has been repeated three times with the same frame and the same questions is a survey that measures the same thing three times, and if the business's decision has moved on, the repeat measures something nobody needs. The check is short: does the decision rule still name the decision the business faces, and does the survey still inform it? If the answer is no, the annual rhythm is the moment to redesign rather than to repeat, and the redesign is a new study with a new baseline.

KEY CONCEPT A survey repeated with the same frame and questions can show a change, and a survey changed cannot

The central idea of this section is that a comparison needs the same frame, the same questions and the same mode, and that each wave of a probability sample prints its own margin. Two intervals that overlap show no change, as the case's first two awareness measures do: nineteen of sixty, about thirty-one point seven percent with a margin of about plus or minus eleven point eight points, and thirty-five of ninety-two, about thirty-eight percent with a margin of about plus or minus nine point five points, overlap, and the rise is not shown. A number of interviews that was enough in one study is that study's own and is not a rule: the abstract of one study reports saturation within the first twelve interviews in sixty in-depth interviews with women in two West African countries (Guest et al., 2006, at abstract only), and that is one study of one population. No source read for this book gives a cadence for repeating a study, and the annual rhythm of this chapter is posed by this book.

10.6 The Research Task Table

The routine is only as good as the tasks it is made of, and a task is only as good as its definition. This section sets the nine tasks of the routine in the format of the application's task table, which Book N2 (Piloting Tasks, Executions and Projects) describes through its ten questions: what the task is, what it is for, what triggers it, how often it runs, how much of it is automated, what goes in, what comes out, how completion is verified, who does it and what tools it uses. The ten questions are the reason the routine is checkable. A task with a trigger and a verification criterion is a task that happens and that can be shown to have happened. A task without them is an intention.

The table is split in two because a table of ten columns does not read aloud, and this book is also published as audio. The first table gives the first five fields of each task: its name, its objective, its trigger, its frequency and its level of automation. The second gives the remaining five: its input, its output, its verification criterion, its role and its tools. Both tables open with the Task column, so that a listener can follow a single task across the two.

The table below sets nine tasks beside their objective, their trigger, their frequency and their level of automation. The routine is posed by this book and is never a norm, and no task in it is run by software on its own: the automation level says what a reminder does and where a person takes over.

Table 10.2 — The tasks of the routine: objective, trigger, frequency and automation

The table below gives, for the same nine tasks, the input, the output, the criterion that proves the task done, the role and the tool by function. Every criterion is observable, and every tool is described by what it does rather than by a name.

Table 10.3 — The tasks of the routine: input, output, verification and role

Two features of the two tables are deliberate. The first is that every task has a role, and no task has two roles that could both assume the other is doing it. The second is that every verification criterion is something a person can check by looking, which is what Book J2 (Processes, SOPs and the Definition of Done) means by a definition of done. A criterion that cannot be checked is a criterion that will be satisfied by a feeling, and a routine built on feelings is a routine that stops without anybody noticing.

WATCH OUT A task without a verification criterion, a calendar without a name and a reminder without a person are three ways a routine stops

Three traps stop a routine, and each has a question that catches it. A task that nobody can check: what would show that it is done? If the answer is a feeling, the task has no criterion, and it will be closed on a busy week without being done. A task that belongs to everyone: whose name is on it, and who is the backup? A task with two owners has none, and a task with one owner and no backup stops the week that owner is away. A reminder that nobody reads: who reads it, and when? A recurring alert that arrives in a mailbox nobody opens is a reminder that has been mistaken for a system. The three traps are the book's reasoning and are illustrative; no source read gives the frequency of any of them in any business. The task table of this section gives each task a role and a criterion, and the criterion is the answer to the first question, the role is the answer to the second, and the trigger is the answer to the third.

THE RULE Every research task has a name, a date and a way to check that it is done

This is the book's own rule for this chapter, drawn from its reasoning and from the definition of done that Book J2 (Processes, SOPs and the Definition of Done) holds: a task is done when its verification criterion is met, and a task that nobody can check, or that belongs to no one, is not on the calendar. The rule is the book's own; it is neither a standard nor a legal duty, and no source read states it. The example is the weekly round, whose criterion is that every source has been read and every change logged on the day it was seen. A task that cannot be given a criterion is not ready to go on the calendar, and a task that has been postponed for a month may be too large, in which case the remedy is to split it rather than to lower the criterion.

10.7 Sharing Research Work in Three Sizes of Business

The routine does not change with the size of the business. The four rhythms exist in a solo business, in a small firm and in an established company, and the tasks inside each rhythm are the same tasks. What changes is how many people touch them and how much of the work is prepared by somebody other than the person who decides. This section sets the four rhythms beside who does the work in three sizes, and the first and third columns are the book's suggestion rather than anything in the case's model. No hour is promised to a reader, and no size of business is told that it needs a particular rhythm.

In a solo business, the routine is compressed rather than abandoned. The owner does the round, the review, the brief and the study, and the compression is in the frequency rather than in the content. A solo owner who tries to hold a weekly round and a monthly review may drop one of them, and a routine that has been dropped leaves a stale log, so that the next reader cannot tell where it stopped. The book's suggestion for a solo business is one weekly sitting that covers the round and the review together, and a quarterly brief written in the same sitting as the source-list review. The tasks are the same; they are simply held on fewer days.

In a small firm like the case, the work divides. The administrator holds the weekly round, keeps the log, updates the competitor profiles and files the repository, and the owner joins the monthly review, writes the quarterly brief and decides. The fieldwork service runs the survey, and the answers stay in the research file. The division works because the person who prepares the meeting is the person who knows what the log says before anybody reads it, and because the person who decides is not the person who did the reading. A firm of this size also needs a named backup for the round, because a small firm has no slack and an absence is felt in the first week.

In an established business, one person runs the routine for the team. The tasks are the same nine, and the risk is different: in a larger business the routine can become a reporting obligation rather than a decision-making practice, and the monthly review can turn into a status report from each department. The remedy is the one this book repeats, which is that the review starts from a prepared log and ends with a decision flag on every flagged signal, and that the person who prepares the log is not the person who chairs the review. A larger business also has more sources to read, and the source list is where the growth shows: a list that has doubled without a review is a list that has doubled the round's length without doubling its value.

The table below sets the four rhythms beside who does the work in a solo business, in the case's firm and in an established business. The first and third columns are the book's suggestion and are not part of the case, and the case's firm is Nordec Entretien inc., whose routine is posed by this book.

Table 10.4 — The routine in three sizes of business

Four rules hold in every size, and they are the rules that make the routine a research routine rather than a reading habit. Only public sources are read, which is the line of Chapter 7. The log is dated, so that a signal can be placed in time. No brief is sent and no decision is taken without a person reading it first, which is the rule this book applies to every document the routine produces. And a backup is named for the person who runs the round, because the routine's weakest point is the week its holder is away.

WHO DOES THIS The administrator runs the round, the owner reads and decides, and the fieldwork service repeats the survey

The roles are the case's, and the division is posed by this book. The administrator holds the weekly round of thirty minutes, keeps the log, updates the competitor profiles, files the repository and checks that respondents stay apart from prospects. Sophie Nadeau, the owner, joins the monthly review of sixty minutes, writes the quarterly brief and decides. The fieldwork service repeats the survey with the same frame and the same questions. In a solo business the owner does all of it in one weekly sitting, which is the book's suggestion and not part of the case. In a larger business a person runs the routine for the team, which is also the book's suggestion and not part of the case. No hour is promised to a reader, and no role is told that the work takes a particular time. The one rule that holds in every size is that the person who reads the signal and the person who decides are both people, and that the routine never decides anything by itself.

10.8 Automating the Routine Without Losing Judgment

A routine has parts that a machine does well and parts that a machine does badly, and the value of separating them is that the first kind can be made reliable and the second kind can be protected. What a machine does well is remind, list, template and repeat. A recurring alert that rings on the same day every week is a machine's work, and it is better at it than a person, because it does not forget and does not decide that this week can be skipped. A task list with a date, a person and a repeat rule is a machine's work. A template with three sections, one for findings, one for interpretations and one for recommendations, is a machine's work. A folder that keeps each study with its dates and its retention date is a machine's work. None of these requires judgment, and all of them fail when they are left to memory.

What stays with a person is the choice of sources, the reading of a signal, the threshold and the decision. The choice of sources is a judgment about what is worth watching, and it changes as the business's question changes. The reading of a signal is a judgment about what a change means, and it is the judgment the monthly review exists to make. The threshold is a judgment written in advance, in the decision rule of Chapter 2, and it is read rather than computed. The decision is the owner's, and it is taken with the brief in hand and the log behind it. A routine that automates any of these four has automated the part that made the routine worth running.

An assistant has a place in the routine, and the place is narrow. The administrator asks an AI assistant for a first summary of a long public page, then checks the summary against the page, and the owner approves it before it enters the log. The assistant produces a draft; a person verifies it; the owner approves it. No signal is read, no brief is sent and no decision is taken without a person reading it first, and the same rule applies to a first draft of a brief, a first pass at a set of notes or a first list of candidate sources. Book K4 (Artificial Intelligence for Business) treats the assistants themselves, and this chapter uses them only as a drafting aid inside a routine that a person runs.

Automated collection of signals lies outside this book, and Book K7 (Web Scraping, Data Enrichment and Market Signals) treats collection methods, enrichment and legal boundaries in its chapters 3, 4 and 5. This chapter assumes that a person opens the sources, and it says so because the assumption is what makes the rest of the routine honest: a log whose entries were collected by a process nobody read is a log whose entries cannot be trusted, and a signal that nobody read is not a signal.

The application copy read for this book carries the routine's reminders and none of its judgment, as the box below and the closing section set out.

IN LA MACHINE Alerts and tasks remind by hand, links keep the sources, and no page watches the market

Productivity → Alerts (productivity/alerts) rings at a chosen time or on an event of the product, and Productivity → Tasks (productivity/tasks) holds the task and its verification criterion, so the weekly round, the monthly review and the quarterly brief are recurring reminders and dated tasks, and nothing collects a signal by itself (La Machine, 2026). Productivity → Links (/liens) keeps team bookmarks of the public sources, dated by the team, which is where the statistics agency's tables, the enterprise register, the property assessment roll and the public tender system can sit. The home page (/dashboard) shows the business's own activity and no market indicator. La Machine is described and never sold, and it is one worked example of a business platform rather than the tool of the case's study. The case holds its weekly watch in it only as reminders. Check it in your installation, because what an installation shows is what that installation shows.

10.9 What Breaks a Research Routine

A routine breaks in a small number of ways, and each one is worth naming before it happens, because the remedy is easy in a quiet week and impossible in the week the break arrives. The first break is that the administrator goes away and no backup is named. The round stops, the log goes stale, and the monthly review reads a month of nothing. The remedy is a named backup who has run the round once, on an ordinary week, with the real source list in front of them. The second break is that the round is skipped in a busy week. One skipped week is harmless; three in a row is a habit, and the habit is what ends the routine. The remedy is to keep the round and shorten it: a ten-minute sweep that reads the two most important sources is better than a thirty-minute sweep that does not happen.

The third break is that the monthly review is postponed until a decision has already been taken. A review that happens after the decision is a review that records rather than informs, and it teaches the business that the routine is a formality. The remedy is to hold the review on a fixed day and to keep it short, because a review that is scheduled when it becomes urgent is a review that happens too late. The fourth break is that a source disappears and nobody notices. A page that has moved, a register that has changed its address, a tender system that has reorganized its listings: each of these leaves a source list that looks complete and produces nothing. The remedy is the quarterly review of the source list, with a date read against every line.

The fifth break is that the log is not dated. An undated log cannot be read months later, because a signal without a date cannot be placed against a threshold or against another signal. The remedy is the round's verification criterion, which requires the date on every entry. The sixth break is that the owner becomes the only reader of the log. A log read by one person is a log that dies with that person's attention, and the routine's whole design assumes two readers: the administrator who writes and the owner who decides. The remedy is the monthly review, which puts a second reader in the room. The seventh break is that the brief grows until nobody reads it. A brief of five pages is a brief that is opened once and filed, and the remedy is the one-page test: if the owner cannot decide from the page alone, the page is too long.

The eighth break is the one that costs the most and shows the least: the annual study changes its questions and loses the comparison. A survey whose wording has been improved between two waves is a survey that cannot be compared with itself, and the improvement is usually invisible until somebody tries to read the two waves side by side. The remedy is the annual study's verification criterion, which requires that the frame and the questions are unchanged and that both margins are printed. The case keeps the same frame and the same questions on day 365, a figure posed by this book, and no result of the second wave is printed here.

When a task is missed, the remedy is short and it is the same in every case. Do the task the next day rather than the next month, write down that it was missed and why, and find the cause rather than the fault. A missed round may point to a scheduling problem, a missed review to a review that was too long, and a missed brief to a brief that had grown past a page. The cause is fixed in the task table, not in a resolution to try harder, and the fix is written into the criterion so that the next holder of the task inherits it.

DO THE MATH The watch hours of the case in a quarter, in five steps

Illustrative figures; the case is fictional. Step 1 — The weekly round is 30 minutes, held by the administrator. Step 2 — Thirteen weeks in a quarter make thirteen rounds, which is six and a half hours. Step 3 — The monthly review is 60 minutes, three times in a quarter, which is three hours. Step 4 — Six and a half hours and three hours add to 9.5 hours a quarter, as the model states. Step 5 — The model poses no hour for the quarterly brief or for the annual study, and this book prints none. The sums are exact and no rounding is needed. The figures are posed by this book, they are hours and not a norm, and they credit no contract and no sale to the routine.

CASE IN POINT Nordec Entretien inc. runs a weekly round, a monthly review, a quarterly brief and an annual study

Nordec Entretien inc. is a fictional business created for this encyclopedia; its owner, its employees, its customers, its competitors, its figures and the events described are invented, and any resemblance to a real business is coincidental. The routine is a weekly round of thirty minutes held by the administrator, a monthly review of sixty minutes held by the owner with the administrator, a one-page quarterly brief written by the owner, and an annual study that repeats the survey with the same frame and the same questions. The weekly rounds and the monthly reviews come to about nine and a half hours a quarter. All of it is posed by this book, illustrative and not a forecast, and no result of the routine is printed here. The routine credits no contract and no sale to the watch, and the decision rule decides.

WHAT CHANGED On day 365 the second survey repeats the first with the same frame and questions

The dated change in the case is counted in days and is posed by this book. On day 365 the watch has been kept 49 weeks, 4 quarterly market briefs have been written and 38 signals are in the log, and the survey is repeated with the same frame and the same questions. Nothing is credited to the study, and no result of the second wave is printed. The application code of La Machine was read on October 10, 2026 (La Machine, 2026), and the dates of sources are never dates of the case, which is counted in days and never in calendar years or months. A change of this kind is the routine's own anniversary: the day the baseline is reset and the weekly round begins watching a picture that is current again.

Chapter 11 reads who helps with this work and when to get help: the owner as question-setter and decider, the person who runs research day to day, research firms and independent researchers, panels, recruiters and fieldwork services, analysts, and the legal and privacy advice a business may need.

Costly Mistakes

• Treating research as a project that ends when the brief is filed. A study is a photograph of a market on a date, and the sources it read keep moving; a business with no routine works from a picture that is quietly out of date until the next study.
• Writing the log as readings instead of facts. A log entry that says what a change means cannot be re-read months later, and a reading that turns out to be wrong contaminates every decision built on it.
• Letting the weekly round read anything other than what competitors publish. Calling a competitor, posing as a customer or exchanging a price, a bid or a plan crosses the line that the Competition Act, s. 45(1) and (2), marks, and Book H4 holds the rule.
• Giving a task no verification criterion and no name. A task that nobody can check is closed on a busy week without being done, and a task with two owners has none; the task table of this chapter gives each task a role and a criterion.
• Reading a difference between two survey waves as a change when the intervals overlap. Each wave of a probability sample prints its own margin, and two overlapping intervals show no change, as the case's first two awareness measures do.
• Changing the frame, the questions or the mode between two waves of a survey. A repeat is comparable only if all three are unchanged; a survey that has been improved is a second study, not a repeat, and the two cannot be read side by side.
• Letting the quarterly brief grow past a page. A brief of five pages is opened once and filed; the test is that the owner can decide from the page alone, and a page that fails the test is fixed by cutting rather than by adding.

Key Takeaways

• A study is a project with an end, but a watch is a routine that runs all year, and the routine is what keeps the study's picture current enough that a decision taken in month nine is not taken on month-one evidence.
• The routine has four rhythms: a weekly monitoring round of thirty minutes held by the administrator, a monthly signal review of sixty minutes held by the owner with the administrator, a one-page quarterly market brief written by the owner, and an annual study that repeats the survey with the same frame and questions.
• The weekly round reads only what competitors publish, logs each change as a dated signal with its source, and keeps the reading apart from the fact; the monthly review reads the log against the thresholds of the decision rule and writes a decision flag on every flagged signal.
• A survey repeated with the same frame, the same questions and the same mode can show a change, and a survey changed cannot; each wave of a probability sample prints its own margin, and two intervals that overlap show no change.
• Every task in the routine has a name, a date, an input, an output, a verification criterion and a role, and the routine is posed by this book rather than drawn from any source, because no source read measures a cadence of research work.
• The routine breaks in a small number of ways — no backup, a skipped week, a postponed review, a dead source, an undated log, a single reader, a brief that grows, a study that changes its questions — and each break has a remedy that is easy in a quiet week and impossible in the week it arrives.

Your Action Plan

1. Put the four rhythms on the calendar with a name on each. Place the weekly round, the monthly review, the quarterly brief and the annual study on a shared calendar, each with a time, a duration and a named holder, and name a backup for the round. Deliverable: dated entries for the four rhythms and a named backup. Verification: every entry has a holder and a backup, and the round's holder has run it once on an ordinary week.
2. Write the source list and the log before the first round. List the public sources of the watch with their addresses, and open a log with columns for the date, the competitor, the kind of change and the source. Deliverable: a dated source list and an open log. Verification: every source on the list is a page a competitor or a public body publishes, and the log has no column for a reading.
3. Run the first monthly review against the thresholds of the decision rule. Read the month's log, mark every signal that meets a threshold, write the reading of each marked signal apart from the fact, and decide for each whether it calls for an alert, a note on a profile or nothing. Deliverable: a dated reading of the month and a list of flagged signals. Verification: every flagged signal has a reading and a decision flag, and the next review is on the calendar.
4. Write the first quarterly brief on one page and test it. Build the page from the three monthly readings, with findings first, interpretations second and recommendations third, and a line for what was not seen. Deliverable: a one-page brief with three visible sections. Verification: the owner can decide from the page alone, without opening the log.
5. Set the annual study's criterion before the repeat. Confirm that the frame, the questions and the mode are unchanged from the first wave, and write the criterion into the task: both margins are printed and the two intervals are compared before any change is claimed. Deliverable: a dated task for the repeat with its criterion. Verification: the frame and the questions are identical to the first wave, and no change is reported from two overlapping intervals.

