Research starts with a decision
The question to ask before anything else is what the business will do differently depending on the result. If the answer is nothing, the study is decoration. If the answer is "we will know more", the study is curiosity, which can be worth funding but should be called what it is. A study worth designing has a decision attached: enter the market or not, lead with this proposition or that one, price at this level or lower, keep the segment or drop it, launch in spring or wait.
Naming the decision does three things. It sets the deadline, because research that arrives after the decision has no value however good it is. It sets the standard of evidence, because a reversible decision with a small budget can be made on a handful of interviews, while an irreversible one with a large budget justifies a representative survey and a longer timeline. And it forces the sponsor to state, before fieldwork, what finding would change their mind. That last point is the one most often skipped and most often regretted. A sponsor who has not said in advance what would make them abandon a plan will find reasons to discount any finding that threatens it.
Write the decision as a sentence with a verb and a date. "We will decide by the end of the quarter whether to extend the subscription to small businesses" is a decision. "Understand the small-business opportunity" is a topic, and a brief built on a topic produces a report, not an answer.
From decision to research questions
A research question is a question the study can answer with evidence, phrased so that the answer bears on the decision. The move from decision to questions is where design skill shows, because the obvious questions are usually the wrong ones. "Would small businesses buy our subscription?" invites a hypothetical answer that people are poor at giving. "How do small businesses currently solve the problem the subscription addresses, what do they pay for it, and what would make them switch?" can be answered by asking about behaviour that has already happened. The gap between stated and revealed preference is the reason: what people say they would do is a weaker guide than what they have already chosen.
| Decision | Weak question | Research question that can be answered |
|---|---|---|
| Extend the product to a new segment | Would they buy it? | What do they use today, what does it cost them, and what has made them switch in the past? |
| Choose between two propositions | Which do they prefer? | Which proposition do people recall and describe accurately after a single exposure, and which objections does each raise? |
| Set a price | What would they pay? | At what price points do people say the product is too cheap to trust, good value, expensive but worth considering, and out of the question? |
| Enter a new market | Is the market attractive? | How many buyers meet our serviceable definition, how is demand moving, and who holds the share we would need to take? |
| Fix a leaking FunnelModel of the customer journey stages. step | Why do people drop off? | What did people who stopped at that step believe was happening, and what did they do next? |
| Keep or cut a channel | Does the channel work? | What happens to enquiries and conversion rate (CVRShare of visits or clicks that convert.) when the channel is withheld from a comparable group? |
The last row is an experiment, and experiments have their own discipline: a hypothesis with a predicted direction, a success metric and a sample large enough to see the difference. The writing hypotheses guide covers the form, and the sample size and significance guide covers the arithmetic. Marketing research borrows both when the question is causal and otherwise leaves them where they are.
Rank the questions. A brief with one primary question and two or three secondary ones gets a study that answers the primary question well. A brief with eight equal questions gets eight shallow answers.
Primary or secondary first
Secondary research is evidence someone else has already gathered: official statistics, regulator reports, trade body data, published surveys, academic work, analyst reports, and the business's own analytics, sales records and previous studies. Primary research is evidence gathered for this question. The order is almost always secondary first, for three reasons. It is cheaper. It narrows the primary question, so the expensive work is aimed at what is genuinely unknown. And it provides the context that makes primary findings interpretable, because a survey result means little without knowing how the market is moving around it.
The discipline in secondary work is judging the source. Who collected the data, how, when, from whom, and why? A trade association's estimate of its own industry's growth was produced by people with an interest in the answer. A free report from a software vendor was produced to generate leads. Official statistics have documented methods but may lag the market by a year or more. None of this makes a source useless; it sets the weight it carries. The market and competitor section covers the judging of sources for sizing and trend questions in more depth.
Secondary research is enough on its own when the decision turns on facts about the market and a credible, recent source has measured them. It is almost never enough when the decision turns on why the business's own customers behave as they do, because nobody else has studied them. Most briefs end up with a short secondary stage that sets the scene and a primary stage that answers what remains. Answers generated by an AI model in place of customers are neither, and the guide to synthetic personas and simulated respondents explains why they can prepare a study but never stand in for its evidence.
Choosing the method family
With the questions ranked and the secondary scan done, the method follows from the kind of answer needed. Questions that begin with why or how, or that the team cannot yet phrase precisely, call for qualitative methods: interviews, groups, observation, open feedback. Questions that begin with how many, how much or which is larger call for quantitative methods: surveys, behavioural data, experiments. Questions about cause call for a controlled comparison, an A/B TestControlled comparison of two variants. or a holdout. The qualitative vs quantitative comparison goes through the choice in detail; the design point here is that most studies with more than one question need more than one family, run in sequence.
Three further questions shape the design within a family. Exploratory, descriptive or causal: are we discovering what matters, measuring how things stand, or testing whether one thing changes another? Each has different standards for sampling and analysis. One-off or tracked: a single waveOne round of fieldwork in a repeated survey or tracker, drawn, worded and coded the same way as earlier rounds so the results can be compared over time. answers the question now; repeated waves show direction, cost more and need a stable instrument. In-house or external: running the study yourself is cheaper and faster for small qualitative work; an agency brings sample access, fieldwork capacity and distance from the sponsor's hopes, which matters most on large or politically charged questions. The guide to commissioning a research agency or freelancer covers what to buy, how to judge the proposals and what the deliverables should contain.
