What customers can and cannot tell you
Customers are reliable witnesses to their own recent behaviour and poor forecasters of their future behaviour. Asked what they did the last time they bought in the category, where they looked, who they asked, what nearly stopped them, most people give accurate, detailed answers. Asked whether they would buy a product that does not exist yet, or what they would pay for it, most people give polite, optimistic answers that bear little relation to what they later do.
They are also poor at explaining their own motives in the abstract. "Why did you choose us?" tends to produce the reasons people think they should have, such as price or quality, rather than the sequence of events that actually led to the purchase. Good customer research works around this by asking about specific past episodes and reconstructing the story, which is the core of the jobs-to-be-done approach: what was happening, what were they trying to get done, what did they try first, and what made them switch.
Customers cannot tell you how many people share their view; that needs a measured sample. And they cannot tell you how they would behave in a situation they have never faced; that needs an experiment or careful observation. Choosing the method is mostly choosing which of these limits to work around.
Interviews, surveys, observation
Three primary methods cover most customer research. The choice follows the question, as the marketing research hub explains in its overview of the research process.
| Method | Best for | Typical scale | Main weakness |
|---|---|---|---|
| Interviews | Why and how: motives, context, language, decision journeys | A small number of carefully chosen people | Cannot say how common a view is; easy to lead |
| Surveys | How many and how much: prevalence, preference, segment sizes | Enough respondents to read differences between groups | Only answers the questions asked; wording shapes answers |
| Observation | What people actually do, in context or on screen | A few sessions or a stream of behavioural data | Shows behaviour, not reasons; can be intrusive |
Most useful studies combine them. Interviews find what matters and the words customers use; a survey then measures how widespread each finding is; observation or behavioural data checks that stated behaviour matches real behaviour at each FunnelModel of the customer journey stages. stage. None of the three can be replaced by a model answering as a customer; the guide to synthetic personas and simulated respondents sets out the few preparation jobs such answers can do.
Focus groupsA moderated discussion among a small set of people who share an experience, used to hear how people talk about a category together and react to material. and diary studiesA qualitative method in which a small set of people record, at the time it happens, what they do during an episode that unfolds over days or weeks. sit beside interviews. The focus group tutorial covers when a group shows what an interview cannot and how to moderate dominant voices; the diary study tutorial records behaviour over time as it happens rather than months later in an interview.
The core guides
- Customer interviews, step by step — a tutorial for a small interview study: setting the learning goal, deciding who to talk to, recruiting and screening, writing the discussion guide, running and recording sessions with consent, synthesising and sharing the results so they lead to a decision.
- Survey and questionnaire design — the craft of a survey as a measurement instrument: question types and scales, wording and order effects, length and invitations, sampling and weighting, consent and anonymity, and how to analyse and report the results.
Segmentation and personasA research-based profile of a typical customer type: goals, problems and buying behaviour., the strategy outputs built from this evidence, belong to the strategy section. The audience research guide covers the lighter, continuous customer understanding drawn from analytics, CRM data and platform insights; this section covers the deliberate studies that go beyond what existing data can show.
Further guides
Further guides take the groundwork and the methods further: finding and screening the right people and keeping a pool of them, writing a survey from interview findings, drawing and testing it, cleaning and tabulating its answers, coding what respondents write, running groups and diaries, and adapting the work to business buyers:
- Recruiting research participants and writing a screener — a tutorial for recruiting people for interviews, groups and usability sessions: the sample frame, a neutral screener, quotas, invitations and no-shows.
- Building a research participant pool — a tutorial for a standing pool of customers who agreed to research contact: joining routes, what joiners are told, profile fields, a list kept apart from marketing, rotation, participation records and review.
- Turning interview findings into survey questions — a tutorial for carrying interview findings across into a survey: sorting findings by the question they raise, keeping with the interviews what cannot be rewritten as something that happened, writing options in participants' words, ordering the questions so that the options do not prime the answers and tracing each question back to the finding and the output behind it.
- Drawing a survey sample from a customer list — a tutorial for defining the population, cleaning the frame, applying contact rules, stratifying and comparing respondents with the frame.
- Pretesting a questionnaire before it goes out — a tutorial for think-aloud sessions with the target audience, routing and device checks, a soft launch and fixing what each one finds.
- Cleaning and tabulating survey responses — a tutorial for turning a raw export into clean data and tables: exclusion rules written in advance, a cleaning log, bases and the planned comparisons.
- Coding open-text survey answers by hand — a tutorial for building a codeframe from a sample, defining each code, applying it consistently, checking agreement and counting with bases.
- Running a focus group that produces evidence — a tutorial for when a group beats interviews, setting the mix, a guide built for a group, moderating dominant voices, consent and honest analysis.