In La Machine

The routine's reminders can live in the application copy read for this book, and its judgment cannot (La Machine, 2026). A recurring alert at Productivity → Alerts (productivity/alerts) puts the weekly round on the calendar, a dated task at Productivity → Tasks (productivity/tasks) carries the criterion that shows it done, and Productivity → Links (/liens) holds the dated bookmarks of the public sources. The home page (/dashboard) shows the business's own activity and no market indicator, and the three competitors sit as company records with dated notes. Check it in your installation.

What the platform does not do belongs in the same breath, because a business that expects the routine to be built into the software will not build it. La Machine collects no signal, watches no competitor, has no approval and no deliverable milestone for research, and shows no market indicator. What the reader needs elsewhere is a short list: a list of public sources, a dated signal log, a page holding the thresholds of the decision rule, a one-page brief template, a named backup for the person who runs the round, and a person who reads the signal and decides. The thinking happens in the documents, the reading happens in the routine, and the platform carries the reminders.

Self-Check

1. Why is research a routine rather than a project, and what does each of the four rhythms keep current?
2. What does the weekly monitoring round read, what does it never do, and what is its verification criterion?
3. What are the five steps of the monthly signal review, and what does the review write down?
4. Under what conditions can two waves of a survey be compared, and what does an overlap between two intervals show?
5. Name four ways a research routine breaks and the remedy for each.

Answers. 1. A study is a project with an end, but the sources it read keep moving on their own dates, so a business with no routine works from a picture that is quietly out of date; the weekly round keeps the sources read and the signals logged, the monthly review keeps the signals read and the alerts set, the quarterly brief keeps the market picture current in one page, and the annual study resets the baseline by repeating the survey. 2. The round reads the public sources of the watch — the competitors' published pages, the public tender system, the enterprise register, job postings and public reviews — and it never calls a competitor, never poses as a customer, a candidate or a buyer, and never exchanges a price, a bid, a territory or a plan; its criterion is that every source has been read and every change logged on the day it was seen. 3. The five steps are to open the month's log and read each signal with its date and its source, to read the thresholds of the decision rule and mark each signal that meets one, to write the reading of each marked signal apart from the fact, to decide for each marked signal whether it calls for an alert, a note on a profile or nothing, and to close with the criterion that every flagged signal has a reading and a decision flag; the review writes a dated reading of the month and a list of flagged signals. 4. Two waves can be compared only if the frame, the questions and the mode are unchanged, and each wave of a probability sample prints its own margin; two intervals that overlap show no change, as the case's first two awareness measures do, and a business that reads the two point estimates as a trend has read a change the data do not support. 5. The administrator goes away with no backup, and the remedy is a named backup who has run the round once; the round is skipped in a busy week, and the remedy is to keep it and shorten it; the review is postponed until a decision is already taken, and the remedy is a fixed day and a short meeting; a source disappears and nobody notices, and the remedy is the quarterly review of the source list with a date read against every line.

Chapter References

Competition Act, R.S.C. 1985, c. C-34, s. 45(1) and (2) (Canada). Justice Laws Website, current to September 21, 2026 (last amended March 26, 2026) — read October 10, 2026. https://laws-lois.justice.gc.ca/eng/acts/C-34/

Guest, G., Bunce, A., & Johnson, L. (2006). How many interviews are enough? Field Methods, 18(1), 59–82. https://doi.org/10.1177/1525822X05279903

La Machine. (2026). Documentation — route reference inventory. Internal product documentation, read October 10, 2026, together with the code of the pages this book describes: productivity/alerts, productivity/tasks, /liens and /dashboard.

Chapter 11 — Expertise, Roles and When to Get Help

Research is a question set by the owner and a body of work done by several people, and this chapter says which skills the work needs, what the owner keeps, when to get help, how to read what a helper delivers and what stays inside the firm.

What You'll Be Able to Do

• Name the five skills a market study needs — setting the question, running the work day to day, talking to people, handling numbers and reading the rules — and say which of them a small business already has.
• Decide whether a task stays inside the firm or is bought, using three questions: does it need a skill the firm lacks, does it need independence, and does it need capacity.
• Keep the question, the decision rule and the decision with the owner, and delegate the desk study, the counts, the log and the fieldwork without delegating the decision.
• Describe a research firm, an independent researcher, a panel, a recruiter and a fieldwork service by what each does and by the form of its fee, never by an amount.
• Read a research proposal and a report against a fixed list of questions — the question, the method, the frame, the sample, the margin, the data and the deliverables — and write the questions to ask before signing.
• Recognize the work that stays inside the firm, the work that is bought, and the legal and privacy questions that belong to a lawyer.

Why It Matters

A market study is described as a method, and the method is the part this book has spent nine chapters setting out. But a method is performed by people, and the division of the work among them shapes the study. The question is written by somebody who has to live with the answer. The desk study is read by somebody who has an afternoon. The interviews are held by somebody the customer will talk to. The survey calls are made by somebody who is not the owner. The numbers are checked by somebody who can see an arithmetic mistake. The rules are read by somebody who is allowed to give an opinion on them. When those six somebodies are the same person, the study is slow and thin. When they are nobody in particular, the study does not happen at all. This chapter is about the division of that work.

Nordec Entretien inc., the fictional business followed through this book, is a small firm with a small team, and its study was run by four people and one outside service. Sophie Nadeau, the owner, wrote the decision and the rule and decided at the end. The administrator ran the desk study, the survey frame and the watch log. The lead technician and the owner held the interviews. The dispatcher checked the zones walked against the routes. A fieldwork service made the survey calls. An outside lawyer read the line. Every one of those roles and every hour attached to it is posed by this book and illustrative, and nothing in the study is credited to it. The point of the example is not the size of the team. The point is that a study of this kind has five or six distinct pieces of work, and that a business which names them in advance can decide, piece by piece, which ones it does itself and which ones it buys.

The reason the division matters is that the wrong division produces two opposite failures. In the first, the owner tries to do everything: writes the questionnaire, makes the calls, reads the statistics, interprets the law. The study takes three times as long as it should, the calls are made by the person the customer buys from, and the answers are shaped by the relationship. In the second, the owner buys everything: a firm is hired to design the study, run it and report, and the owner reads a document at the end that answers a question the owner never quite asked. The first failure is a study that is too close to the business. The second is a study that is too far from it. The remedy in both cases is the same, and it is the rule this chapter keeps returning to: the owner keeps the question, the decision rule and the decision, and buys only the work.

There is a third reason to think about roles before thinking about vendors. A business that knows what it is buying can read a proposal. A business that does not know what it is buying reads a proposal as a document, and a document can be long, well bound and beautifully written while answering nothing. The list of questions in section 11.8 exists because a proposal is read for its content and not for its length, and because a report is read against a fixed set of disclosures rather than against the impression it leaves. None of that requires the owner to become a methodologist. It requires the owner to know which pieces of work exist, who is doing each one, and what each helper is being paid for.

In This Chapter

Section 11.1 sets out the five skills a market study needs and the three questions that decide whether a task stays inside the firm or is bought. Section 11.2 describes the owner as question-setter and decider, and the four things the owner never delegates. Section 11.3 describes the person who runs the research day to day, what that person needs and what that person is not. Section 11.4 covers research firms and independent researchers, what they bring and how they are paid. Section 11.5 covers panels, recruiters and fieldwork services, and the difference between a panel and a sample. Section 11.6 covers analysts, statisticians and data specialists, and when a number needs one. Section 11.7 covers legal and privacy advice on research, and what to bring to a lawyer. Section 11.8 gives a method for reading a research proposal and its report. Section 11.9 sets out the work that stays inside the firm and the work that is bought.

11.1 The Skills Research Needs

A market study needs five skills, and they are worth naming separately because a business may have some of them and not others. The first is setting the question: tying the work to a decision that somebody has to take, and writing the rule that will decide it before any fieldwork begins. The second is running the work day to day: keeping the desk study, the frame, the log, the files and the task list in a state where the study moves without being pushed. The third is talking to people: holding an interview without leading it, asking about what a person did rather than what a person intends, and writing notes that another person can code. The fourth is handling numbers: drawing a sample, computing a size, reading a count against a plan, and knowing when a difference is a difference. The fifth is reading the rules: knowing which questions about consent, calls, competitors and copyright belong to a lawyer and which belong to the business.

A small business may already have more of these skills than it thinks. It knows its customers, because it serves them every week, and that knowledge is the raw material of a good interview guide. It reads a contract, because it signs them, and that habit is what makes a proposal readable. It counts, because it invoices, and counting is most of what a market count is. A small business may lack the fourth and fifth skills at depth: a person who can draw a probability sample and compute its margin, and a person who can say what a rule requires. Those two can be bought in small quantities, a few hours of a statistician or a lawyer, rather than a firm on retainer.

The decision to do a task inside or to buy it turns on three questions, and the questions are asked one task at a time rather than once for the whole study. The first question is whether the task needs a skill the firm lacks. Drawing a probability sample from a frame and computing its margin is a skill; if nobody in the firm has it, the task is bought or the design is changed to something the firm can do honestly. The second question is whether the task needs independence. An interview with a customer who buys from the owner is not the same interview when the owner is in the room, and a survey of a list the business holds is not the same survey when the business makes the calls. Independence is not a virtue of the helper; it is a property of the design, and it is bought when the design needs it. The third question is whether the task needs capacity. The 360 numbers the case draws make a task that is not hard and is large, and a business with four people cannot call them all without stopping the work that pays.

The three questions are the book's reasoning and not a finding. No source read for this chapter measures how often a small business does a research task itself or buys it, and no share of businesses that hire help is printed anywhere in this book. What the three questions give is a way to make the choice deliberately rather than by habit, and the habit they replace is the one in which a business buys the whole study because it cannot see the pieces, or does the whole study itself because it has not noticed which piece it is bad at.

A skill is a task done well, and a role is a name on a task. The distinction sounds like a definition and it is a working tool. A business that lists its skills will find that it has three of the five and lacks two. A business that lists its roles will find that one person is carrying four of them and that the fourth is the one that keeps slipping. The two lists are different, and the second is the one that gets work done, because a task with no name on it is a task that waits. The rest of this chapter is about putting names on the tasks of a market study, and about the moment when the right name is not somebody inside the firm.

DEFINITION Question-setter, research firm, panel, fieldwork service and fee form, as this book uses the words

These are this book's working words, and no source read for this chapter defines them in this sense. The question-setter is the person who ties the research to a decision and writes the rule that will decide it, before any fieldwork begins. A research firm is a business that designs and runs studies for clients, and an independent researcher is the same work done by one person. A panel is a list of people who agreed to answer surveys, and an opt-in panel is a non-probability source: its members chose to be there, so no margin of error is printed for it. A fieldwork service makes the calls or runs the platform for a study, on a script and a list that somebody else designed. A fee form is the way a helper is paid — a fixed price per study, time and materials, a price per complete, or a subscription — and this book prints the form and never an amount. The words are used consistently through the chapter so that a reader can follow a proposal without translating it.

11.2 The Owner as Question-Setter and Decider

The owner writes the decision and the decision rule before any fieldwork begins, approves the one-page research brief, decides at the end and holds every purchase. Chapter 2 owns the design of that brief and the writing of the rule, and this section does not redo it; what this section adds is the list of things the owner keeps, and the reason the list is short. The owner keeps the question, because a question written by somebody else is a question the owner will not recognize at the end. The owner keeps the rule, because a rule written after the results are in is not a rule but a rationalization. The owner keeps the decision, because the decision is the thing the business answers for. And the owner approves what leaves the firm: the interview guide, the questionnaire, the invitation, the report that goes to a partner. Four items, and none of them can be handed over without the study losing its point.

What the owner can delegate is everything else. The desk study is a reading task and it can be done by anybody with an afternoon and a list of sources. The counts are arithmetic and they can be done by anybody who can use a spreadsheet. The log is a habit and it can be kept by whoever holds the calendar. The fieldwork is a set of calls and it can be made by a service. The owner is not asked to be a statistician, and the owner is not asked to become one. The owner is asked to write two sentences before the work starts — what decision this serves, and what result would change it — and to read the brief at the end. That is the whole of the owner's method, and it fits on an index card.

Figure 11.1 lists, question by question, which role answers it in the case's study and what that role does. The figure has six rows, each beginning with a teal question, then an orange box for the role in the case, then a gray box for what that role does; the lawyer's row is in steel gray. The owner writes the decision and the rule and decides at the end, and the administrator keeps the desk study, the survey frame and the watch log across 38 hours. The owner and the lead technician hold fourteen interviews, and the fieldwork service makes the survey calls, for 92 completed answers at $25 each, for research only and with no offer. The dispatcher checks the zones walked against the routes in 3 hours, and the outside lawyer answers on the survey invitations, the line of the watch and the data kept, with Books H2 to H5 holding the rules.

Figure 11.1 — Who does what. The owner writes the decision and the rule and decides at the end; the administrator runs the desk study, the survey frame and the watch log in 38 hours; the owner and the lead technician talk to the field in fourteen interviews; the fieldwork service makes the survey calls, 92 completed answers at $25 each; the dispatcher checks the zones walked in 3 hours; the outside lawyer answers whether a step is lawful.

In the case, Sophie Nadeau holds the question and the rule, joins the fourteen interviews with the lead technician, and reads the decision brief at the end. Every one of those facts is posed by this book and illustrative. The owner's 22 hours are counted and not priced, as Book B5 (Content Marketing and the Content Factory) does with the owner's time, and the counting is deliberate: a study that prices the owner's hours at a rate the business does not pay would report a cost the business never incurred, and a study that ignores them would pretend the work was free. The owner's time is a cost even when it is not priced, and the honest way to carry it is to count the hours, name them, and leave them out of the cash total.

KEY CONCEPT The owner keeps the question, the decision rule and the decision, and buys only the work

The central idea of this chapter is that a helper can run fieldwork, count, analyze and report, but the decision belongs to the person who answers for it, and a rule written by the owner before the fieldwork keeps a helper's findings from deciding by themselves. A firm that is hired to design a study will design a good study; it will not know which of two answers the business can act on, and it will not be the one who lives with the choice. That is why the question and the rule are written first, in the owner's words, and why the brief that carries them is approved before the first call is made. The case's roles and hours are posed by this book and illustrative. No source read for this chapter measures how often businesses hire help, and no share of businesses that do research is printed anywhere in this book. The idea is the book's reasoning, and it is offered as a way to divide the work rather than as a finding about how businesses behave.

11.3 The Person Who Runs the Research Day to Day

The person who runs the research day to day is the person who keeps the study moving between the moments when somebody decides something. In the case that person is the administrator, who keeps the desk study, the survey frame, the signal log, the repository and the task list, and who is the same person the cadence of Chapter 10 already relies on. The role is not a research role in the professional sense; it is a coordination role. A study with a named coordinator has a next step at every moment. A study without one has a next step whenever the owner remembers.

What the person needs is short and specific. The first need is a task list in which every task has an owner, a date and a verification criterion, which is the same list the weekly review of Chapter 10 reads. The second is a place for the files: the brief, the guide, the frame, the responses, the notes, the codes, the log and the report, in one folder that another person can find without asking. The third is a backup, named in advance, who can run the weekly review and keep the list current during an absence. None of the three is expensive and all three are easy to skip, and a study that skips them is a study that depends on one person's memory for its whole length.

What the person is not is a decision-maker. The coordinator does not decide whether a channel is cut, whether the third territory is opened or whether a finding is strong enough to act on. The coordinator writes the question down and puts it in front of the person who owns it. This boundary is not a demotion; it is what makes the role sustainable. A coordinator who is asked to decide will start to shade the findings toward the decision that seems expected, and a coordinator who is asked only to run the work can report a result that contradicts the plan without it being a personal matter.

In the case, the administrator runs the desk study, the survey frame and the watch log across 38 hours, posed by this book, and keeps the answers people give for the research only. That last check is small and it is the one most easily lost. The answers a person gives in a survey serve the research, and they are never used to put that person on a mailing list or to call back with an offer. The code of self-regulation that governs research asks for a clear distinction between research and non-research activity, and for the person to be told before collection if the data will be used for anything else, as section 11.9 sets out with its article. The administrator is the person who keeps that distinction in practice, and Book H2 (Privacy and Personal Information) holds the rules on personal information and consent, with Book N4 (CRM, Contacts and Prospect Data) holding the rules of contacts in La Machine.

WHO DOES THIS The owner decides, the administrator runs the work, and the outside lawyer reads the line

In the case, and posed by this book, Sophie Nadeau, the owner, writes the decision and the rule, joins the interviews, reads the brief and decides; the administrator runs the desk study, the frame and the log across 38 hours; the lead technician checks the technical words in 4 hours; the dispatcher checks the zones against the routes in 3 hours; the fieldwork service makes the survey calls; and the outside lawyer reads the script, the invitation, the line of the watch and the retention of the files. In a solo business the same work is done by the owner, who buys the calls and keeps the rest, and that is the book's suggestion rather than a fact of the case. In a larger business the work is held by a manager of research and an analyst, which is also the book's suggestion and not part of the case. No hour is promised to any reader, and the hours of the case are the case's own.

11.4 Research Firms and Independent Researchers

A research firm brings four things to a study, and none of them is a better answer. The first is method: a firm designs studies for a living, and its design may be tighter than a design written in an afternoon. The second is a team: a firm can put several people on a study at once, which is what makes a short schedule possible. The third is independence: a firm is not the business, and a respondent who is asked a question by a firm is not being asked by the person who sells to them. The fourth is capacity: a firm can put more people on the calls than a small team can spare. What a firm does not bring is the question. A firm will answer the question it is given, and if the question is vague the answer will be vague in a well-formatted way.