Decide the method last, not first. A brief that starts "we want a survey" has already closed off the possibility that the question is better answered another way. When the decision is about why visitors fail to buy or enquire on the website, the sequence is well worn: the tutorial on running a conversion research sprint starts from the business's own analytics and then points recordings, polls and usability sessions at the step those numbers locate.
Sampling, timing and budget trade-offs
Every design trades the same three things against each other, and the brief should say which it is protecting. Sample is who the study hears from and how well they stand for the population the decision concerns. Timing is when the findings land relative to the decision. Budget is the money and the people's time it consumes. Improving any one usually costs one of the others. A representative sample of a hard-to-reach audience takes longer and costs more than a convenience sample; a study that must report in three weeks cannot wait for slow recruitment; a tight budget buys either depth with a few people or breadth with a short instrument, not both.
The way through is to decide what the decision can tolerate. A reversible decision tolerates a weaker sample, because a wrong call can be corrected. A decision that will set a KPIA metric tied directly to an objective. target for the year, or commit a large budget, cannot. The one-way and two-way door framework sorts decisions by how hard they are to undo, which settles which of the two a given choice is. Timing is rarely negotiable, so it usually sets the method: when the deadline is close, a well-run set of interviews and a short survey of existing customers beat a representative study that arrives late. Budget is more negotiable than teams assume, because the cost of a study is small against the cost of most decisions it informs; the argument to make is the cost of being wrong. When a study is already set to report after its decision, the diagnosis for research taking too long sets the decision date beside the plan stage by stage and separates a missing date, an oversized sequence, recruitment from scratch, late consent work and a small study sent to an agency.
Whatever is traded, write it down. "The sample is drawn from existing customers and will under-represent people who have never bought from us; findings about non-customers are directional only" is a legitimate design. The same sample presented without the caveat is a trap for whoever reads the report next year. When that trap has already closed and a finding is disputed, the diagnosis for survey results that are not representative checks the frame, the respondents, the wording, the groups and the weights to say what the survey can still claim.
Designing a study under a deadline
Kestrel Outdoor, a mid-sized retailer of walking equipment, has eight weeks to decide whether to open a third store in a coastal town. The commercial director's first request is "a survey of the town". The research lead reframes the decision (open by spring or defer a year) and writes three questions: how many households within a thirty-minute drive buy walking equipment at least yearly, where they buy it now, and what would make them switch. Secondary work takes a week: official population data for the drive-time area, footfall counts the council publishes, and the retailer's own online orders, which show 1,840 customers already in the catchment. Primary work is a short online survey sent to those customers (520 responses) and twelve interviews with people recruited outside the two existing competitors. The sample over-represents existing customers, which the brief states plainly; the compensating evidence is the interview recruitment. The board gets a range (an estimated 4,000 to 6,500 buying households, with the retailer's online customers making up roughly 30% of the lower figure) in week six. The whole study costs £3,200 in incentives and panel fees.
Writing the research brief
The brief is the design written down for the people who will execute it. It serves the same purpose as the campaign brief template does for a campaign: it stops good work being done on the wrong problem. A usable brief fits on two pages and covers the following, in this order.
| Section | What it contains |
|---|---|
| Background | The situation in a paragraph: what the business sells, what has changed, why the question has arisen now |
| The decision | One sentence with a verb and a date; who makes it; what finding would change it |
| Research questions | The primary question and up to three secondary ones, ranked |
| What is already known | The secondary evidence reviewed, what it settled and what it left open |
| Audience and sample | Who the study must hear from, who it must exclude, the sample frame and the known gaps |
| Method guidance | The family the business expects, stated as a preference open to challenge, with the reasons |
| Constraints | Deadline, budget, markets, languages, access to customers, anything the study cannot do |
| Ethics and data | Consent approach, incentives, how personal data is handled and for how long, any sensitive topics |
| Deliverables | What arrives and in what form: findings, not a slide count; the decision meeting it feeds |
| Ownership | Who signs off, who is consulted, who answers the researcher's questions during fieldwork |
Two sections deserve a note. Under method guidance, state a preference and invite challenge rather than specifying a technique, because the people executing the study often know a better route to the question. Under ethics and data, the ESOMAR and MRS codes set the professional standard (voluntary and informed participation, no selling under the guise of research, confidentiality, extra care for vulnerable participants) and data protection law adds legal duties wherever personal data is held. The general direction is to collect the minimum, say what will happen to it, separate identities from answers early and delete identifying material when the study closes; the specifics depend on jurisdiction and belong with legal counsel. The tutorial on participant information and consent turns this field into the documents and records a study needs before the first invitation.
The brief is also the record. When the decision plays out, the brief and the findings together show what was known, what was assumed and what was decided, and that record is what makes the next study better than this one. The methods section collects the other design guidance this page draws on. The research brief template turns these sections into a copyable form, with instructions for each field and a complete fill.
Common mistakes
- Starting from a method. "We need a survey" before anyone has written the question the survey would answer.
- Topics instead of questions. A brief that says "understand the youth market" and a report that understands it in no useful direction.
- No threshold for changing course. The sponsor never says what finding would stop the plan, so no finding does.
- Skipping the secondary stage. Paying for primary research into something a regulator or a trade body measured last year.
- Too many questions. A dozen equal questions, each answered thinly.
- Convenience samples presented as representative. The caveat known to the researcher and invisible to the reader.
- Research that arrives after the decision. Timing treated as negotiable when it is the one constraint that is not.
- Ethics as an afterthought. Consent, incentives and data retention worked out during fieldwork rather than in the brief.
