Before you start
- A decision. Interviews should inform something specific: a positioning choice, a messaging brief, the fix for a leaking FunnelModel of the customer journey stages. step, the next feature. If no decision depends on the answer, postpone the study.
- What you already know. Read the existing evidence first. The audience research guide covers the analytics, CRM and platform data that should be reviewed before talking to anyone, so interviews go after what the data cannot explain.
- Confirmation that interviews are the right method. Interviews answer why and how. If the question is how many, the qualitative vs quantitative comparison explains when to survey instead, or to do both in sequence.
- Access to participants. A customer list you are permitted to contact for research, a CRM with lost prospects, or a budget for a recruitment panel.
- Tools. A video-call or meeting tool with recording, a transcription service your organisation has approved, a consent form, and a shared space for notes.
- Time. Interviewing, note-taking and synthesis take longer than the sessions themselves; block time for all three.
Set the learning goal
Write one sentence that names the decision and the thing you need to learn to make it. "Before we rewrite the onboarding emails, learn what new customers expected in their first week and where that expectation broke" is a learning goal. "Understand our customers better" is not. Add up to three specific questions the study must answer, ranked. Then write down what you currently believe the answer is, and what finding would change your mind. That last line keeps the team honest later, when it is tempting to hear only the findings that confirm the plan. Where the decision is a software product's onboarding emails, the tutorial on building onboarding emails shows where the findings go: one email for each step at which new users stall.
Define who to talk to
Decide whose experience answers the question. For onboarding, that is recent sign-ups; for churn, people who left recently; for acquisition, people who considered you and chose someone else. Describe them by behaviour and recency, not by demographics alone: "bought for the first time in the last two months" is a usable criterion, "women aged thirty to forty-five" rarely is. Where groups differ in ways that matter, such as new and long-standing customers or high and low LTVTotal value a customer brings over time. accounts, plan a few interviews per group so you can compare them. Recent behaviour matters because memory fades quickly and people fill the gaps with rationalisations. A study recruited only from current customers falls into survivorship bias, hearing only from those who stayed, so a churn question needs people who left in the sample.
Recruit and screen participants
Choose a source that reaches the people you defined: a segmented customer list, CRM records of lost deals, an invitation on the website or in the product, a research panel, or a community where the audience gathers. Write a short screener of a few questions that confirms each candidate fits, without revealing which answers qualify. Invite more people than you need, because some will not respond and some will cancel. Offer an incentive appropriate to the audience's time, in a form that does not bias the result, such as a voucher rather than a discount on your product. Schedule sessions with gaps between them for notes. The tutorial on recruiting research participants works through the frame, quotas, a screener question by question, exclusions and no-shows in more detail.
Write the discussion guide
The guide is a plan for the conversation, not a script. Open with an easy warm-up, move to the specific episode you want to reconstruct, and spend most of the time there. The jobs-to-be-done approach is a reliable structure: what was happening when the need arose, what they were trying to get done, what they tried first, what they compared, what nearly stopped them, and what tipped the decision. Write open questions that ask about past behaviour ("tell me about the last time…"), and list probes such as "what happened next?" and "why was that important?". Keep hypothetical questions and your own product to the end, if at all. The difference between what buyers say and what their choices reveal is why past episodes make better evidence than predictions. The template table below gives a working structure.
Run the session
Start by explaining the purpose, confirming consent and saying there are no right answers. Then listen. Aim for the participant to talk for most of the session. Ask one question at a time, allow silences, and follow the participant's story rather than the order of the guide when it leads somewhere useful. Ask for specifics: dates, places, the words someone used, the tab they had open. When people generalise ("I always…"), bring them back to a particular occasion. Avoid leading questionsA survey or interview question whose wording contains or suggests its expected answer, producing measurement bias rather than evidence., avoid explaining or defending the product, and do not pitch. Close by asking whether there is anything you should have asked and did not.
Record and take notes with consent
Record only with explicit agreement given before recording starts, and tell participants what is recorded, who will see it, how long it will be kept and how to withdraw. Data protection law treats recordings and transcripts as personal data, so collect the minimum, store it securely, restrict access and delete it on schedule; agree the specifics with whoever handles data protection. Have a second person take notes so the interviewer can concentrate. Immediately after each session, write a short summary while it is fresh: the story in a few lines, notable quotes, surprises, and anything that contradicts what you expected.
Synthesise the findings
Once the sessions are done, read the summaries and transcripts against the learning goal. Tag passages by theme, such as triggers, alternatives considered, anxieties, and moments of friction, then look for patterns across participants rather than memorable single stories. Keep three things apart: what people did, what they said, and what you infer. For each theme, note how many participants it came from and which groups, so the strength of a finding is visible. Write each finding as an observation, the evidence, and the implication for the decision. Two people synthesising together catch more and argue less later. If a model helps with the tagging, the guide to using AI for research synthesis covers redaction, a codeframe the model applies and how to keep every theme traceable to a transcript. When the findings are meant to feed advertising, the guide to finding creative angles in customer evidence codes interview passages alongside reviews and ad comments into an angle bank for Meta ads.
Share and decide
Present findings to the people who own the decision, in a short form: the learning goal, who you spoke to, the handful of findings with evidence and quotes, and a recommendation. Go back to the line you wrote in the first step about what would change your mind and say whether it did. Where a finding needs measuring before the business commits, propose the follow-up, often a survey. The tutorial on turning interview findings into survey questions shows how to carry the themes and the participants' own words into it. Record the decision taken and the KPIA metric tied directly to an objective. that will show whether it worked. Store the summaries where future teams can find them, so the next study starts from this one. The guide to reporting research findings covers the write-up in more depth, including how to describe interview themes as reasons rather than as shares.