What a firm costs in time is worth counting before it is hired. A study run by a firm needs a brief written by the owner, a proposal read and questioned, a design approved, a questionnaire or a guide reviewed, a fieldwork period monitored, a report read and a set of findings discussed. That is not a small amount of the owner's time, and it is the part of the cost that a fee does not show. A business that hires a firm to save time and then spends the same time managing the firm has bought method and independence rather than hours, which may still be the right purchase and should be the purchase it knows it is making.

How a firm or an independent researcher is paid is described by the form of the fee, and this book prints no amount. The forms are a fixed price per study, time and materials, a price per complete, and a subscription. A fixed price per study is quoted for a defined piece of work and it puts the risk of overrun on the firm. Time and materials is billed by the hour or the day and it puts the risk on the client. A price per complete is paid for each finished interview or questionnaire and it is a form a fieldwork service may use. A subscription is a recurring fee for access to a panel or a platform. Each form suits a different kind of work, and a business comparing a fixed price with an hourly rate is comparing two different things rather than two prices.

The industry group that lists the polling industry does not break out research firms, and no count specific to research firms was read. The industry profile of other professional, scientific and technical services lists the industry of marketing research and public opinion polling under its group, and its own page returned no data on the day it was read, October 10, 2026, so the page supports no count of research firms and no market size of the research industry (Innovation, Science and Economic Development Canada, n.d.). A business that wants to know how many firms exist in its region will not find the number in that page, and this book prints none.

The council's code is the Canadian Code of Market, Opinion and Social Research and Data Analytics, adopted from the international code with the support of the international association, and its accredited and associate agency members must comply with a published international standard on market research; the standards page was read on October 10, 2026 (Canadian Research Insights Council, n.d.-b). The text of that standard was not read for this book, so no requirement of it is printed here, and a certificate is not a measure of quality. The international code itself applies to everyone involved in research, including users of self-service platforms, and it makes the client responsible for seeing that its contractors comply (International Chamber of Commerce & ESOMAR, 2025, at the level read). A code binds its members and the associations that adopt it, and it is never a law. The contracting of an agency or a freelancer in general is Book I8 (Freelancers, Agencies and Outsourcing), and the contract itself is Book H1 (Contracts and Commercial Agreements).

In the case, no research firm is hired, posed by this book. The study is run by the team, with the survey calls bought from a fieldwork service, and that choice is the case's and not a recommendation.

WHAT CHANGED The international research code was revised in 2025 after its 2016 version

The dated change here is in the sources and not in the case. The international code of market, opinion and social research and data analytics was revised in 2025 after its 2016 version, it applies to everyone involved in research including users of self-service platforms, and it asks that the client be informed when artificial intelligence or synthetic data are used (International Chamber of Commerce & ESOMAR, 2025, at the level read, articles 1 to 8). The council's standards page, read on October 10, 2026, lists guiding principles on the use of artificial intelligence in market research among the documents it publishes (Canadian Research Insights Council, n.d.-b). A code is self-regulation: it binds its members and the associations that adopt it, never a business by law. The dates of sources are never dates of the case, which is counted in days, and no requirement of a document whose text was not read is printed here.

11.5 Panels, Recruiters and Fieldwork Services

Three services can be confused with one another, and the confusion costs money. A panel is a list of people who agreed to answer surveys. A recruiter finds participants for interviews, usually against a profile the business describes. A fieldwork service makes the calls or runs the platform for a study, on a script and a list that somebody else designed. The three are different purchases. A panel supplies respondents. A recruiter supplies interviewees. A fieldwork service supplies labor. A business that wants answers to a short questionnaire is buying fieldwork, and it may or may not also be buying access to a panel.

An opt-in panel is a non-probability sample, and no margin of error is printed for it. That sentence is the whole of the technical point, and it has a practical consequence: a business that runs a survey through an opt-in panel gets a count of answers and not an estimate with a known precision. The council's page on reporting precision recalls that, under the standards of the Government of Canada, no other measure of precision may be quoted for a non-probability sample, and that no statement may suggest a sampling error equivalent to that of a probability sample of the same size; the page was read on October 10, 2026 (Canadian Research Insights Council, n.d.-a). The page is self-regulation of the council's members and it binds no business that is not a member, and it states what an opt-in panel can and cannot support.

The comparison that a business should have in mind is a study of American adults, read at abstract only, in which survey results were compared with benchmarks: probability-sample telephone and internet surveys were consistently more accurate than non-probability internet surveys, even after demographic post-stratification, which helped some non-probability surveys and harmed others, and higher completion and response rates of the surveys were associated with less accuracy (Yeager et al., 2011, at abstract only). The study concerns American adults and it supports no figure for Canada and none for a business population, and it is printed here as one study of one setting. What it supports is a caution about non-probability samples of the kind an opt-in panel supplies, and nothing more.

The calls themselves have a Canadian feature worth knowing. Calls made by organizations conducting surveys, public opinion polls or market research are exempt from the National Do Not Call List Rules, survey and market research calls made solely for those purposes are not considered telemarketing, and market research and polling firms are not required to keep internal do-not-call lists (Canadian Radio-television and Telecommunications Commission, 2018, page modified April 9, 2018). The exemption is for research calls and it does not cover a call that also sells or books an appointment; that call is telemarketing, and Book H3 (Anti-Spam and Telemarketing Compliance) holds the rule. The fieldwork service says who is calling and for whom, and the call offers and sells nothing.

In the case, the fieldwork service makes the survey calls, 92 completes at a price per complete of $25, posed by this book, from a frame of 1,259 buildings and 360 numbers drawn. The price per complete is the case's own and it is not a norm, not a benchmark and not a rate any reader should expect. The frame, the number drawn and the completes are all posed by this book, and the response the case reads is its own count on its own day.

QUÉBEC & CANADA NOTE Calls made solely for surveys, polls or market research are exempt from the National Do Not Call List Rules

The national do-not-call page for organizations, modified April 9, 2018 and read on October 10, 2026, says that calls made by organizations conducting surveys, public opinion polls or market research are exempt from the National Do Not Call List Rules, that such calls made solely for those purposes are not considered telemarketing, and that market research and polling firms are not required to keep internal do-not-call lists (Canadian Radio-television and Telecommunications Commission, 2018). A call that also sells or books an appointment is telemarketing, and Book H3 (Anti-Spam and Telemarketing Compliance) holds the rule. The page does not give calling hours or identification rules, and this book prints none. Whether a particular call falls on one side of the line is a question for a lawyer, and this book explains the page and recommends no vendor. The box does not tell any reader what their own calls require.

11.6 Analysts, Statisticians and Data Specialists

Some research tasks need a person who handles numbers, and the tasks are narrower than they sound. Drawing a sample size, weighting a set of answers, testing whether a difference could be chance, and reading a table of counts for a pattern are four of them. The question is not whether the business needs a statistician on staff. It is whether a specific task needs one, and a count a spreadsheet can make may not, while a sample size and a weighting may.

The manual of the statistics agency is the reference this book uses for the arithmetic, and it is a methodology manual of a public agency rather than a rule binding a private business. It establishes that there are two types of sampling, that sampling error decreases as the sample grows, and that nonsampling errors exist in samples and censuses alike, which means that a larger sample does not reduce a systematic bias (Statistics Canada, 2010, archived manual). The same manual gives the simple-random-sample size for a margin of plus or minus 5 points at 95 percent confidence when the estimated proportion is one half, and its equation brings the population size in. Those are a statistician's tools, and a business that uses them is using a method rather than following a rule.

In the case, the sample size for plus or minus 10 points at 95 percent confidence is 90 completes, computed with the manual's equation for a finite population and a z value of 1.96, and checked by a person, and that computation belongs to Chapter 5. The number is posed by this book and it is the case's own; it is not a typical size, not a benchmark and not a recommendation for any other study. What matters here is who did the arithmetic: the administrator, with a spreadsheet, and a person reading the result before it was used. A computation that nobody checks is a computation that can be wrong in a way that only shows up at the end, when the answers arrive and the margin turns out to have been computed on the wrong population.

What to ask of an analyst is short. Ask for the formula, so that the business can see what was computed. Ask for the assumptions, so that the business can see what was assumed about the population and the design. Ask for the code of the work in a form the firm can read, so that the computation can be repeated next year without hiring the same person again. An analyst who answers those three questions has done the job, and an analyst who answers with a number and no formula has left the business unable to check anything. In the case, no analyst is hired, posed by this book, and the administrator computes with a spreadsheet. A statistician's help is for a question the firm cannot answer, not for a number the firm already has.

11.7 Legal and Privacy Advice on Research

A market study touches four areas of law, and the business does not decide any of them. The first is whether an email invitation to a survey is a commercial electronic message, which is Book H3's question and is never decided here. The second is whether a recording or a set of answers needs a given consent, and how long the data may be kept, which is Book H2's. The third is whether an exchange with a competitor can raise concerns under competition law, which is Book H4's; the Bureau's guidelines say that exchanging pricing information, costs, trading terms, strategic plans or marketing strategies between competitors can raise concerns under the Act, that exchanges of current or future information are riskier than exchanges of historical information, and that information shared through a third party in aggregated form is generally less concerning than information shared directly (Competition Bureau, 2021, guidelines dated May 6, 2021). The guidelines give no verdict on a given exchange, and this book gives none either. The fourth is whether a purchased report may be copied or circulated, which is Book H5's.

What to bring to a lawyer is a short list, and bringing it makes the advice cheaper and better. Bring the script: the interview guide and the questionnaire, so that the questions themselves can be read. Bring the invitation: the exact text of the email or the opening words of the call. Bring the consent text and the privacy notice, so that what a person is told before answering can be checked. Bring the list of data kept and the retention date: what is stored, where, for how long, and who sees it. Bring the line of the watch: what the business reads about its competitors and what it does not. Five items, and a lawyer who has them can answer in an hour what would otherwise take a meeting to frame.

The questions a lawyer answers are not the questions the book answers, and the difference is worth stating plainly. This book describes published rules on research, personal information, competition and copyright and is not legal advice. It does not tell a reader whether a recording, an invitation, a call, a watch or a report is lawful, and it does not declare any practice compliant. What it does is name the question and send the reader to the book that owns the rule and to a lawyer. Book H2 holds personal information and consent. Book H3 holds electronic messages and calls. Book H4 holds agreements between competitors, section 45 of the Competition Act and the Bureau's guidelines. Book H5 holds trade secrets, section 391 of the Criminal Code, and fair dealing for research, section 29 of the Copyright Act. Book H1 holds the contract with a helper.

In the case, the outside lawyer reads the script of the interviews, the survey invitation, the line of the watch and the retention of the files, posed by this book, and no fee is printed. The lawyer's reading is a step in the plan and not a verdict on the study, and the case records that the step was taken rather than what the lawyer said.

11.8 Reading a Research Proposal and Its Report

A proposal is read for seven things, and the order matters less than the completeness. The first is the question it answers, and whether that question is the business's question and not a question the firm finds convenient. The second is the method: what will be done, in what order, by whom. The third is the sample and its frame: who will be asked, how they were selected, and what the frame leaves out. The fourth is the fee form and what it covers, without comparing amounts of different forms. The fifth is the deliverables: what the business receives, in what form, on what dates. The sixth is who owns the data and who may see the answers people give. The seventh is the dates, including the date the fieldwork closes and the date the report arrives. A proposal that answers all seven can be signed or refused on its content. A proposal that answers four of them cannot be judged at all.

A report is read against a list of disclosures, and the list is the one a United States association publishes for the release of results. When results are released, the sponsor and the conductor, the exact wording of the questions and their context, the population, whether the sample is probability-based or not with the frame and its gaps, the recruitment and selection, the modes and languages, the dates, the sample sizes, the margin of sampling error for probability samples, the weighting, the data quality checks and a statement on unmeasured error are disclosed; for non-probability samples, measures of precision are reported only if defined and accompanied by the model and its assumptions (American Association for Public Opinion Research, 2021, disclosure standards revised April 2021). The list is a United States association's and it binds no Canadian business; this book uses it as a list of questions to ask of a report, and a report that answers them is a report a reader can check.

Two further points come from the self-regulation of research. The international code asks that findings be distinguished from interpretation and recommendations, that limitations be disclosed, and that the client be informed when artificial intelligence or synthetic data are used (International Chamber of Commerce & ESOMAR, 2025, at the level read). The council's page requires of its members that the results of a probability sample carry an estimate of the error that sampling introduces, with the limits from other sources of error and potential bias disclosed (Canadian Research Insights Council, n.d.-a). The same page, read on October 10, 2026, asks that open-link surveys disclose their limits, since such surveys carry risks of bias, fraudulent and duplicate answers, manipulation and lack of representativeness. Both are self-regulation: they bind members and adopting associations, never a business by law.

The table below sets six parts of a proposal and a report beside the question to ask about each, a sign of trouble, the source of the question and the book that holds the rule. It is the book's method and not a finding, and it prints no fee. Read it as a checklist to carry into a meeting: the middle column is what to say out loud, and the last column is where to go when the answer matters legally.

Table 11.1 — What to read in a proposal and in a report

WATCH OUT A proposal read for its length, a panel read as a sample and a fee read as quality are three ways help is chosen badly

Three traps catch a business that is buying help, and each has a question that catches it. The first trap is a long proposal taken for a good one; the question is whether the proposal answers the owner's question and names the method and the frame, and a short proposal that does is worth more than a long one that does not. The second trap is a panel's answers taken for a sample of the market; the question is whether the sample was drawn at random from a named frame, and whether a margin is printed for it, and the answer for a panel is that it was not and that none is. The third trap is a high fee or a certificate taken for quality; the question is what was read and what it supports, and a certificate names a standard whose text this book did not read. The three traps are the book's reasoning and illustrative, and no frequency of any trap in any business is printed.

METHOD Reading a research proposal in five steps

First, find the question and the decision it serves, and check that the decision rule is the owner's and was written before the fieldwork. Second, read the method, the frame and the sample, and ask who drew the sample and how. Third, read the fee form and what it covers, without comparing amounts of different forms. Fourth, read who owns the data, how long it is kept, and who sees the answers people give. Fifth, read the deliverables against the list of what a report discloses, and write the questions to ask before signing. The five steps are the book's reasoning and not a tested procedure, and they print no fee and no figure beyond the case's own. A proposal that survives the five steps can still be refused on its content, and a proposal that fails one of them is worth a question before it is refused.

TOOLS OF THE TRADE What any proposal checklist must offer, whatever it is called

A checklist for reading a proposal needs six functions and no brand name. It needs a line for the question and the decision it serves, so that the owner's rule is on the page. It needs a line for the frame, the sample and who drew it, so that the selection can be seen. It needs a line for the fee form, with no amount compared across forms. It needs a line for who owns the data, who may see the answers and how long they are kept. It needs a list of deliverables with their dates, so that the report can be checked when it arrives. And it needs a place to write the questions to ask before signing, with the date of the review. A checklist with those six functions can be a page in a notebook, and a checklist without them is a formality.

11.9 Research Work That Stays Inside the Firm

Some research work stays inside the firm, and it stays there for two reasons rather than one. The first is knowledge: the question, the rule and the decision need the firm's own understanding of its market, and no helper can supply that. The second is cost: the desk study, the log and the watch are cheap in hours and expensive to explain, and a business that buys them pays a helper to learn what it already knows. The work that stays inside is therefore the work that is closest to the decision and the work that is cheapest to do. The work that is bought is the work that needs a skill, an independence or a capacity the firm does not have.

What a firm may do itself with care is more than it thinks. Interviews with its own customers are the clearest case, and Book B10 (Customer Experience, Retention and Loyalty) holds the interviews with loyal, new and lost customers as a practice of its own, in chapter 4 of that book. A survey of a list the firm holds is also possible, with the caveat that a list the firm holds is a frame with known gaps, and that the answers describe the list rather than the market. The desk study is a reading task and it is the firm's by default. The watch is a habit and it is the firm's by default. What is not the firm's by default is a probability sample drawn from a frame the firm does not hold, and an analysis beyond the firm's skill.

The table below sets seven tasks of a market study beside whether the case keeps them inside the firm or buys them, the reason, and who decides. The case's choices are posed by this book and illustrative, and the reasoning in the middle columns is the book's and not a finding. Read it as a division of labor rather than a rule: a different business with different skills will draw the line in a different place, and the line is drawn task by task rather than once for the study.

Table 11.2 — Inside the firm or bought, by task

In the case, the owner decides, the administrator runs the work, the fieldwork service makes the calls and the outside lawyer reads the line, and every one of those facts is posed by this book. The study cost $3,903 in all, of which $1,603 was internal and $2,300 was cash, and the revenue at stake was $162,279, of which the study is 2.4 percent. Those figures are the case's own, posed by this book, illustrative, and not a norm, not a benchmark and not a rate any reader should expect. Nothing is credited to the study: no contract, no sale and no share is attributed to the research, the insight or the watch, and the decision rule decides.

DO THE MATH What the case paid for the study, in five steps

Illustrative figures; the case is fictional. Step 1 — The internal cost is $1,603, which is the administrator's 38 hours at $35, the lead technician's 4 hours at $42 and the dispatcher's 3 hours valued at the administrator's rate. Step 2 — The cash cost is $2,300, which is the fieldwork service's 92 completes at $25 each. Step 3 — The two add to $3,903. Step 4 — The owner's 22 hours are counted and not priced, as Book B5 (Content Marketing and the Content Factory) does, so the study's cost is a floor and not a full price. Step 5 — The revenue at stake is $162,279, and the $3,903 is 2.4 percent of it, rounded to the nearest tenth of a percent. The figures are posed by this book, are not a norm, and credit no contract or sale to the study.

CASE IN POINT Nordec Entretien inc. keeps the question and buys the calls

Nordec Entretien inc. is a fictional business created for this encyclopedia; its owner, its employees, its customers, its competitors, its figures and the events described are invented, and any resemblance to a real business is coincidental. Sophie Nadeau, the owner, writes the rule and decides. The administrator runs the work across 38 hours. The lead technician and the dispatcher help in 4 and 3 hours. The fieldwork service makes the survey calls for research only, and the calls offer and sell nothing. The outside lawyer reads the line. Every one of those facts is posed by this book, illustrative and not a forecast, and no result of the study is reported here. The answers people give serve the research only, and no owner hidden in the assessment roll is ever identified.