- Running a diary study — a tutorial for recording behaviour over time as it happens: recruiting, writing prompts, keeping people going, consent and analysis.
- Researching business buyers and buying committees — reaching each role in a buying committee, recruiting without sales bias, confidentiality, incentives and small populations.
Recruiting participants and incentives
Recruitment decides whose voice the research carries, and it is where most customer research quietly goes wrong. The easiest people to reach, loyal customers who answer emails, are rarely the people whose behaviour the business most needs to understand. A study of why trials fail needs people who abandoned a trial; a study of why customers leave needs people who have left.
Recruit from a sample frame that matches the question: customer lists segmented by behaviour, recent churners, lost prospects from the CRM, website intercepts, research panels, or communities where non-customers gather. Screen every candidate with a few short questions that confirm they fit, and keep the screener neutral so it does not reveal the answer that qualifies. The recruitment tutorial works through the frame, quotas, the screener, invitations and no-shows; for a survey drawn from the business's own list, the sampling tutorial cleans the frame, applies contact rules and compares respondents with it. Where the participants are business buyers, the guide to researching business buyers covers reaching each role in a buying committee, recruiting through account owners without sales bias and small populations.
Incentives compensate people for their time and raise participation, especially among the busy and the unengaged. Pay them in a form that does not bias the result: a voucher or a charity donation rather than a discount on the product being researched, which attracts people who like the product. Professional codes, including the ESOMAR code, require that research is not used as a disguise for selling, that participation is voluntary and informed, and that personal data is handled with care. Customer value matters too: a business studying its high-LTVTotal value a customer brings over time. customers should take extra care not to over-survey them.
Synthesis: from notes to insights
Raw research is a pile of transcripts, notes and response tables. Synthesis turns it into a few findings the business can act on. Start by tagging the material against the research questions, then look for patterns that recur across participants rather than vivid stories from one. Separate what people did, what they said, and what the team infers, because mixing the three is how opinions get passed off as evidence. For open-text survey answers, the hand-coding tutorial builds a codeframe from a sample, applies it consistently and counts respondents against the base.
Frame each finding as an observation, the evidence for it, and its implication for the decision. A value proposition canvas is a useful structure for findings about customer jobs, pains and gains, as long as each entry is traced back to the research rather than to the team's assumptions. Where a finding should change a target or a KPIA metric tied directly to an objective., say so explicitly.
A pet insurer combines interviews and a survey
Hollin Pet Cover wants to know why a large share of customers cancel in their second year. The research lead recruits 12 recent cancellers from the CRM, offering a £30 voucher, and runs interviews built around the moment each decided to leave. Three patterns recur: a renewal price rise noticed only at the payment date, a claim experience that felt slow, and a cheaper quote found through a comparison site. A survey of 1,100 cancellers then measures how common each is: 46% cite the renewal price, 21% the claim experience, and 18% a comparison quote. The team changes the renewal reminder timing first and plans a follow-up study on claims.
Common customer research problems
| Problem | Usual cause | Guide |
|---|---|---|
| Interviews produced compliments, not insight | Questions about opinions and the future instead of specific past behaviour | Customer interviews |
| Findings from a handful of loyal fans | Recruiting from whoever answered, not from a frame that matches the question | Customer interviews, Recruiting research participants |
| The same few customers turn up in study after study | Recruitment from whoever replies, with no pool, rotation or record of who took part | Building a research participant pool |
| A survey that confirms what the team believed | Leading wording, missing options, or order effects | Survey design |
| Survey options came from the team's labels, and the interviewees' words never reached the survey | No hand-off from the interview findings to the first draft of questions; every theme became a question or none did | Turning interview findings into survey questions |
| Survey results that cannot be generalised | A self-selected sample with no check on representativeness | Survey design, Drawing a survey sample from a customer list |
| Respondents misread a question once the survey was live | Questionnaire reviewed by colleagues but never pretested with the audience on their own devices | Pretesting a questionnaire |
| A figure changes depending on who cleaned the file, or shares appear without a base | Exclusions decided while reading the results; no cleaning log; bases left off the tables | Cleaning and tabulating survey responses |
| Open-text answers reported through a few vivid quotes | No codeframe; mentions counted instead of respondents, with no base | Coding open-text answers |
| Business-buyer research heard only from friendly accounts | Account owners chose the participants; only the champion interviewed | Researching business buyers |
| Unsure whether to interview or survey | The question type was never matched to the method | Qualitative vs quantitative research |
| Research that never changed a decision | No decision named before fieldwork; findings not framed as implications | Research design and brief |
| Personas nobody uses | Research outputs written as descriptions rather than as segments tied to strategy | Audience research |
When a survey is already in the field with too few answers, or a round of interviews has finished without anything a decision can use, the marketing research troubleshooting section reads the completion funnel or the transcripts first and separates the causes from there.