Discussion guide template
A working structure for a session of around an hour. Adjust the time on each part to the learning goal; the reconstruction of a real episode should always take the largest share.
| Section | Purpose | Example questions | Notes |
|---|---|---|---|
| Introduction | Set expectations and confirm consent | "We're learning how people choose…; there are no right answers." | Confirm recording consent before starting |
| Warm-up | Build rapport and context | "Tell me a little about your role and what a typical week looks like." | Keep short |
| The trigger | Find what started the episode | "Think back to when you first started looking for something like this. What was going on?" | Ask for dates and circumstances |
| The search | Understand alternatives and sources | "What did you look at? Who did you ask? What did you try first?" | Note every alternative, including doing nothing |
| The decision | Find what tipped the choice | "What made you go with that option? What nearly stopped you?" | Probe anxieties and trade-offs |
| The experience | Compare expectation and reality | "What happened in the first few days? What surprised you?" | Follow friction stories in detail |
| Reflection | Gather language and priorities | "How would you describe it to a colleague?" | Capture exact words for messaging |
| Close | Catch what was missed | "Is there anything I should have asked and didn't?" | Explain next steps and the incentive |
Interview findings feed content as well as positioning; the audience research for content guide shows how customer language from sessions like these shapes topics and briefs.
Recruiting and consent checklist
- Sample frame matches the question: the people recruited have the experience the learning goal is about, recently.
- Screener is neutral: it confirms fit without signalling the qualifying answers.
- Groups are planned: where groups differ, each has enough participants to compare.
- Incentive is fair and neutral: proportionate to the time asked, not a discount on the product.
- Permission to contact: the list used for recruiting may be used for research under your privacy notice and the professional codes the MRS publishes.
- Informed consent: participants know the purpose, what is recorded, who sees it, how long it is kept and how to withdraw, and agree before recording.
- No selling: the session is research only; sales follow-up is not triggered by participation.
- Data minimised and secured: recordings and transcripts stored with restricted access and deleted on a stated schedule.
- Identities separated: findings shared with quotes attributed by role or group, not by name, unless the participant agreed otherwise.
A meal-planning app learns why trials stall
Tamsin, the marketing lead at Sprigg, a meal-planning app, needs to rewrite the trial email sequence. Analytics show that 58% of trial users never create a second weekly plan. Her learning goal: find what new users expected in week one and where that broke. She recruits 10 people who started a trial in the last month, 5 who converted and 5 who did not, with a £25 voucher each. In the sessions, a pattern emerges that analytics could not show: 7 of the 10 had expected the app to build a shopping list automatically and had not found that feature, and those who did not convert described the first plan as "more work than writing a list". She rewrites the second and third trial emails around the shopping list feature and plans a follow-up survey to measure how widespread the expectation is.
Interviews with people who have cancelled need a different guide; switch interviews for retention using Jobs-to-be-Done shows how to reconstruct a departure and turn each lost job into a lever. If a round has already finished and produced only compliments and anecdotes, the diagnosis for customer interviews that are not producing insight separates recruitment, questioning, synthesis and counting before the next round is booked.
Verify it worked
- Reread the learning goal. It should name one decision, up to three ranked questions, your current belief and the finding that would change your mind.
- Check the participant list against your criteria. Each person should match the behaviour and recency you defined, with every group you planned for.
- Open the notes for each session. A short summary should exist, written straight afterwards, with quotes, surprises and anything that contradicts what you expected.
- Check the consent records. Agreement should come before recording, and retention and access should be agreed with whoever handles data protection.
- Read the synthesis. Each theme should show how many participants and which groups, with what people did, what they said and what you infer kept apart.
If it goes wrong
- Participants give compliments or predictions. The questions are probably hypothetical. Rewrite them around "tell me about the last time…" and probe for dates and specifics.
- Everyone sounds loyal and satisfied. The sample may be only people who stayed. Recruit people who left or never bought, and compare the groups.
- Sign-ups are thin or sessions get cancelled. Invite more people than you need, check the incentive suits their time and leave gaps between sessions.
- Findings are single vivid stories. Return to the summaries, tag themes across participants and note how many each came from.
- Sessions drift into selling. Stop explaining or defending the product, and keep sales follow-up out of the research.
Next steps
- Read the survey and questionnaire design guide to measure how many people share a view you heard.
- Read the qualitative vs quantitative comparison to decide when to survey, interview or do both in sequence.
- Read the audience research for content guide to turn customer language into topics and briefs.
Common mistakes
- Asking about the future. "Would you use…?" and "How much would you pay?" produce optimism, not evidence. Ask about what people did.
- Talking to the easy people. Interviewing loyal customers who reply to every email when the question is about people who left or never bought.
- Leading the witness. Questions that contain the answer, or reactions that reward the answers the team hoped for.
- Pitching in the session. Explaining or defending the product turns research into a sales call and stops the participant being candid.
- Taking notes from memory. Writing up days later, when the details have blurred into what the team expected to hear.
- Counting anecdotes. Treating one vivid story as a finding, or reporting interview tallies as if they measured prevalence.
- No decision at the end. A well-received readoutThe meeting or short document where a study's or test's results are presented against the decision it was run for, ending in a recorded decision. that changes nothing because nobody owned the decision.