THE RULE The client answers for its helpers, and the question, the rule and the decision are never delegated

The rule is drawn from the code of self-regulation as this book reads it: the code applies to everyone involved in research, including users of self-service platforms, and the client is responsible for making its contractors comply (International Chamber of Commerce & ESOMAR, 2025, at the level read). The book's own rule for this chapter adds that the owner keeps the question, the decision rule and the decision. A code binds its members and the associations that adopt it, and it is never a law; the rule here is the book's reading and neither a standard nor a legal duty. The example is the fieldwork service, whose calls offer and sell nothing, and whose answers serve the research only. Whether a given arrangement satisfies a given rule is a question for a lawyer, and this book decides none of it.

IN LA MACHINE Contacts hold every person who submits a form, and nothing separates a respondent from a prospect

CRM → Contacts business/marketing/contacts holds every person who submits a form, and nothing separates a respondent from a prospect automatically (La Machine, 2026). This book keeps research data outside the prospecting file, as the code asks (International Chamber of Commerce & ESOMAR, 2025, art. 1(d)-(e)), and Book N4 (CRM, Contacts and Prospect Data) holds the rules of contacts. La Machine recruits no participant, runs no panel, draws no sample and computes no margin of error, so a questionnaire hosted in a form reaches only the people the business already has a way to contact, and each of them becomes a contact. Check it in your installation. La Machine is one worked example of a business platform, described and never sold, and it is not the tool of the case's study: the case does not use it for the survey.

Chapter 12 reads the tools of this work by function, and Chapter 13 turns the whole book into a thirty-day plan.

Costly Mistakes

• Buying the whole study because the pieces cannot be seen. A business that cannot name the five skills of a study buys a firm to do all of them, and then reads a report that answers a question the owner never quite asked.
• Doing the whole study inside because the firm is proud of its knowledge. A business that draws its own sample from a list it holds, makes its own calls and reads its own numbers has a study of its own list, and it will report the result as a study of the market.
• Handing the decision rule to a helper. A rule written by the person who runs the fieldwork is a rule that will fit the fieldwork, and the owner will discover at the end that the threshold moved.
• Reading a proposal for its length and a certificate for its quality. A long proposal can answer nothing, and a certificate names a standard whose text this book did not read; the question is what the proposal answers and what the certificate supports.
• Printing a margin of error for a panel, an open link or a set of interviews. A margin belongs to a probability sample with a named frame, and an opt-in panel is a non-probability source whose members chose to be there.
• Asking a fieldwork service for the names of the people who answered. The answers serve the research only, and nobody is added to a mailing list or called back with an offer because of an answer.
• Leaving the legal questions to the end. The invitation, the script, the consent text, the retention and the line of the watch are read before the fieldwork, not after a complaint, and the questions belong to Books H2 to H5 and to a lawyer.

Key Takeaways

• A market study needs five skills — setting the question, running the work day to day, talking to people, handling numbers and reading the rules — and a small business may hold three of them and buy the other two in small quantities.
• The decision to do a task inside or to buy it turns on three questions asked one task at a time: does it need a skill the firm lacks, does it need independence, and does it need capacity.
• The owner keeps the question, the decision rule, the decision and the approval of what leaves the firm, and delegates the desk study, the counts, the log and the fieldwork.
• A research firm brings method, a team, independence and capacity, and it is paid by a fee form — a fixed price per study, time and materials, a price per complete or a subscription — never by an amount this book prints.
• an opt-in panel is a non-probability source and no margin of error is printed for it; calls made solely for surveys, polls or market research are exempt from the National Do Not Call List Rules, while a call that also sells is telemarketing and Book H3's.
• A proposal is read for the question, the method, the frame, the fee form, the deliverables, the data and the dates, and a report is read against a fixed list of disclosures, with findings kept apart from interpretation and recommendations.

Your Action Plan

1. Name the five skills and mark which ones the firm has. Write the five skills on one page, mark each as held, partly held or absent, and note the task that depends on each. Deliverable: a one-page skills sheet. Verification: every skill has a name beside it and every absent skill has a task attached to it.
2. Write the question and the rule before anything is bought. Put the decision and the decision rule in the owner's own words, approve the one-page brief, and date it. Deliverable: a dated brief with the rule in it. Verification: the rule names a threshold and the brief was approved before any fieldwork began.
3. Apply the three questions to each task. For every task of the study, ask whether it needs a skill the firm lacks, whether it needs independence and whether it needs capacity, and record the answer. Deliverable: a task list with an inside-or-bought decision on each line. Verification: every bought task names the fee form and every inside task names the person who runs it.
4. Read the proposal against the seven points and the report against the disclosure list. Carry the checklist into the meeting, ask the questions out loud, and write the answers down before signing. Deliverable: a marked-up proposal and a list of questions asked. Verification: the question, the method, the frame, the fee form, the deliverables, the data and the dates each have an answer on the page.
5. Send the legal questions to a lawyer before the fieldwork. Bring the script, the invitation, the consent text, the list of data kept and the retention date, and the line of the watch. Deliverable: a dated note of what was sent and what came back. Verification: the five items were sent together, and no answer was treated as a verdict on the study.

In La Machine

The application copy read for this book holds a contact for every person who submits a form, and it holds nothing that separates a respondent from a prospect (La Machine, 2026). CRM → Contacts (business/marketing/contacts) is the page where those people live, and the fact that a person answered a questionnaire is not recorded there as a research fact, so the separation is a rule the business keeps itself, with Book N4 (CRM, Contacts and Prospect Data) holding the rules of contacts. Check it in your installation, because what a given installation shows depends on the roles and the settings it carries.

What a reader needs for the work of this chapter lives elsewhere, and it is worth naming plainly. A research file kept apart from the prospect list, a named person who runs the work day to day, a proposal read against a checklist before it is signed, and a lawyer for the questions that belong to Books H2 to H5. La Machine recruits no participant, runs no panel, draws no sample and separates no respondent from a prospect by itself. It is one worked example of a business platform, described and never sold, and the case does not use it for its survey.

Self-Check

1. What are the five skills a market study needs, and which of them does a small business usually buy?
2. What does the owner keep, and what can the owner delegate?
3. What are the three questions that decide whether a task stays inside the firm or is bought?
4. What is the difference between a panel, a recruiter and a fieldwork service, and why is no margin of error printed for an opt-in panel?
5. What is read in a proposal, and what is a report read against?

Answers. 1. The five skills are setting the question, running the work day to day, talking to people, handling numbers and reading the rules; a small business may hold the first three in some measure and buy the fourth and fifth in small quantities, a few hours of a statistician and a few hours of a lawyer. 2. The owner keeps the question, the decision rule, the decision and the approval of what leaves the firm, and can delegate the desk study, the counts, the log and the fieldwork. 3. The three questions are whether the task needs a skill the firm lacks, whether it needs independence, and whether it needs capacity, and they are asked one task at a time rather than once for the whole study. 4. A panel is a list of people who agreed to answer surveys, a recruiter finds participants for interviews, and a fieldwork service makes the calls or runs the platform; an opt-in panel is a non-probability source whose members chose to be there, so no margin of error is printed for it, and the council's page recalls that no other measure of precision may be quoted for a non-probability sample under the standards of the Government of Canada. 5. A proposal is read for the question it answers, the method, the sample and its frame, the fee form, the deliverables, who owns the data and the dates; a report is read against the list of disclosures that a United States association publishes for the release of results, with findings kept apart from interpretation and recommendations.

Chapter References

American Association for Public Opinion Research. (2021). AAPOR code of professional ethics and practices: Disclosure standards (Revised April 2021). Retrieved October 10, 2026, from https://aapor.org/standards-and-ethics/disclosure-standards/

Canadian Radio-television and Telecommunications Commission. (2018). Exemptions [National Do Not Call List, information for organizations]. (Page modified April 9, 2018.) Retrieved October 10, 2026, from https://lnnte-dncl.gc.ca/en/Organization/Exemptions

Canadian Research Insights Council. (n.d.-a). CRIC requirements for reporting survey precision and on the use of open-link surveys. Retrieved October 10, 2026, from https://www.canadianresearchinsightscouncil.ca/cric-requirements-for-reporting-survey-precision-and-on-the-use-of-open-link-surveys/

Canadian Research Insights Council. (n.d.-b). Standards. Retrieved October 10, 2026, from https://www.canadianresearchinsightscouncil.ca/standards/

Competition Bureau. (2021). Competitor collaboration guidelines. Government of Canada. (Enforcement guidelines dated May 6, 2021; page modified June 27, 2024.) Retrieved October 10, 2026, from https://competition-bureau.canada.ca/en/how-we-foster-competition/education-and-outreach/competitor-collaboration-guidelines

Innovation, Science and Economic Development Canada. (n.d.). Other professional, scientific and technical services – 5419: Businesses [Canadian Industry Statistics]. Retrieved October 10, 2026, from https://ised-isde.canada.ca/app/ixb/cis/businesses-entreprises/5419

International Chamber of Commerce & ESOMAR. (2025). ICC/ESOMAR international code on market, opinion and social research and data analytics. ESOMAR. Retrieved October 10, 2026, from https://community.esomar.org/uploads/public/knowledge-and-standards/codes-and-guidelines/ICCESOMAR-International-Code_English.pdf

La Machine. (2026). Documentation — route reference inventory. Internal product documentation, read October 10, 2026, together with the code of the pages this book describes: business/marketing/contacts.

Statistics Canada. (2010). Survey methods and practices (Catalogue no. 12-587-X; originally published October 2003; archived). Retrieved October 10, 2026, from https://www150.statcan.gc.ca/n1/pub/12-587-x/12-587-x2003001-eng.htm

Yeager, D. S., Krosnick, J. A., Chang, L., Javitz, H. S., Levendusky, M. S., Simpser, A., & Wang, R. (2011). Comparing the accuracy of RDD telephone surveys and internet surveys conducted with probability and non-probability samples. Public Opinion Quarterly, 75(4), 709–747. https://doi.org/10.1093/poq/nfr020

Chapter 12 — Tools: Functions and Popular Tools

A tool is a function before it is a product, and the research file is the thing to keep, so this chapter says what the tools around a market study and a watch do by function, reads each function in a dated table of examples, places the pages of La Machine in their place and says what is missing, and reads how to choose a tool and what to read in a pricing page and a contract.

What You'll Be Able to Do

• Name the seven functions that tools serve around a market study and a watch, and say for each one what it sees and what it cannot settle.
• Read a tool table as a dated list of examples rather than as a ranking, and say what a pricing model is without confusing it with a price.
• Separate what a tool holds, counts or shows from the judgment that stays with the people who wrote the question, drew the frame and read the signal.
• Choose a tool in five steps, starting from what the business already has and ending with a written choice and a date on which it will be reviewed.
• Read a pricing page and a contract for the term, the renewal, the export, the exit and the ownership of the answers and the transcripts.
• Place the pages of La Machine in their place, and say plainly what the platform does not do for a study and for a watch.

Why It Matters

A business that has decided to run a market study will meet the tools before it meets the method. Each tool page is written to be convincing and offers a free trial, and the risk is that a business chooses the tool first and discovers the question second, so that the study begins with whatever the tool happens to do well. A survey product produces a questionnaire, a database produces a list, and a monitoring service produces a stream of alerts, and none of the three tells the business whether it is answering the question it needed to ask.

The remedy is to describe the work before describing the product. A market study and a watch have seven recurring pieces of work, and each has a family of tools behind it: surveys and forms, interviews and transcription, qualitative analysis, search and web analytics, listening and reviews, competitor and website monitoring, and enrichment and business data. Naming the function first tells the business what to look for and makes visible the parts of the work that no tool will do. A tool delivers a questionnaire, but the frame and the random draw that make a sample are done by a person, and so are the consent that allows a recording and the decision that a mention matters.

A tool is also a commitment that outlives the study. The answers a survey tool holds, the transcripts a transcription tool stores, the list a database compiles and the log a monitor keeps are records about people, and records about people carry obligations that do not end when the study does. Reading the pricing model and the contract before the trial is the same discipline this book applies to a source, applied to a supplier.

In This Chapter

Section 12.1 sets out the seven functions that tools serve around a research question and what each cannot settle. Section 12.2 covers survey and form tools, section 12.3 interview, transcription and qualitative analysis tools, section 12.4 search and web analytics tools, section 12.5 listening and review monitoring tools, section 12.6 competitor and website monitoring tools, and section 12.7 data enrichment and business database tools. Section 12.8 places the pages of La Machine in their place and says what is missing. Section 12.9 covers spreadsheets, statistics and visualization. Section 12.10 sets out how to choose a tool and what to read in a pricing page and a contract.

12.1 What Research Tools Do, by Function

A research tool is easier to choose once it is described by what it does rather than by what it is called. The market study of this book has seven recurring pieces of work, each with a family of tools behind it. A survey and form tool builds a questionnaire, sends it, collects the answers and exports them. An interview, transcription and qualitative analysis tool turns a conversation into a draft, stores transcripts, codes passages and shows themes. A search and web analytics tool shows the relative interest in a term, estimates how often it is searched, or measures the visits to the business's own site. A listening and review monitoring tool collects public mentions and reviews and shows counts, themes and sentiment. A competitor and website monitoring tool watches a page for changes and rings when one occurs. A data enrichment and business database tool adds fields to a record or lists companies and contacts. A spreadsheet, statistics and visualization tool holds the counts, computes the interval, draws the chart and keeps the log.

A function is not a product: one product can serve several functions, and one function can be served by several products, by a spreadsheet, or by hand. The seven families are a way of asking what work the study needs, not a shopping list.

The second thing to say is what no tool in any of the seven families does. No tool decides the research question, which is written with its decision rule before any tool is opened (Chapter 2). No tool draws a valid sample, because a sample is made by a frame and a random draw and not by a link (Chapter 5). No tool reads a signal (Chapter 7), and no tool writes a finding, because a finding is a sentence a person is prepared to defend (Chapter 9). A tool holds, counts or shows, and the judgment stays with the people.

The third thing to say is what this chapter is not. This book explains published rules on research, personal information, competition and copyright, and is not legal, tax or accounting advice. It recommends no vendor, no panel, no research firm, no fieldwork service, no data provider, no platform, no service and no tool, and where a question of personal information arises it sends the reader to Book H2 (Privacy and Personal Information) and to a lawyer.

Figure 12.1 places seven tool functions around the research question and says, for each, what La Machine covers and what it does not.

Figure 12.1 — Seven tool functions around the research question. Surveys and forms, interviews and transcription, qualitative analysis, search and web analytics, listening and reviews, competitor and website monitoring, and enrichment and business data each have a place; La Machine offers a form, a meeting, an alert and a company record, and nothing for the others, and spreadsheets and statistics sit under all of them.

The tables of this chapter are read in one way. A name in a row is an example of a function, never a recommendation. Every pricing cell states a model and no amount, because an amount is a unit, a currency and a billing period on a date. A row is set as of the day the vendor's public page was read, and a vendor's status can change: a product is acquired, a service is renamed, a plan is withdrawn. No row is read against another, and no table is a ranking.

The table below sets six neighbors beside what each holds and what this chapter holds, and it is the boundary table that opens the chapter.

Table 12.1 — Where this chapter stops and the books around it start

DEFINITION Tool, function, pricing model and example, as this book uses the words

These are this book's working words, and no source read defines them in this sense. A tool is software or a service that holds, counts or shows something for the study. A function is what a tool does for the study; one product can serve several functions, and one function can be served by several products, by a spreadsheet or by hand. A pricing model is the way a vendor charges, described by its form alone, such as a free tier, a subscription, a plan by usage, a plan on request or a one-time license, and stated without any amount. An example is a product named in a row of a dated table as an instance of a function, never a recommendation, never a ranking and never a claim about quality. The rows of the tables of this chapter were read on the vendors' public pages on October 10, 2026.

12.2 Survey and Form Tools

A survey and form tool builds a questionnaire, sends it by a link or an invitation, collects the answers and exports them. What it brings is speed, skip logic that shows a question only to the people for whom it applies, and a table of answers that can be exported and read.

What it cannot do is the part that decides whether the answers mean anything. A survey tool draws no sample. It delivers a questionnaire to whoever receives the link, and a link posted in a tool, on a page or in a social feed is an open-link survey, with the limits that the research council's page lists for its members, as self-regulation and never as a law, and that section 5.3 sets out: bias, duplicate and fraudulent answers, manipulation and a lack of representativeness, and the requirement that its limits be disclosed and that it never be combined with sample data. A sample is made by a frame and a random draw, and no margin of error is printed for an open link, an opt-in panel, a focus group or a set of reviews (Chapter 5).

A form that asks for an email address or a telephone number creates a record the moment an answer is stored, and a tool that finds or creates a record from every submission has put an answer into a prospecting file. The rule of this book keeps the two files apart: what a person says in a study stays in the research file, and nobody is put on a mailing list or called back with an offer because of an answer. The personal information a form collects belongs to Book H2 (Privacy and Personal Information), and whether an email invitation to a survey is a commercial electronic message is a question this book poses and sends to Book H3 (Anti-Spam and Telemarketing Compliance), never decided here.

The case buys no survey tool. The survey of Nordec Entretien inc. is run by the fieldwork service on its own platform outside the application, and the answers come back as a table. The frame of 1,259 buildings, the 360 numbers drawn at random and the 92 completes are posed by this book, and they are the case's own counts, not a norm and not a typical response rate; the price per complete of the fieldwork service is posed by this book.

The table below presents examples of survey and form tools, read on the vendors' public pages on October 10, 2026, with the pricing model stated and no amount printed; it is the only place in this book where such a tool is named.

Table 12.2 — Survey and form tools: examples of the function, read on October 10, 2026

KEY CONCEPT A survey tool sends questions, and the sample is made by the person who draws the frame

A tool delivers a questionnaire to whoever receives the link, and that is the whole of what it does with a sample. A sample is made by a frame and a random draw: the frame is the list of units the study wants to reach, and the draw is the rule that decides which of them are asked. A tool does not draw the frame, does not draw the units and does not know what the frame left out. A link posted in a tool is an open-link survey, and the council's page lists its limits and asks that they be disclosed (section 5.3). No margin of error is printed for an open link. The frame of 1,259 buildings and the 360 numbers drawn of the case are posed by this book and are not a norm.

12.3 Interview, Transcription and Qualitative Analysis Tools

The second family covers three pieces of work that are worth separating. A transcription tool turns a conversation into a draft of text. A qualitative analysis tool stores transcripts, lets a person code passages and shows the themes that the codes form. A tool that applies an AI model to a transcript is an AI assistant, described by function, and Book K4 (Artificial Intelligence for Business) treats that family in depth. The family brings a searchable text, a coding grid and one place for the material of a study.

What the family cannot do is decide anything. It decides no theme, because a theme is a judgment about what a set of passages has in common. It gives no number of interviews that is enough, because saturation is a property of a study and not of a tool, and the two studies read on it, each at its abstract only, are given with their numbers in section 4.7. Neither number is a rule for a business study.

A recording needs the participant's consent before it starts, given and noted, with the purpose said before any question and the participant free to stop at any time; the code of self-regulation asks for that, as Chapter 4 sets out, and the personal information of a participant is Book H2's subject, called here and never rewritten (Personal Information Protection and Electronic Documents Act, Schedule 1, clauses 4.2 to 4.5). The case records nothing: its 14 conversations are written up by hand in notes of 15 minutes each, with no recording and no transcription tool. That is a choice the case made and not a rule for a reader.

The rule that governs every draft is that a transcription tool produces a draft, and a person checks every quotation against the notes before the quotation is used. An AI assistant can be asked for a first pass over a transcript, and the administrator rewrites what comes back and Sophie Nadeau, the owner, approves it. No questionnaire is fielded, no finding is reported and no decision brief is sent without a person reading it first.

The case buys no tool of this family. Its 14 conversations, 30 codes and last new code at interview 9 are posed by this book. A spreadsheet with one row per conversation and one column per code is a coding grid, and the question to ask is whether a person can still find a passage six months later.

The table below presents examples of interview, transcription and qualitative analysis tools, read on the vendors' public pages on October 10, 2026, with the pricing model stated and no amount printed; it is the only place in this book where such a tool is named.

Table 12.3 — Interview, transcription and qualitative analysis tools: examples of the function, read on October 10, 2026

WATCH OUT A transcription tool, an AI assistant and a coding tool each produce a draft that a person must check line by line

Three traps sit in this family, and each has a question that catches it. The first is a transcript that nobody reads against the notes: what would show that the quotation is exact? Without a person with the notes and the draft side by side, the quotation is the tool's text and not the participant's words. The second is a theme proposed by an AI assistant that the team accepts at once: which passages support it, and which contradict it? A theme that cannot be shown in passages is a guess with a label on it. The third is a code applied by a tool to every passage without a person reading the grid: who decided what each code means? A code is a decision, and a decision that nobody made is not a finding. This book prints no frequency of any of the three traps in any business.

12.4 Search and Web Analytics Tools for Research

The third family covers three different things. A tool that shows the relative interest in a term over time is a search-interest tool, a tool that estimates how often a term is searched is a keyword tool, and a tool that measures the visits to the business's own site is a web analytics tool. The first two are objects of study in this book, and the third belongs to Book B7 (Marketing Analytics, Attribution and Dashboards).

What the search-interest function brings is a direction over time, and what it cannot do is give a volume. The provider's own help page says that its data are a largely unfiltered sample of actual search requests, anonymized, categorized and aggregated; that each point is divided by the total searches of its place and time and scaled from 0 to 100, so that the values show relative popularity and not search counts; that the data are not a scientific poll and reflect searches only; and that noise is added for privacy, so that low-interest queries can show misleading spikes (Google, n.d.-a). An index is therefore read as an index: a peak in a month says that interest was higher in that month than in the others on the same chart, and nothing about how many people searched or bought.

What the keyword function brings is a range, and what it cannot do is give a count. The provider's help page on its keyword planner says that the planner shows estimates of the number of searches a keyword gets each month; that keywords with very low search volume are not discoverable or forecastable; and that forecasts are refreshed daily and based on the last 7 to 10 days, adjusted for seasonality (Google, n.d.-b). Whether volumes are shown as ranges for some accounts is not stated in the part of the page that was read, and this chapter says so rather than assuming it. An estimate is read as an estimate, and no statement that the estimates are counts is supported by the page read.

The case's index of Chapter 6 peaks in months 4 and 10, posed by this book, and is an index and not a volume; keyword research for a page is Book C3's (Search Engine Optimization). The case buys no search tool and no analytics tool for research.

The table below presents examples of search and web analytics tools, read on the vendors' public pages on October 10, 2026, with the pricing model stated and no amount printed; it is the only place in this book where such a tool is named.

Table 12.4 — Search and web analytics tools: examples of the function, read on October 10, 2026

THE RULE A relative index or a keyword estimate is never a volume, and the tool's number stays the tool's number

The rule is drawn from the two help pages as this book read them. The trends data are a largely unfiltered sample of actual search requests, scaled from 0 to 100, and they show relative popularity, not search counts (Google, n.d.-a). The keyword planner shows estimates of the number of searches a keyword gets each month, and keywords with very low search volume are not discoverable or forecastable (Google, n.d.-b). The book's own rule follows from the two: a number read in a tool is printed with the tool's description of it, never as a count of searchers, buyers or customers, and never as a market share. The rule is this book's reading of two help pages; it is neither a standard nor a legal duty. The example is the case's index, which peaks in months 4 and 10 and is an index and not a volume.

12.5 Listening and Review Monitoring Tools

The fourth family collects public mentions and reviews of a name and shows counts, themes and sentiment. What it brings is a stream of mentions, sorted by source, with a count per period and a label for the tone.

What the family cannot do is turn that stream into a sample of a market. Reviews and posts are written by those who chose to write, and the people who chose to write are not a random draw from the people who bought. The study read on this subject, cited at its abstract only in section 6.5, reports that the average online rating is biased by two self-selection biases, an acquisition bias, because mostly people predisposed to like a product buy it and review it, and an underreporting bias, because people with extreme opinions review more, which together produce the J-shaped distribution of reviews; consumers try to correct for the biases but cannot fully. That study concerns product reviews and not business services, and it is one study of one setting. The collection of reviews belongs to Book C9 (Reviews, Testimonials and Social Proof), and the listening of the business's own accounts to Book C7 (Social Media Marketing and Syndication, section 9.5).

A second limit is the sentiment label, which applies the tool's own model to a sentence and is an interpretation that a person can accept or reject. The case reads its 40 public reviews by hand and codes them into four themes, and the counts are posed by this book.

The case buys no listening tool and no review monitor. The 40 public reviews of Chapter 6 were read and coded by hand, and the counts (response time 14, billing surprises 9, the same technician each visit 8, other 9) are posed by this book. A business that reads far more reviews may want a tool, and the tool will still be reading what people chose to write.

The table below presents examples of listening and review monitoring tools, read on the vendors' public pages on October 10, 2026, with the pricing model stated and no amount printed; it is the only place in this book where such a tool is named.

Table 12.5 — Listening and review monitoring tools: examples of the function, read on October 10, 2026

WHO DOES THIS The administrator chooses and keeps the tools, the owner approves, and the outside lawyer reads the terms

The roles are the case's, and they are posed by this book. The administrator reads the vendor pages, keeps the shared spreadsheet and the log, and proposes a choice with its reason and its date. Sophie Nadeau, the owner, approves every choice and every purchase, and decides whether the study needs the tool at all. The outside lawyer reads the contract terms and answers the questions of privacy and use that a tool raises, which belong to Book H2 (Privacy and Personal Information). The fieldwork service runs the survey on its own platform, outside the application. In a solo business the owner does all of it, and in a larger business a person holds the stack for the team, which is Book K1's subject (The SME Technology Stack); both are this book's suggestions and not facts of the model. No hour is promised to any reader.

WHAT CHANGED On October 10, 2026, the vendor pages read showed an acquisition, an absorbed product and a renamed service

On the day the pages of this chapter were read, one page announced that a larger company had completed the acquisition of a search and marketing platform, another that a social listening vendor had absorbed another product and renamed a third, and another that a qualitative analysis tool had been acquired by another company. These statuses are the vendors' own words on that day, and they may have changed since. This is the reason every row of every table in this chapter carries the date on which it was read, and the reason no row is read as a permanent fact about a product. The state of La Machine in this chapter is that of the application copy of September 7, 2026, read on October 10, 2026 (La Machine, 2026). The dates of sources are never dates of the case, which is counted in days.

12.6 Competitor and Website Monitoring Tools

The fifth family watches a page or a source for changes and rings when one occurs. What it brings is a reminder: a page that changed, with the date and often a saved copy of what it said before. The weekly round of the case takes 30 minutes and is read by hand, and a business with more competitors or more pages will find that the round grows.

What the family cannot do is say what the change means. A tool reports that a page changed; whether the change matters is the judgment that Chapter 7 describes. Book K7 (Web Scraping, Data Enrichment and Market Signals) treats the sources, collection, enrichment and legal boundaries of public market signals. A tool that logs in to a competitor's system, posts as a customer, poses as a candidate or collects what is not public crosses a line that this book calls and never rewrites: agreements between competitors and the exchange of competitively sensitive information belong to Book H4 (Consumer Protection, Advertising Law, Competition and Language Requirements), and trade secrets and the obtaining of them by deceit belong to Book H5 (Intellectual Property).

The watch of Chapter 7 reads only what competitors publish. Nobody poses as a customer, a candidate or a buyer, no confidential price, bid, customer list or document is obtained by deceit, and no price, bid, territory or customer is ever discussed or agreed with a competitor. A monitoring tool does not change those rules; it changes only how quickly a public change is noticed. A tool that would require the business to accept terms of use that forbid what the business intends to do is a tool that should not be started, and the reading of those terms is part of the choice described in section 12.10.

The case buys no page-change monitor. The watch reads three competitors (the regional integrator, Competitor B and Competitor C), and the 11 signals in ninety days, the 4 public tenders seen and the 2 won by the regional integrator are posed by this book. The weekly round of 30 minutes and the monthly review of 60 minutes are read by hand, and the log lives in the shared spreadsheet.

The table below presents examples of competitor and website monitoring tools, read on the vendors' public pages on October 10, 2026, with the pricing model stated and no amount printed; it is the only place in this book where such a tool is named.

Table 12.6 — Competitor and website monitoring tools: examples of the function, read on October 10, 2026

TOOLS OF THE TRADE What any monitoring tool must offer, whatever it is called

The functions are what matter, and no brand name and no price belongs on this list. A monitoring tool watches only public pages and logs in nowhere. It shows what changed and when, and it keeps a saved copy of the page as it was. It rings a person on the day the change occurs, and not only on a dashboard that somebody has to remember to open. It lets the team write the reading of a change apart from the fact of the change. It exports the log in a form another tool can read, because a log that cannot leave the tool is a log the business does not own. And it can be paused or stopped without losing the log.

12.7 Data Enrichment and Business Database Tools

The sixth family adds fields to a record or lists companies and contacts. What it brings is a starting list and a field to complete, faster than a directory read by hand.

What the family cannot do is give a count of a market. A database lists what it lists, its coverage is the vendor's own claim, which this book never prints, and a count of rows is a count of rows. The case counts its market by two paths, from the property assessment roll and by walking zones, and its counts of 1,512 buildings in the size band on the roll, 44 of 50 records eligible, 1,331 adjusted, 262 buildings walked in 6 of 30 zones and 1,310 are posed by this book (Chapter 8).

The second thing this family raises is a question of personal information, and the question is Book H2's, as the Québec and Canada note below sets out. Whether a given database or a given enrichment raises it, this book never decides.

The third thing is the rule that the answers of a study never enter the prospecting file. A database of companies is a prospecting tool, and the two files are kept apart by a decision of the business and not by a feature of any tool. Book N4 (CRM, Contacts and Prospect Data) holds the rules of contacts in La Machine, and Book K7 (Web Scraping, Data Enrichment and Market Signals) treats the collection and enrichment of market signals in its chapter 1.

The case buys no database: the market is counted from the roll and by walking zones, and the Business Explorer of La Machine is a prospecting page and never a count of a market, as section 12.8 says.

The table below presents examples of data enrichment and business database tools, read on the vendors' public pages on October 10, 2026, with the pricing model stated and no amount printed and no coverage claim of any vendor; it is the only place in this book where such a tool is named.

Table 12.7 — Data enrichment and business database tools: examples of the function, read on October 10, 2026

QUÉBEC & CANADA NOTE Personal information in a business database is a privacy question under Canadian and Québec law

The Personal Information Protection and Electronic Documents Act, S.C. 2000, c. 5, Schedule 1, clauses 4.2 to 4.5, in Canada, current to September 21, 2026, asks that the purposes of a collection be identified at or before it, that knowledge and consent be obtained, that collection be limited to what the purposes need, and that information be kept only as long as necessary. The Act respecting the protection of personal information in the private sector, CQLR c P-39.1, ss. 13 and 14, in Québec, up to date to August 12, 2026 and reused from the books that print it and not re-read, asks for a clear, free, informed and specific consent before personal information is communicated to a third person. Whether a given database or a given enrichment raises the question is Book H2's (Privacy and Personal Information), and this book never decides it. No amount of a penalty is printed. A reader with a question is sent to Book H2 and to a lawyer.

12.8 La Machine: Forms, Companies and Alerts in Their Place

La Machine is one worked example of a business platform, described and never sold, and it is not the tool of the case's study. What exists for the work of this chapter is four things. Marketing → Forms business/marketing/forms builds custom forms whose page describes them as going beyond simple email opt-ins, naming discovery calls, intake and surveys among their uses; the fields are text, email, telephone, number, date, list, long text, checkbox and call to action, and there is no rating scale field. Every submission requires an email address or a telephone number and finds or creates a contact, so an anonymous survey is not possible as such. CRM → Companies business/marketing/companies holds one record per company, with its sector, its size, its site and notes, and there is no field for a competitor's price, position or strengths, so a competitor profile is kept as dated notes on a company record. Reseller → Business Explorer business/marketing/business-explorer, reserved to the administrator and reseller roles, searches public business listings, stages the listings in a transit table with an industry classification code and a source, and exports the listings or converts the listings into companies with the prospect status; it is never a market database and never a count of a market, and Book B1 (Marketing Fundamentals for Business Owners), chapter 3, sends the reader here. Productivity → Alerts productivity/alerts rings at a chosen time or on an event of the product, and Productivity → Tasks productivity/tasks holds the task and its verification criterion, which is where the weekly round of a watch can live as a recurring reminder and a dated task.

What is missing belongs in the same breath, in full sentences and without blaming anyone. There is no study or research module. There is no sampling, no quota, no weighting and no margin of error. There is no rating scale field. There is no panel and no recruitment of participants. There is no recording, transcription or coding of interviews. There is no listening to social networks and no reading of competitors' reviews. There is no monitoring of competitors' pages and no alert on public tenders. There is no market data, no market count and no forecast. Check it in your installation.

The table below sets the seven functions of Figure 12.1 beside what La Machine offers for each and where to do the rest. Read it as a map of one platform and not as a comparison.

Table 12.8 — What La Machine offers for each function of this chapter, and where to do the rest

The case runs its survey outside the application, on the fieldwork service's own platform, and the answers stay in the research file. The three competitors are kept as company records with dated notes, the weekly watch is held as an alert and a task, and the counts, the survey tables and the signal log live in a shared spreadsheet.

IN LA MACHINE Forms hold a questionnaire, companies hold competitors, alerts remind, and no page runs a study

Marketing → Forms business/marketing/forms builds custom forms with text, email, telephone, number, date, list, long text, checkbox and call to action fields, with no rating scale, and every submission finds or creates a contact. CRM → Companies business/marketing/companies holds a company with notes. Reseller → Business Explorer business/marketing/business-explorer, reserved to the administrator and reseller roles, stages public business listings for prospecting and is never a count of a market. Productivity → Alerts productivity/alerts and Productivity → Tasks productivity/tasks hold the reminder and the task. La Machine draws no sample, computes no margin of error, offers no rating scale, records or transcribes no interview, listens to no social network, monitors no competitor and holds no market data. Check it in your installation. La Machine is described and never sold.

12.9 Spreadsheets, Statistics and Visualization

The seventh family is the one that sits under all the others. A spreadsheet holds the counts, a statistics tool computes the interval, a visualization tool draws the chart, and a log keeps the record. What the family brings is one place for the numbers and a formula that can be checked: a reader who doubts a total can open the cell and see how it was made.

What the family cannot do is give a margin for a sample that has no margin. A margin of error is printed only for a probability sample, with its formula, its confidence level and its population, and no margin, no confidence interval and no other measure of precision is printed for a non-probability sample, an opt-in panel, an open link, a focus group, a set of reviews or a set of interviews. A spreadsheet will compute whatever it is asked to compute, including a margin for a sample that does not support one, and the tool will not object. The discipline is in the person, and the discipline is the one the council's page states: for a non-probability sample the report says that the sample is non-probability, and a measure of precision is allowed only if it is defined and supported by a detailed description of the model, its validated assumptions and its calculation, as section 5.3 sets out; the page binds the council's members as self-regulation and is never a law.

The second thing the family cannot do is keep a chart honest. An axis that does not start at zero, a truncated scale or a line drawn through three points can mislead as easily as a sentence, so the chart is read by somebody who did not draw it, and the number behind it is written beside it.

The spreadsheet of the case holds the counts of Chapter 8, the survey tables of Chapter 5 and the signal log of Chapter 7, all posed by this book. The formulas are told in words and never as symbols: the adjusted count is the raw count multiplied by the share of records found eligible, and the sample size is the size for a large population divided by one plus that size over the population, as section 5.4 gives it from the statistics agency's manual.

The case uses a shared spreadsheet, and the choice is posed by this book; the license and the export of a free or a paid tool are read as section 12.10 sets out.

The table below is the only place in this book where a spreadsheet, statistics or visualization tool is named, and its rows are examples read on the vendors' public pages on October 10, 2026. The pricing model leaves every amount to the vendor's page, and the table is not a recommendation.

Table 12.9 — Spreadsheets, statistics and visualization tools: examples of the function, read on October 10, 2026

12.10 Choosing Research Tools, and the Pricing and Contract Terms Behind Them

Choosing a tool is a small decision that is easy to make badly, and the five steps of the method below are the whole of the book's advice. Read first what the business already has, because the money saved on a tool is available for the fieldwork, the interviews or the time to read. Write the burden in one sentence, name the function, read the pricing model and the contract before any trial, because the trial is where the terms are accepted, and write the choice with the date of its review. Book K1 (The SME Technology Stack) holds the stack of a small business and is cited here and not redone.

The pricing models are described here and never priced, and the price per complete of the case's fieldwork service is posed by this book.

What to read in a contract is a short list and a long habit. The term and the renewal, because a subscription that renews automatically is a decision taken by default. The data and their export, because a tool that holds the answers and will not release them has taken the study. The exit, because a business should know before it starts what it must do to stop. Who owns the answers and the transcripts, because the material of a study belongs to the business that ran it. Where the data are stored, a question for Book H2 (Privacy and Personal Information) and Book K3 (Cybersecurity, Backups and Business Continuity), each cited here in one sentence. And the vendor's rights over what is uploaded, because a clause that lets a vendor use the material of a study for its own purposes is worth reading twice.

The table below sets six pricing models beside what each means, what to read in the contract and a question to ask. It is the book's reasoning and not a finding, and no cell carries an amount.

Table 12.10 — Pricing models and contract terms, and what to ask

The case buys no tool, and that is posed by this book: its study runs on a fieldwork service, a shared spreadsheet and hand-kept notes, at the costs that the box below sets out. Nothing is credited to the study, and the decision rule decides.

METHOD Choosing a tool for the study in five steps

First, read what the business already has — a spreadsheet, a form, a calendar — and what the work really needs, because the tool the business needs may be the one it is already using. Second, name the function and write the one part of the work that is a burden, in a sentence. Third, read the vendor's pricing model and its contract terms before any trial, on the day, and note the date. Fourth, try the tool on one real, small piece of the study and keep the export, because a tool that cannot export the piece cannot hold the study. Fifth, write the choice with the date on which it will be reviewed, so that a tool chosen for one study is not carried into the next by inertia. This is the book's reasoning and not a tested procedure.

DO THE MATH What the case's study cost, and what a tool would have to replace

Illustrative figures; the case is fictional. Step 1 — The internal cost of the study is $1,603, which is the administrator's 38 hours at $35, the lead technician's 4 hours at $42 and the dispatcher's 3 hours valued at the administrator's rate. Step 2 — The cash cost is $2,300, which is the fieldwork service's 92 completes at $25 each. Step 3 — The two add to $3,903, and the owner's 22 hours are counted and not priced, as Book B5 does. Step 4 — The revenue at stake is $162,279, the last year's 19 signings at about $8,541 a year, and the study is 2.4 percent of it, rounded to the nearest tenth. Step 5 — A tool can be weighed only against the part of the $3,903 that it would replace, at a price read on the vendor's page that day, and this book prints none. The figures are posed by this book, are not a norm, and credit no contract or sale to the study.

CASE IN POINT Nordec Entretien inc. runs a study on a fieldwork service, a spreadsheet and hand-kept notes

Nordec Entretien inc. is a fictional business created for this encyclopedia; its owner, its employees, its customers, its competitors, its figures and the events described are invented, and any resemblance to a real business is coincidental. The survey is run by the fieldwork service on its own platform outside the application, and the answers come back as a table. The 14 interviews are held by the owner and the lead technician and written up by hand in notes of 15 minutes each, with no recording and no transcription tool. The 40 public reviews are read and coded by hand. No listening tool, no review monitor, no page-change monitor and no business database is bought. All of it is posed by this book and illustrative, and no result of the study is credited to any of it.

Chapter 13 gives the plan, a thirty-day first market study in three slices of ten days with gates on days 10, 20 and 30, and the tools of this chapter enter that plan only as choices with dates.

Costly Mistakes

• Choosing the tool before naming the function. A business that starts from a product page starts from whatever that product does well, and discovers its question afterwards.
• Reading a tool table as a ranking. A row is an example of a function, read on a vendor's public page on a day, and a table of examples is not a comparison of quality.
• Taking a displayed price for the cost of a tool. A price is a unit, a currency and a billing period on a date, and the cost of a tool adds the seats, the volume, the export and the hours of the person who runs it.
• Letting a link stand in for a sample. A link posted in a tool is an open-link survey, no margin of error is printed for it, and a sample is made by a frame and a random draw.
• Using a transcript or a theme that no person has checked. A quotation that nobody has read against the notes is the tool's text and not the participant's words.
• Reading a relative index or a keyword estimate as a volume. An index shows relative popularity and an estimate is a range, and neither is a count of searchers, buyers or customers.
• Accepting a trial before reading the contract. The term, the renewal, the export, the exit and the ownership of the answers are decided at the moment the trial is accepted.

Key Takeaways

• A tool is a function before it is a product, and the seven functions around a market study and a watch are surveys and forms, interviews and transcription, qualitative analysis, search and web analytics, listening and reviews, competitor and website monitoring, and enrichment and business data.
• No tool decides the research question, draws a valid sample, reads a signal or writes a finding; a tool holds, counts or shows, and the judgment stays with the people.
• The tables of this chapter name products only in their rows, read on the vendors' public pages on October 10, 2026, with a pricing model stated and no amount printed, and a name in a row is an example and never a recommendation.
• A relative search-interest index and a keyword estimate are never read as a volume, a demand or a market share.
• A business reads first what it already has, names the one part of the work that is a burden, reads the pricing model and the contract before any trial, and writes the choice with the date on which it will be reviewed.
• La Machine can host a simple questionnaire as a form, keep competitors as company records with dated notes and hold a recurring reminder, and it draws no sample, computes no margin of error and holds no market data.

Your Action Plan

1. List what the business already has before opening any vendor page. Write down the spreadsheet, the form, the calendar and the folder of notes already in use, and say for each of the seven functions whether they can carry it. Deliverable: one page listing the seven functions and the existing tool beside each. Verification: every function has an entry, and at least one entry names something the business already owns.
2. Write the one part of the work that is a burden. Name the function and the piece of work that is slowing the study, in one sentence. Deliverable: one sentence naming a function and a burden. Verification: the sentence names a function from the seven and a piece of work, not a product.
3. Read the pricing model and the contract before any trial. Open the pricing page and the terms on the same day, note the date, and write down the term, the renewal, the export, the exit and who owns the answers and the transcripts. Deliverable: one dated page of contract notes per tool considered. Verification: the page names the pricing model without an amount and answers all five questions, or records that the vendor does not answer one.
4. Try the tool on one small piece of the study and keep the export. Use it on a single questionnaire, transcript or page, and export the result in a form another tool can read. Deliverable: one exported file. Verification: the export opens outside the tool and holds the same content as the tool shows.
5. Write the choice with the date of its review. Record the function served, what the tool replaces, its pricing model and the day the choice will be looked at again. Deliverable: one dated line in the research file for each tool chosen. Verification: every tool in use has a dated line.

In La Machine

The application copy read for this book offers a form, a company record, the Business Explorer and a recurring alert with a task, described in section 12.8 and in the box above, and the case uses none of them for its study (La Machine, 2026). Their routes are Marketing → Forms (business/marketing/forms), CRM → Companies (business/marketing/companies), Reseller → Business Explorer (business/marketing/business-explorer), Productivity → Alerts (productivity/alerts) and Productivity → Tasks (productivity/tasks). Check it in your installation.

What a reader needs elsewhere is a survey platform or a fieldwork service for the questionnaire, a spreadsheet for the counts, the survey tables and the signal log, a transcript checked by a person, a source of public counts, and a lawyer for the questions of privacy and use that a tool raises, which belong to Book H2 (Privacy and Personal Information). La Machine draws no sample, computes no margin of error, offers no rating scale, records or transcribes no interview, listens to no social network, monitors no competitor and holds no market data. The state of the application copy is that of September 7, 2026, read on October 10, 2026.

Self-Check

1. What are the seven functions that tools serve around a market study and a watch, and what does each one fail to settle?
2. Why is a link posted in a survey tool not a sample, and what does the council's page say about an open-link survey?
3. What does a transcription tool produce, and what must a person do before a quotation from it is used?
4. Why is a relative search-interest index never read as a volume, and what does the provider's help page say about its data?
5. What are the five steps of choosing a tool, and what are the five things to read in a contract?

Answers. 1. The seven functions are surveys and forms, interviews and transcription, qualitative analysis, search and web analytics, listening and reviews, competitor and website monitoring, and enrichment and business data; none of them decides the research question, draws a valid sample, reads a signal or writes a finding. 2. A sample is made by a frame and a random draw, and a tool delivers a questionnaire to whoever receives the link; the council's page lists the limits of an open-link survey, bias, duplicate and fraudulent answers, manipulation and a lack of representativeness, and asks that they be disclosed (section 5.3). 3. A transcription tool produces a draft, and a person checks every quotation against the notes before it is used. 4. The provider's help page says that the data are a largely unfiltered sample of actual search requests, scaled from 0 to 100 so that the values show relative popularity and not search counts (Google, n.d.-a). 5. The five steps are to read what the business already has, to name the function and the burden, to read the pricing model and the contract before any trial, to try the tool on one small piece and keep the export, and to write the choice with its review date; the five things to read in a contract are the term and the renewal, the data and their export, the exit, who owns the answers and the transcripts, and where they are stored.

Chapter References

Act respecting the protection of personal information in the private sector, CQLR c P-39.1, ss. 13 and 14 (Québec). LégisQuébec, up to date to August 12, 2026 — read October 9, 2026. https://www.legisquebec.gouv.qc.ca/en/document/cs/P-39.1

Google. (n.d.-a). FAQ about Google Trends data. Trends Help. Retrieved October 10, 2026, from https://support.google.com/trends/answer/4365533

Google. (n.d.-b). Use Keyword Planner. Google Ads Help. Retrieved October 10, 2026, from https://support.google.com/google-ads/answer/7337243

La Machine. (2026). Documentation — route reference inventory. Internal product documentation, read October 10, 2026, together with the code of the pages this book describes: business/marketing/forms, business/marketing/companies, business/marketing/business-explorer, productivity/alerts and productivity/tasks.

Personal Information Protection and Electronic Documents Act, S.C. 2000, c. 5, Schedule 1, clauses 4.2 to 4.5 (Canada). Justice Laws Website, current to September 21, 2026 (last amended March 4, 2025) — read October 10, 2026. https://laws-lois.justice.gc.ca/eng/acts/P-8.6/page-7.html

Chapter 13 — Practical Advice, Pitfalls and Your Action Plan

A first market study is a plan before it is a report and a practice before it is a plan, so this chapter says what matters most in doing research honestly, names the ten costliest mistakes, lays out a first study in thirty days in three slices of ten days with gates on days 10, 20 and 30, reads the work of each slice, says which few readings show research working and how to keep the practice alive, and then applies all of it to the thirty days and to the day 365 of Nordec Entretien inc.

What You'll Be Able to Do

• State the seven things that decide whether a study is honest, and name the chapter of this book that owns each one.
• Recognize the ten most costly mistakes of a first market study and say, for each, what to do instead and which chapter answers it.
• Run a first market study in thirty days, in three slices of ten days, with gates on days 10, 20 and 30 and a missed gate tried again 3 days later.
• Read the work of each slice — the decision, the question and the desk study; the interviews, the survey and the signals; the synthesis, the decision brief and the watch — and check it against its gate.
• Judge whether the practice is sound by six readings that are signs rather than results, and say what each reading does not show.
• Keep the practice alive after the first study, by holding the routine, naming a person on each task, reading the sources on their public sites and repeating the survey with the same frame and questions.

Why It Matters

A business that has read the twelve chapters before this one knows how to write a decision rule, how to read a property assessment roll, how to hold an interview, how to draw a sample, how to read a search-interest index for what it is, how to watch three competitors without crossing a line, how to count a market by two paths and how to write a decision brief. What it does not yet have is an order. Every one of those pieces is a method, and a method is not a plan: a business can hold fourteen interviews and never write the rule they were meant to test, or draw a frame and never field the survey, or read a year of signals and never take a decision. The gap between knowing the pieces and doing them in the right order, inside a limit of time, with a check at the end of each stretch, is what this chapter closes.

The order matters more than any single method, and it matters for a reason that is easy to state and hard to practice. The decision comes first, the rule is written before the fieldwork, and the brief comes last. A business that reverses the order — that starts with a survey because a survey is what research looks like, or with a tool because a tool was recommended — will collect evidence and then look for a question to hang it on. The evidence will be real, the question will be invented afterward, and the answer will be whatever the business already believed. The thirty-day plan of this chapter is a way of preventing that reversal, and its three gates are the moments at which the reversal is caught.

The practice matters more than the first study, and this is the part that a business discovers only after a year. A first market study is a project: it has a beginning, a middle and an end, and at the end there is a decision brief. A research practice is a routine: a weekly round of the watch, a monthly review of the signals, a quarterly market brief and an annual study, held by named people, on dates fixed in advance. The project produces a decision; the routine produces a business that keeps deciding with evidence rather than with memory. Nordec Entretien inc., the fictional business followed through this book, ends its first year with a decision taken on day 30 and a practice still running on day 365, and the second of those is the more valuable of the two.

There is a third reason the order and the practice matter together, and it is about honesty rather than about efficiency. A study is honest when its limits are visible: when the rule was written before the evidence arrived, when every number carries a source and a date, when a margin appears only on the probability sample, when the watch log is dated, when the brief was read by the person who decides and when the answers are kept apart from the selling list. None of those conditions is a result, and none of them can be checked at the end of a study if the study was run without them. They are checked at the gates, in the middle, while there is still time to correct the work. That is what a gate is for, and it is why this chapter puts the checks inside the plan rather than at the end of it.

In This Chapter

Section 13.1 sets out the seven things that decide whether a study is honest, each with the chapter that owns it. Section 13.2 names the ten most costly mistakes and sets each beside why it costs, what to do instead and the chapter that answers it. Section 13.3 lays out the thirty-day plan in three slices of ten days, with its three gates, its retry of 3 days and its roles. Section 13.4 reads the work of days 1 to 10, the decision, the question and the desk study. Section 13.5 reads the work of days 11 to 20, the interviews, the survey and the signals. Section 13.6 reads the work of days 21 to 30, the synthesis, the decision brief and the watch. Section 13.7 gives the few readings that show research working and says what each one does not show. Section 13.8 says how to keep the practice alive after the first study. Section 13.9 is the case one year later, on day 365 of this book.

13.1 What Matters Most

Seven things decide whether a study is honest, and each belongs to a chapter of this book. The first is that the decision comes first and the decision rule is written before the fieldwork. A study that begins with a method — a survey, a set of interviews, a tool — has no way to know when it is finished, because it has not said what answer would change anything. Chapter 2 owns the question and the rule, and a rule is a sentence with a threshold in it: open the third territory only if its adjusted count is at least 400 buildings in the size band and the regional integrator won fewer than half of the public maintenance tenders seen there. Written before the fieldwork, that sentence decides; written after, it describes what was already decided. The difference is not a matter of style. A rule written in advance can be contradicted by the evidence, and a rule written afterward cannot.

The second is that public sources are read before people are asked. A desk study costs hours rather than dollars, and it answers questions that would otherwise be put to a respondent who has no reason to answer them. Chapter 3 owns the desk sources: the property assessment roll, the enterprise register, the public tenders and the official counts. A business that reads them first asks better questions, wastes fewer conversations and knows which of its assumptions the public record already contradicts. The case's desk study is posed by this book: the roll gives 1,512 buildings in the size band, a check of 50 records finds 44 eligible, and the adjusted count is 1,331.

The third is that a margin is printed only for a probability sample. A margin of error is a statement about sampling error, and sampling error exists only where units were selected at random with a calculable probability. Chapter 5 owns the survey and its frame, and the case prints one margin, for the probability sample of the survey, with its formula, its confidence level and its population. An opt-in panel, an open link, a focus group, a set of reviews and a set of interviews have no margin, and printing one for them is the error this book refuses. A margin printed where none exists gives a number an authority it has not earned.

The fourth is that an index is never a volume, reviews are never a sample and a stated intention is never a sale. A search-interest index is relative, from 0 to 100, and it measures searches rather than demand. Public reviews are written by those who chose to write, and they are evidence of what those writers said rather than a sample of a market. A stated intention overstates, and the case keeps a posed fraction of it. Chapter 6 owns the digital signals and Chapter 8 owns the forecast, and each says plainly what its evidence can carry and what it cannot.

The fifth is that the watch reads only what competitors publish. Chapter 7 owns the watch, and the line runs in every direction at once: nobody poses as a customer, a candidate or a buyer, no confidential price, bid, customer list or document is obtained by deceit, and no price, bid, territory or customer is ever discussed or agreed with a competitor. The watch reads public tenders and awards, job postings, changes to a public price or offer page, a new establishment in the enterprise register and a burst of public reviews, and it reads nothing else. The case's watch records 11 signals in ninety days, and every one of them came from something a competitor published.

The sixth is that the answers given to a research question are kept apart from the selling list. An answer given to a research question is not a lead, and a person who answered a survey is not a person to call with an offer. Chapter 5 owns the survey and Chapter 9 owns the repository, and the separation is a rule of the practice rather than a courtesy: the answers stay in the research file, and nobody is put on a mailing list or called back with an offer because of an answer.

The seventh is that nothing is credited to the study. No contract, sale, signing or share is attributed to the research, an insight or the watch without a comparison group or a comparison period, and the case credits none. The decision rule decides, and the study informs the rule. A business that credits its study with a sale has stopped measuring and started believing, and the belief will survive the next study that contradicts it.

This book describes published rules on research, personal information, competition and copyright, describes practices, sources and a method of decision, and is not legal, tax or accounting advice; it does not replace the advice of a lawyer, an accountant or a research professional; it recommends no vendor, no survey tool, no panel, no research firm, no fieldwork service, no data provider, no platform, no service and no tool; and it promises no sale, no contract, no customer and no result.

DEFINITION Gate, slice, retry, reading and practice, as this book uses the words

Gate is this book's working word for the check, at the end of a slice, that the listed work exists. Slice is ten days of the thirty-day plan. Retry is the second attempt at a missed gate, 3 days later. Reading is a sign that the practice is sound, never a result. Practice is the routine that outlives the first study. No source read for this book defines these words in this sense; the thirty-day plan, its three slices and its gates are the book's own, posed by this book for the case of Nordec Entretien inc., and a business that adopts the plan adopts a working vocabulary rather than a standard. A gate is not a guarantee of quality, and a reading is not a result: the gate checks that the work exists, and the reading says that the practice is being kept. The words are used in this chapter and in the plan of section 13.3, and nowhere else in this book do they carry this sense.

13.2 The Ten Most Costly Mistakes

Ten mistakes are worth naming before a first market study begins, and each is a mistake of order rather than of skill. The first two are starting with the method and not writing the decision rule, and both are answered in Chapter 2. The next three are skipping the desk study, interviewing only happy customers and asking about the future, and they are answered in Chapters 3 and 4. The following three are publishing a margin for an open panel, reading an index as a volume and taking reviews for a sample, and they are answered in Chapters 5 and 6. The last two are crossing the line of the watch and delivering a report without a decision, and they are answered in Chapters 7 and 9.

The table below sets the ten mistakes beside why each one costs, what to do instead and the chapter that owns the answer. It is the book's reasoning rather than a finding, and it prints no frequency of any mistake in any business, because no source read for this book gives one.

Table 13.1 — The ten most costly mistakes

Three pitfalls documented by the application copy read for this book are general lessons, without blame: a form taken for an anonymous survey, the Business Explorer taken for a count of the market, and the workflow register taken for a register of enterprises. The box below sets them out.

WATCH OUT A form taken for an anonymous survey, the Business Explorer taken for a count of the market and the workflow register taken for a register of enterprises are three pitfalls of the code

The application copy read for this book documents three pitfalls, and each is a general lesson rather than a fault of any person. Marketing → Forms (business/marketing/forms) requires an email address or a telephone number on every submission and finds or creates a contact, so a form cannot be an anonymous survey. Reseller → Business Explorer (business/marketing/business-explorer) stages public business listings for prospecting and is never a count of a market. Control center → Workflows (/registre) is the register of automated workflows and is not a register of enterprises, which is read on its public site. The question that catches each one is the same: what does this page actually do, and what does it not do? Check it in your installation, because a page can change.

13.3 Running a First Market Study in Thirty Days

The plan of this chapter runs thirty days in three slices of ten days, with a gate on days 10, 20 and 30, and it is posed by this book. The first slice produces the decision, the research question, the decision rule, the hypotheses, the one-page research brief and the desk study, and it draws the survey frame. The second slice produces the interviews and their codes, the survey fielded and closed, the digital signals read and the watch log opened. The third slice produces the synthesis, the market counted by two paths, the forecast given as a range, the decision brief read by the owner and the weekly watch on the calendar. The three slices are one project with internal checks, because a study that runs thirty days without a check will discover its problems on day 30.

A gate is a list of work that must exist at the end of a slice, checked against a verification criterion rather than against a feeling that the work is done. The definition of done that a gate applies is Book J2's (Processes, SOPs and the Definition of Done), and the management of the project that carries the gates is Book J3's (Project and Task Management). A gate is not a guarantee of quality. It checks that the listed work exists, and it says nothing about whether the work was done well; a gate that is passed is a gate that was checked, not a gate that was declared. A gate that is not met is tried again 3 days later, and the next slice waits, so a missed gate moves the end of the plan by 3 days and no more is promised. A second missed gate moves it by 3 days again.

The roles are the roles of the case, posed by this book. Sophie Nadeau, the owner, approves the one-page research brief at gate 1 and reads the decision brief at gate 3. The administrator runs the desk study, draws the frame, keeps the log and holds the repository. The lead technician and the technicians help with the interviews and supply the technical words. The dispatcher checks the walked zones against the routes. The fieldwork service makes the survey calls, and the outside lawyer reads the script, the invitation and the line of the watch. The hours are the model's, at the rates Book B5 (Content Marketing and the Content Factory) prints: the administrator's 38 hours at $35 an hour, the lead technician's 4 hours at $42 an hour, the dispatcher's 3 hours, and the owner's 22 hours counted and not priced. The internal cost is $1,603, the cash cost paid to the fieldwork service is $2,300, and the total is $3,903, all posed by this book.

Figure 13.1 draws the thirty-day plan on a line from day 1 to day 30, with its three gates on days 10, 20 and 30 and what each stretch produces.

Figure 13.1 — The thirty-day plan and its three gates. Days 1 to 10 produce the decision, the question, the rule, a one-page brief and the desk study; days 11 to 20 the 14 interviews, the survey closed at 92 answers, the search index and reviews, and the watch log; days 21 to 30 the triangulated count, the decision brief and the weekly watch. A missed gate is retried three days later.

The table below sets the three gates beside the day on which each falls, what must be true for the gate to be passed, what happens if the gate is missed and who checks it. The plan is posed by this book, and no result is promised by any of the three rows.

Table 13.2 — The three gates of the thirty-day plan

KEY CONCEPT A first study is a plan of thirty days with three gates, and the decision rule written first decides

The central idea of this chapter is that a first market study is a plan before it is a report. The decision and the decision rule come first, the evidence comes second and the brief comes last, and each slice of ten days ends with a gate that checks that the listed work exists. A gate that is not met is tried again 3 days later, and the next slice waits, so a missed gate moves the end of the plan by 3 days and no more is promised. The thirty days are the book's own plan, posed by this book for the case, and no source read gives a length for a first market study. The case's figures are posed by this book and are illustrative, never typical and never a benchmark. The rule decides; the study informs the rule.

METHOD Passing a gate in five steps

First, list the work that the gate requires, taken from the plan and not from memory. Second, check each item against its verification criterion, which is the field that says how completion is checked, and not against a feeling that the work is done. Third, write what is missing, if anything, and name the person who will do it. Fourth, if the gate is not met, set the retry 3 days later and let the next slice wait; the plan moves by 3 days and no more is promised. Fifth, write the date on which the gate was passed and move on to the next slice. The method is the book's reasoning and not a tested procedure, and the definition of done that a gate applies is Book J2's (Processes, SOPs and the Definition of Done). The days 10, 20 and 30 of the case are posed by this book, and the gate 2 example requires 14 interviews coded, the survey closed at 92 completes, the index and the 40 reviews read and the watch log opened. A gate that is passed is a gate that was checked, not a gate that was declared.

DO THE MATH The thirty days and a missed gate, in five steps

Illustrative figures; the case is fictional. Step 1 — The plan is three slices of 10 days, and 3 × 10 = 30 days. Step 2 — The gates fall at the end of each slice, on days 10, 20 and 30. Step 3 — If gate 1 is missed on day 10, it is tried again 3 days later, on day 13. Step 4 — The next slice waits those 3 days, so the end of the plan moves from day 30 to day 33, and a second missed gate would move it by 3 days again, to day 36. Step 5 — The plan carries no hour of its own; the hours of the case are the model's, the administrator's 38 hours, the lead technician's 4, the dispatcher's 3 and the owner's 22, counted and not priced, at the rates Book B5 (Content Marketing and the Content Factory) prints. The figures are posed by this book, are days and hours and not a norm, and credit no contract or sale to the study. The sums are exact and no rounding is needed.

THE RULE A gate is passed when its listed work exists, and a gate that is not met is tried again 3 days later

This chapter's own rule is that a gate is passed when the work the plan lists for it exists, checked against a verification criterion, and that a slice does not begin before its gate is passed. A gate that is not met is tried again 3 days later, and the next slice waits. The rule applies the definition of done that Book J2 (Processes, SOPs and the Definition of Done) holds, and it is the book's own rule, neither a standard nor a legal duty. The example is gate 2 on day 20, which requires 14 interviews coded, the survey closed at 92 completes, the search-interest index and the 40 public reviews read and the watch log opened. A gate is not a guarantee of quality: it checks that the listed work exists, and it says nothing about whether the work was done well. The plan of thirty days, its three gates and the retry of 3 days are posed by this book for the case of Nordec Entretien inc., and a business that adopts the rule adopts a discipline rather than a promise.

13.4 Days 1 to 10: The Decision, the Question and the Desk Study

The first slice contains no fieldwork at all. Its work is to write the decision, the research question and the decision rule, to write the hypotheses and the one-page research brief, to read the desk sources and to draw the survey frame. Chapter 2 owns the decision, the question and the rule. The decision is a choice the business must make anyway, with or without research: where next year's growth goes, deeper into the two territories or into a third territory next to the eastern one. The question is what the business needs to know in order to make that choice well. The rule is the sentence that converts the answer into a decision, and it is written before any fieldwork begins, because a rule written afterward is a description of what was already decided.

The hypotheses belong to the same chapter, and they are the part of the brief that the evidence is allowed to contradict. A hypothesis is a statement the study could show to be false, and a brief without one is a brief that cannot be wrong. The one-page research brief carries the decision, the question, the rule, the hypotheses, the sources to be read, the people to be asked, the sample to be drawn, the days and the cost, and it is approved by the owner before the second slice begins. Chapter 2 owns the brief, and its one-page limit is not a stylistic preference: a brief that fits on a page is a brief that the person who decides will actually read.

The desk sources belong to Chapter 3, and they are read in the first slice because they are cheap and because they change the questions that follow. The property assessment roll gives the buildings in the size band, and its owners are left redacted: the roll's license forbids any attempt to re-identify a redacted owner, and the case counts buildings rather than owners. The enterprise register gives the establishments and their registration status. The public tender system gives the calls for tenders and the notices that public bodies publish. The official counts give the industry group and its establishments. None of the four requires a conversation, and together they answer questions that would otherwise be put to a respondent who has no reason to answer them.

The survey frame belongs to Chapter 5, and it is drawn in the first slice because the frame is a desk product rather than a fieldwork product. The frame is the list of units from which the sample will be drawn, and its quality decides the quality of everything the survey can say. The case's frame is posed by this book: 1,259 buildings not under contract, of which 1,050 have a reachable decision-maker, a coverage of 83.4%. The frame is drawn from the roll and the register, and it is checked against them before the sample is drawn, because a frame with omissions, erroneous inclusions, duplications or misclassifications will produce a sample that carries those faults into every estimate.

Gate 1 falls on day 10, and it requires that the decision, the question, the rule and the hypotheses are written, that the one-page brief is approved by Sophie Nadeau, that the desk study is read and that the frame is drawn. The case's decision is where next year's growth goes, the rule is the case's own, the brief is approved by Sophie Nadeau, and the frame is 1,259 buildings, all posed by this book. If the gate is missed, it is tried again 3 days later, on day 13, and the second slice waits.

13.5 Days 11 to 20: Interviews, Survey and Signals

The second slice is the one that talks to people. Its work is to hold and code the interviews, to field the survey through the fieldwork service, to read the search-interest index and the public reviews for what they are, and to open the watch log on the three competitors. Chapter 4 owns the interviews. Consent is given before any recording starts and is noted, the purpose is said before any question, the participant may stop at any time, and the answers serve the research only. The case's 14 conversations are posed by this book: 6 property managers, 4 owners who run their own building, 2 former customers and 2 customers who renewed, 45 minutes each plus 15 minutes of notes. The new codes by interview are 9, 6, 4, 3, 3, 2, 1, 1, 1, 0, 0, 0, 0, 0, the last new code appears at interview 9, and there are 30 codes in all. The themes are proof of response time, one contact rather than a bundle, planning equipment replacement, and price only after proof.

The survey belongs to Chapter 5, and it is fielded through the fieldwork service rather than by the business. The sample is drawn from the frame, and the call offers and sells nothing: a call that pitches a service is telemarketing rather than research, and the fieldwork service says who is calling and for whom. The case's survey is posed by this book: a frame of 1,259 buildings, 90 completes needed for a margin of plus or minus 10 points at 95% confidence, 360 numbers drawn at random, 92 completes, a response of 25.6%, at $25 per complete paid to the fieldwork service, which is $2,300. The prompted awareness is 35 of 92, which is 38% with a margin of plus or minus 9.5 points, and the two intervals overlap with Book B3's (Branding and Visual Identity) measure of 19 of 60, which is 31.7% with a margin of about plus or minus 11.8 points, so the rise is not shown. The margin is printed once, for the probability sample, with its formula, its confidence level and its population.

The digital signals belong to Chapter 6, and they are read for what they are. The search-interest index is relative, from 0 to 100, and it peaks in months 4 and 10; it is a shape rather than a volume, and no search volume, demand or market share is read from it. The 40 public reviews of the three competitors are read and coded as statements rather than as a sample: response time appears 14 times, billing surprises 9 times, the same technician each visit 8 times, and other themes 9 times. The reviews are written by those who chose to write, and the count of themes is a count of statements rather than a measure of a market.

The watch log belongs to Chapter 7, and it is opened in the second slice on the three competitors: the regional integrator, Competitor B and Competitor C. The log records 11 signals in ninety days, posed by this book: public tenders and awards 4, job postings 3, changes to a public price or offer page 2, a new establishment in the enterprise register 1, and a burst of public reviews 1. Every one of the 11 came from something a competitor published, and the log holds the date of each. Gate 2 falls on day 20, and it requires the 14 interviews coded, the survey closed at 92 completes, the index and the 40 reviews read and the log opened. If the gate is missed, it is tried again 3 days later, and the third slice waits. The answers stay in the research file throughout, and nobody is put on a mailing list or called back with an offer because of an answer.

WHO DOES THIS The owner decides at each gate, the administrator runs the work, and the outside lawyer reads the line

At Nordec Entretien inc., the roles are posed by this book. Sophie Nadeau, the owner, approves the one-page research brief at gate 1, joins the interviews, and reads the decision brief at gate 3. The administrator runs the desk study, draws the survey frame, keeps the watch log and holds the research repository. The lead technician and the technicians help with the interviews and supply the technical words. The dispatcher checks the walked zones against the routes. The fieldwork service makes the survey calls, and the outside lawyer reads the script, the invitation and the line of the watch. In a solo business the owner does all of it inside the same thirty days, which is the book's suggestion and not a figure of the model. In a larger business a person runs the plan for the team, which is also the book's suggestion and not a figure of the model. No hour is promised to a reader, and no role is named beyond the roles the case gives.

13.6 Days 21 to 30: Synthesis, the Decision Brief and the Watch

The third slice turns the evidence into a decision, and it is the slice in which the discipline of the earlier chapters is either kept or lost. Its work is to write the synthesis with findings, interpretations and recommendation kept apart, to count the market by two paths, to give the forecast as a range, to write the decision brief and have the owner read it, and to put the weekly watch on the calendar. Chapter 9 owns the synthesis and the brief. A finding is what the evidence shows, an interpretation is what the business makes of it, and a recommendation is what the business proposes to do; keeping the three apart is what allows a reader to disagree with the recommendation without disputing the finding, and it is what allows the same evidence to be read again a year later without confusion.

The market count belongs to Chapter 8, and it is done by two paths rather than one. The top-down path counts from the roll, and the bottom-up path counts by walking. The case's counts are posed by this book: the roll gives 1,512 buildings in the size band, a check of 50 records finds 44 eligible, which is 88%, and the adjusted count is 1,331; walking 6 of 30 commercial zones counts 262 buildings, which gives 1,310. The range is 1,310 to 1,331, and Book B4's (The Marketing Plan, Budget and KPIs) figure of about 1,400 sits between the adjusted counts and the raw roll. The third territory is 470 raw and 414 adjusted, and the rule's threshold is at least 400 buildings in the size band.

The forecast belongs to Chapter 8 as well, and it is given as a range rather than as a number. The case's forecast is posed by this book: a trend of 19 signings, a funnel of 38 proposals at 54.3%, which is 20.6, and 151 stated switchers kept at one quarter, which is 38, giving a range of 19 to 23 signings with a central value of 21. The stated switchers are kept at a posed fraction because a stated intention overstates, and the fraction is the case's own rather than a rule drawn from any study. The forecast is a range because a range is what the evidence can carry, and a single number would claim a precision the evidence does not have.

The decision brief belongs to Chapter 9, and it is read by the person who decides. The case's brief is read by Sophie Nadeau, and the decision rule decides: the third territory's adjusted count of 414 is at least 400, but the regional integrator won 2 of the 4 public maintenance tenders seen there, which is not fewer than half, so the rule says to deepen the two territories, first among owner-run buildings. The rule decided, and the study informed the rule; nothing is credited to the study. The weekly watch is put on the calendar in the same slice, as recurring alerts and dated tasks that a person follows by hand, and the routine of Chapter 10 owns the weekly round, the monthly review and the quarterly brief. Gate 3 falls on day 30, and it requires the synthesis written, the market counted by two paths, the forecast given as a range, the brief read by Sophie Nadeau and the weekly watch on the calendar. If the gate is missed, it is tried again 3 days later, and the practice begins when the gate is passed.

QUÉBEC & CANADA NOTE Four Canadian lines frame the thirty days, and each belongs to a book of this library

Four Canadian texts frame the thirty days, each called to the book that owns it. The Competition Act, R.S.C. 1985, c. C-34, s. 45(1) and (2), current to September 21, 2026, makes it an offense, with a competitor in respect of a product, to agree to fix the price of its supply, to allocate sales, territories, customers or markets, or to restrict its production or supply; no penalty amount is printed, and Book H4 (Consumer Protection, Advertising Law, Competition and Language Requirements) holds the mechanics. The Criminal Code, R.S.C. 1985, c. C-46, s. 391, current to September 21, 2026, makes it an offense to obtain a trade secret by deceit or other fraudulent means; Book H5 (Intellectual Property) holds the mechanics. The Personal Information Protection and Electronic Documents Act, S.C. 2000, c. 5, Schedule 1, clauses 4.2 to 4.5, current to September 21, 2026, sets out identified purposes, consent, limited collection, limited use and disclosure, and retention only as long as necessary; Book H2 (Privacy and Personal Information) holds the mechanics. The anti-spam legislation of Canada, s. 1(2), defines a commercial electronic message; no provision names surveys, and whether a survey invitation is one is Book H3's (Anti-Spam and Telemarketing Compliance) question, posed and never decided. This book explains the texts; a lawyer answers the questions of a given business.

13.7 The Few Readings That Show Research Working

A business that wants to know whether its research practice is sound does not need a result. A result takes a year and a comparison group, and the case credits none. What a business can read at any moment is a short list of signs, and each sign says something about the practice rather than about the market. The first sign is that the decision rule was written before the fieldwork. It is read in the brief, dated, and it is the sign that the study had a question before it had evidence. The second sign is that every number has a source and a date. It is read in the synthesis, where each figure carries the source it came from and the date it was read, and a figure without a source is a figure that has been remembered rather than checked.

The third sign is that a margin appears only on the probability sample. It is read in the survey report, where one margin is printed, with its formula, its confidence level and its population, and no margin appears for the interviews, the reviews or the panel. The fourth sign is that the watch log is dated. It is read in the log itself, where each of the 11 signals carries the day it was seen and the public page it came from, and a log without dates is a list of impressions. The fifth sign is that the brief was read by the person who decides. It is read in the repository, where the brief carries the date it was read and the decision it produced, and a brief that was sent rather than read is a document rather than a decision. The sixth sign is that the answers given to a research question are kept apart from the selling list. It is read in the research file, where the answers stay, and in the file the business uses for outreach, which holds no one because of an answer.

None of the six is a result, and the table below sets each one beside what it shows, what it does not show and where it is kept. The table is the book's reasoning rather than a finding, and nothing is credited to the study by any of its six rows.

Table 13.3 — Readings that show research working, and what they do not show

No sale, contract or share is read as a sign of working, because nothing is credited to the study. The case's rule decided, the log is dated and the margin is printed once, for the survey, and those three facts are the whole of what the case can say about its own practice.

13.8 Keeping the Research Practice Alive

The practice survives if the routine of Chapter 10 survives, and the routine is small enough to survive. The weekly round of the watch takes 30 minutes, the monthly review of the signals takes 60 minutes, and the quarterly market brief takes the time a brief takes; the case's watch costs about 9.5 hours a quarter, posed by this book. The practice also survives if a name is on each task, because a task without a name is a task that will be done by whoever notices it, which is to say by nobody. It survives if the sources are read on their public sites rather than from a saved copy, because a register, a roll and a tender system change, and a copy ages without saying so. And it survives if the survey is repeated with the same frame and the same questions, because a second measurement is comparable with the first only when the instrument has not changed.

A quarter may be skipped, and the remedy is not to double the next one. The remedy is to do it the next week, write it down, and find the cause. A skipped quarter may have one of three causes: the person who runs it was away, the work was not on a calendar, or the task had no verification criterion and so was never closed. Each cause has a small answer, and the answer is worth finding because a practice that is skipped twice in a row is a practice that has stopped. The same is true of the annual study: a year that passes without one is a year in which the business is deciding from memory, and the memory will be of the last study rather than of the market.

When to ask for help is Chapter 11's subject, and the short answer is that help is worth buying for the work that is technical, the work that is heavy and the work that carries a legal line. A survey with a probability sample is technical, a set of interviews across a region is heavy, and the line of the watch is legal. When to change a tool is Chapter 12's subject, and the short answer is that a tool is changed when a function is missing rather than when a newer one appears. Three books lie outside this book: Book E4 (Consultative Selling, Discovery and Demos) treats discovery, demonstrations and value quantification; Book L2 (Product Development and Innovation Management) treats product discovery, prioritization and release planning; and Book L3 (New Markets and Internationalization) treats market selection, entry modes and localization.

The weekly round, the monthly review and the quarterly brief are held as recurring alerts and dated tasks that a person follows by hand, as the box below sets out, and the task format is Book N2's (Piloting Tasks, Executions and Projects).

TOOLS OF THE TRADE What any plan tracker must offer, whatever it is called

A plan tracker is judged by the functions it offers, not by the name on the box, and no brand belongs on this list. It must hold a task with a date, a person and a verification criterion, so that completion is checked rather than assumed. It must group tasks into a gate and show, at a glance, what is missing from the group. It must send a reminder that reaches the person on the day, not the day after. It must hold a place to write the retry and the date to which the retry moved the plan. It must keep a history of what was done and when, so that the annual reading has evidence rather than memory. It must link to the files of the study, the brief, the frame and the log, so that the work and its record sit together. It must export in a form another tool can read, so that the business is never locked into one product. A tracker that offers these seven functions is enough, and a tracker that offers them without a price is not thereby worse.

IN LA MACHINE Alerts and tasks hold the watch, and three pages are not what they seem

Productivity → Alerts (productivity/alerts) rings at a chosen time or on an event of the product, and Productivity → Tasks (productivity/tasks) holds the task and its verification criterion, so the weekly round, the monthly review and the quarterly brief can live as recurring reminders and dated tasks that a person follows by hand; nothing collects a signal by itself (La Machine, 2026). The three pages of the warning above are the ones to read before use. La Machine has no study module and no plan of thirty days, draws no sample, offers no anonymous survey and counts no market, and the case holds its weekly watch in these two pages only as reminders. Check it in your installation, because a page can change. La Machine is described and never sold.

13.9 The Case, One Year Later

Day 365 of this book is entirely posed by this book, and it is counted in days rather than in calendar years or months. On day 365 the weekly watch has been kept 49 weeks, 4 quarterly market briefs have been written, the survey has been repeated with the same frame and the same questions, and 38 signals are in the log. Nothing is credited to the study, and no contract, sale or share is attributed to the research, the insight or the watch. The second survey prints its own margin, and it is compared with the first only through both margins, because a margin belongs to a probability sample and to the population it measured; a comparison of two margins is a comparison of two intervals rather than a claim that anything rose or fell.

The case begins from Book B4's state, cited as that book's: 72 buildings, 19 signings on 35 proposals and an average contract of about $8,541 a year, all from Book B4 (The Marketing Plan, Budget and KPIs). The case ends here with the decision taken on day 30 and the practice alive on day 365. The decision was to deepen the two territories, first among owner-run buildings, and the rule decided it: the third territory's adjusted count of 414 was at least 400, but the regional integrator won 2 of the 4 public maintenance tenders seen there, which is not fewer than half. The study informed the rule, the rule decided, and the practice is what remains. The costs of the study are posed by this book: internal $1,603, cash $2,300, total $3,903, with the owner's 22 hours counted and not priced, against revenue at stake of $162,279, which makes the study 2.4% of it.

What the year shows is not a result but a habit. A watch that has been kept 49 weeks is a dated habit. Four quarterly briefs are four dated documents that a person can read in sequence, and the survey repeated with the same frame and the same questions is an instrument that can be read twice. The 38 signals in the log are 38 dated observations of things three competitors published, and they are the raw material of the next annual study rather than a verdict on the market. A year of watch does not show that the market is stable, and this chapter does not claim it does; the log holds the signals seen, and nothing more.

The chapter closes the book, and the book closes where it began. A business that has read this far has the whole method: the decision rule, the desk study, the interviews, the survey with its frame, the digital signals, the watch of three competitors, the market count and the forecast, the decision brief, the routine and the plan. What remains is to begin, and the beginning is Chapter 2, where the decision, the question and the rule are written before anything else is done.

CASE IN POINT Nordec Entretien inc. runs its first market study in thirty days and keeps the watch for a year

Nordec Entretien inc. is a fictional business created for this encyclopedia; its owner, its employees, its customers, its competitors, its figures and the events described are invented, and any resemblance to a real business is coincidental. The plan runs thirty days in three slices of ten, with gates on days 10, 20 and 30, and a missed gate is tried again 3 days later. Gate 1 produces the decision, the research question, the decision rule and the hypotheses, the one-page brief approved by Sophie Nadeau, the desk study and the survey frame. Gate 2 produces 14 interviews coded, the survey closed at 92 completes, the search-interest index and the 40 public reviews read and the watch log opened. Gate 3 produces the synthesis, the market counted by two paths, the forecast given as a range, the decision brief read by Sophie Nadeau and the weekly watch on the calendar. The decision rule decides to deepen the two territories, first among owner-run buildings. On day 365 the watch has been kept 49 weeks, 4 quarterly market briefs have been written, the survey has been repeated with the same frame and questions, and 38 signals are in the log. All of it is posed by this book and illustrative, and nothing is credited to the study.

WHAT CHANGED On day 365 the watch has been kept 49 weeks and the survey has been repeated

On day 365 of this book, entirely posed by this book, the weekly watch has been kept 49 weeks, 4 quarterly market briefs have been written, the survey has been repeated with the same frame and the same questions, and 38 signals are in the log. The state at day 1 was Book B4's (The Marketing Plan, Budget and KPIs): 72 buildings, 19 signings on 35 proposals and an average contract of about $8,541 a year, cited as that book's and never recomputed here. What changed is not a result but a practice: the watch is a dated habit, the brief is a quarterly document and the survey is a repeatable instrument. Nothing is credited to the study, and no contract, sale or share is attributed to the research, the insight or the watch. The dates of the sources are never dates of the case, and the case is counted in days and never in calendar years or months. The second survey prints its own margin and is compared with the first only through both margins, because a margin belongs to a probability sample and to the population it measured.

Costly Mistakes

• Starting with the method instead of the decision. A business that chooses a survey, a set of interviews or a tool before it has written the decision and the rule has decided how it will learn before it has decided what it needs to know, and the method will shape the question rather than the question shaping the method.
• Writing the decision rule after the fieldwork. A rule written afterward describes what was already decided, and a study without a threshold cannot be contradicted, because every result can be read as support for what the business already intended.
• Skipping the desk study because it feels like reading rather than research. The public record answers a great many questions for the cost of an afternoon, and a business that skips it pays for the same answers in interviews and survey completes.
• Interviewing only customers who renewed. A set of interviews drawn from customers who stayed will produce themes about satisfaction and will say nothing about why buildings leave.
• Publishing a margin for an open panel, an open link or a set of reviews. A margin is a statement about sampling error, and sampling error exists only where units were selected at random with a calculable probability; a margin printed where none exists gives a number an authority it has not earned.
• Reading an index as a volume, reviews as a sample or a stated intention as a sale. A search-interest index is relative, from 0 to 100; public reviews are written by those who chose to write; and a stated intention overstates, which is why the case keeps a posed fraction of it.
• Delivering a report without a decision. A study that ends with a document rather than with a decision brief has produced reading material, and the reading material will be filed rather than used.

Key Takeaways

• Seven things decide whether a study is honest: the decision comes first and the rule is written before the fieldwork; public sources are read before people are asked; a margin is printed only for a probability sample; an index is never a volume, reviews are never a sample and a stated intention is never a sale; the watch reads only what competitors publish; the answers given to a research question are kept apart from the selling list; and nothing is credited to the study.
• The ten most costly mistakes are mistakes of order rather than of skill, and each one has an answer in a chapter of this book: the rule in Chapter 2, the desk study in Chapter 3, the interviews in Chapter 4, the survey in Chapter 5, the signals in Chapter 6, the watch in Chapter 7, the count and the forecast in Chapter 8 and the brief in Chapter 9.
• A first market study runs thirty days in three slices of ten, with gates on days 10, 20 and 30, and a gate that is not met is tried again 3 days later while the next slice waits; the plan is posed by this book, and no source read gives a length for a first market study.
• A gate is passed when the listed work exists, checked against a verification criterion, and a gate is not a guarantee of quality; the definition of done that a gate applies is Book J2's, and the management of the project that carries it is Book J3's.
• Six readings show that the practice is sound — the rule written first, every number sourced and dated, a margin only on the probability sample, a dated watch log, a brief read by the person who decides, and the answers kept apart from the selling list — and none of the six is a result.
• The practice survives if the routine survives, if a name is on each task, if the sources are read on their public sites and if the survey is repeated with the same frame and questions; a skipped quarter is done the next week, written down, and its cause found.

Your Action Plan

1. Write the decision, the question and the rule before anything else. Put the decision the business must make anyway at the top of a page, write the question the decision needs answered below it, and write the rule as a sentence with a number and a condition in it. Deliverable: one page carrying the decision, the question, the rule and the hypotheses. Verification: the rule names a threshold and a condition, and it is dated before the first fieldwork day.
2. Read the desk sources in the first slice and draw the frame. Read the property assessment roll, the enterprise register, the public tenders and the official counts, leaving the roll's owners redacted, and draw the survey frame from what the desk study gives. Deliverable: a dated desk note and a frame with its coverage stated. Verification: every figure in the desk note carries its source and its date, and the frame states how many units it holds and how many have a reachable decision-maker.
3. Hold the interviews and field the survey in the second slice. Hold the conversations with consent given before any recording and noted, code them as they arrive, and field the survey through the fieldwork service with the sample drawn from the frame and no offer in the call. Deliverable: a coded interview set and a closed survey with its margin, its formula, its confidence level and its population. Verification: the last new code is named with the interview at which it appeared, and the margin appears once, for the probability sample only.
4. Write the synthesis and the decision brief in the third slice, and put the watch on the calendar. Keep findings, interpretations and recommendation apart, count the market by two paths, give the forecast as a range, and write a brief that the person who decides reads and dates. Deliverable: a synthesis, a two-path count, a range forecast and a dated decision brief. Verification: the brief names a decision, a date and a person, and the weekly watch is a recurring reminder and a dated task rather than an intention.
5. Keep the practice alive after the first study. Hold the weekly round, the monthly review and the quarterly brief as recurring alerts and dated tasks, put a name on each task, read the sources on their public sites, and repeat the survey with the same frame and the same questions. Deliverable: a calendar of recurring research work with a named person on each item. Verification: a skipped quarter is done the next week and written down, and the cause of the skip is named rather than repeated.

In La Machine

The recurring work of a research practice can live in two pages of the application copy read for this book, and the study itself cannot (La Machine, 2026). The alerts at Productivity → Alerts (productivity/alerts) and the dated tasks at Productivity → Tasks (productivity/tasks) can carry the weekly round of the watch, the monthly review of the signals and the quarterly market brief as recurring reminders that a person follows by hand, and nothing collects a signal by itself.

What the platform does not do belongs in the same breath, because a business that expects the plan to be built into the software will not build it. La Machine has no study module and no plan of thirty days; it draws no sample, computes no margin of error, offers no rating scale, records and transcribes no interview, listens to no social network, monitors no competitor and holds no market data. Three pages are not what they seem, as the warning of section 13.2 sets out. What the reader needs elsewhere is a plan of thirty days with three gates, a research file, a named person who runs the work and a lawyer for the rules. Check it in your installation, because a page can change. La Machine is described and never sold.

Self-Check

1. What are the seven things that decide whether a study is honest, and which chapter owns each one?
2. Why is starting with the method the first of the ten most costly mistakes, and what does the decision rule prevent?
3. What is a gate, what is it not, and what happens when a gate is not met on its day?
4. What are the six readings that show the practice is sound, and what does each one not show?
5. What does the case show on day 365, and what is credited to the study?

Answers. 1. The decision comes first and the rule is written before the fieldwork (Chapter 2); public sources are read before people are asked (Chapter 3); a margin is printed only for a probability sample (Chapter 5); an index is never a volume, reviews are never a sample and a stated intention is never a sale (Chapters 6 and 8); the watch reads only what competitors publish (Chapter 7); the answers given to a research question are kept apart from the selling list (Chapters 5 and 9); and nothing is credited to the study (Chapter 9). 2. A business that begins by choosing a survey, a set of interviews or a tool has decided how it will learn before it has decided what it needs to know, and the method will shape the question rather than the question shaping the method; the rule prevents that by stating, before the fieldwork, what answer would change the decision, so that the evidence can contradict the business rather than confirm it. 3. A gate is the check, at the end of a slice of ten days, that the listed work exists, checked against a verification criterion; it is not a guarantee of quality, because it says nothing about whether the work was done well; a gate that is not met is tried again 3 days later, and the next slice waits, so the plan moves by 3 days and no more is promised. 4. The rule written before the fieldwork shows that the study had a question before it had evidence, and it does not show that the rule was the right one; every number sourced and dated shows that the figures were checked, and it does not show that the sources are complete; a margin only on the probability sample shows that the precision claimed matches the design, and it says nothing about the interviews or the reviews; a dated watch log shows that the watch was kept, and it does not show that the market is stable; a brief read by the person who decides shows that the decision had a reader and a date, and it does not show that the decision was right; the answers kept apart from the selling list shows that the answers serve the research only, and it does not show that the separation will hold without a rule. 5. On day 365, entirely posed by this book, the weekly watch has been kept 49 weeks, 4 quarterly market briefs have been written, the survey has been repeated with the same frame and the same questions, and 38 signals are in the log; nothing is credited to the study, and no contract, sale or share is attributed to the research, the insight or the watch, because the decision rule decided and the study informed the rule.

Chapter References

An Act to promote the efficiency and adaptability of the Canadian economy by regulating certain activities that discourage reliance on electronic means of carrying out commercial activities, and to amend the Canadian Radio-television and Telecommunications Commission Act, the Competition Act, the Personal Information Protection and Electronic Documents Act and the Telecommunications Act, S.C. 2010, c. 23, ss. 1(1), 1(2), 1(3) and 6(1) (Canada). Justice Laws Website, current to September 21, 2026 (last amended January 1, 2026) — read October 10, 2026. https://laws-lois.justice.gc.ca/eng/acts/E-1.6/page-1.html

Competition Act, R.S.C. 1985, c. C-34, s. 45(1) and (2) (Canada). Justice Laws Website, current to September 21, 2026 (last amended March 26, 2026) — read October 10, 2026. https://laws-lois.justice.gc.ca/eng/acts/C-34/

Criminal Code, R.S.C. 1985, c. C-46, s. 391 (Canada). Justice Laws Website, current to September 21, 2026 (last amended July 18, 2026) — read October 10, 2026. https://laws-lois.justice.gc.ca/eng/acts/C-46/section-391.html

La Machine. (2026). Documentation — route reference inventory. Internal product documentation, read October 10, 2026, together with the code of the pages this book describes: productivity/alerts, productivity/tasks, business/marketing/forms, business/marketing/business-explorer and /registre.

Personal Information Protection and Electronic Documents Act, S.C. 2000, c. 5, Schedule 1, clauses 4.2 to 4.5 (Canada). Justice Laws Website, current to September 21, 2026 (last amended March 4, 2025) — read October 10, 2026. https://laws-lois.justice.gc.ca/eng/acts/P-8.6/page-7.html