3.4 Customer Discovery Interviews
At a Glance
Other names Customer Interview
In Brief
Customer discovery interviews are one-on-one conversations with potential customers designed to uncover their pain points, current behaviors, and purchasing motivations. You ask open-ended questions about past experiences — not hypotheticals — and listen for patterns. The output is qualitative insight into who your customer really is, what problem they struggle with, and how they already try to solve it, which helps narrow your target market and sharpen your value proposition.
Common Use Case
You have an idea for a product but you are not sure who would actually pay for it. You set up one-on-one conversations with people you think might be customers and ask about their daily frustrations, how they currently solve the problem, and what they have tried before. Their answers help you decide whether to pursue the idea or pivot.
Helps Answer
- Who is the person most likely to buy this?
- What problems are they struggling with right now?
- Where do these potential customers spend their time?
Description
Customer discovery interviews are one-on-one conversations with potential customers to learn what problems they actually have, how they currently solve them, and what they would change. They are generative — you are not validating a feature or testing copy; you are mapping the problem space before you build anything. The single most important rule of the method is that questions probe past behavior, not future intent. “What did you do the last time this came up?” produces signal; “Would you use a tool that did X?” does not.
The method has several common variations. Most interviews are 1:1, but Focus Groups (one researcher, multiple participants from the same segment) work when you want to surface how participants describe the problem to each other rather than to you. In-person sessions are the default in B2B and produce the richest observation data — body language, environment, what is on the participant’s screen — but remote sessions are now common and reach people no in-person session ever could. Structured interviews follow a tight script and are easier to compare across participants; semi-structured interviews keep the script as a backbone but follow whatever thread the participant pulls on, and tend to produce the unexpected findings that change product direction. Most experienced interviewers default to semi-structured.
The output of a good interview round is a sharper customer persona, a ranked list of pains worth investigating, and a much shorter list of assumptions you were carrying that the data did not support.
How to
Prep
-
Define the learning goal. Write one sentence: “I want to learn whether [target segment] currently struggles with [problem space], and how they handle it today.” If you cannot finish that sentence, you are not ready to interview yet — go back to your customer hypothesis.
-
Write down your customer-persona assumptions. What you currently believe about who the potential customer is, what they do, and what they care about. The interviews will test these assumptions; you cannot tell what the data is telling you if you have not stated the prior.
-
Draft a screener. Three to four short closed-ended questions that confirm the person you are about to interview matches the segment you are trying to learn about. Ask them on the recruiting form, not at the start of the call.
-
Write the interview guide. Eight to ten open-ended questions focused on past behavior, not hypotheticals. The classic backbone covers:
- “What’s the hardest part about [problem context]?”
- “Can you tell me about the last time that happened?”
- “Why was that hard?”
- “What, if anything, have you done to solve that problem?”
- “What don’t you love about the solutions you’ve tried?”
Treat the guide as a backbone, not a script. Follow whatever thread the participant pulls on. Include at least one fully open-ended catch-all with no hypothesis behind it — “What else should I know about how you handle [context]?” — since the strongest insights often surface where you weren’t looking. Decide the opening framing here too: a one-sentence summary of the purpose plus a line that there are no right or wrong answers — “I want your real experience, not what you think I want to hear.”
-
Recruit the interviewees. Five interviews is a good target for early discovery; 8-12 is the saturation range within a single segment; some practitioners run dozens before drawing conclusions.
-
Set up logistics. Confirm consent before recording (legal requirements vary by jurisdiction). Prepare a notes template, the recording tool you will use, and any thank-you gift (a $5–$50 gift card is typical for B2C consumers; B2B specialists may expect more or refuse compensation).
-
Pilot the guide. Run one practice interview with a colleague or a friendly customer before going live. Pilots almost always surface a leading question, an ambiguous phrasing, or a question whose answer you can already predict.
Execution
1. Frame the interview.
Deliver the opening framing from your guide, then confirm consent for any recording.
2. Qualify.
Ask the screener question first if it wasn’t on the recruiting form. If the participant doesn’t match the segment, end the interview gracefully — bad data is more expensive than a short call.
3. Open with a warm-up.
A low-stakes question about their current role or recent week gets the participant comfortable talking before you ask anything that matters.
4. Listen and follow up.
Let the participant talk. Use “what” and “how” follow-ups; avoid “why” early in the conversation (it can feel accusatory). When a participant mentions a specific incident, slow down and dig into that incident — concrete past events produce far more reliable data than generalizations.
5. Close.
Wrap up by asking if there is anything you should have asked but didn’t. Ask for referrals to other people in the same segment. If applicable, ask permission to follow up.
6. Debrief immediately.
Write notes within an hour while the conversation is fresh. An AI transcription service turns this around in minutes, but read the transcript critically — current models still mis-attribute speakers and hallucinate filler words.
Analysis
Review the transcript and/or listen to the recording, and capture data on each of the following:
- Job: What activities are making the participant run into the problem?
- Obstacle: What is preventing the participant from solving their problem?
- Goal: If they solve their problem, then _____?
- Current solution: How are they solving their problem?
- Decision trigger: Were there pivotal moments where the participant made key decisions about a problem?
- Interest trigger: Which questions did the participant express interest in?
- Persons: Are there any other people involved with the problem or solution?
- Emotions: Is there anything specific that causes the participant to express different emotions?
- Measurement: How is the participant measuring the cost of their problem?
- Leading questions Phrasing that telegraphs the answer (“Wouldn’t it be easier if…?”) gets you agreement instead of truth. Ask what the participant actually did, in neutral wording, and have a colleague or AI flag leading phrasings in your guide before you run it.
- Social desirability bias Participants offer compliments and say what they think you want to hear, especially friends, family, and anyone who knows the idea is yours. Treat praise as a warning sign and steer every answer back to specific past events — what they did, when, and what it cost them.
- The pitch trap The moment you describe your idea, you contaminate every answer that follows. Hold the solution back until the end of the conversation and aim to listen far more than you talk.
- Confirmation bias You hear the evidence that supports your idea and quietly discount the rest. Write your assumptions down before interviewing (Prep step 2), then read each transcript hunting for what contradicts them.
- Future-intent bias People are poor forecasters of their own behavior, so “would you use this?” produces false positives. Anchor every question in what actually happened the last time the problem came up, and discard speculative enthusiasm.
- Selection bias Interviewing whoever is easiest — your own network, enthusiasts, existing customers — skews toward people who already like you or the space. Recruit through several channels, screen for the real target segment, and note which segments you could not reach.
- Anchoring and order effects The sequence in which you ask questions can shape the answers, because an early topic anchors how the participant frames later ones. Vary the question order across interviews so a warm-up topic isn’t always priming the same downstream response.
Learn more
Case Studies
Gothelf: Discovery cadences at Winware, Vistaly, and Penpot
Jeff Gothelf profiles three startups running systematic discovery: Winware’s Steven Cohn did 300+ interviews pre-build at 2–4 per day, Vistaly’s Matt O’Connell runs 15–20 weekly, and Penpot’s Pablo Ruiz-Múzquíz holds 30 interviews per week to validate features.
Airbnb: Hosts’ homes in New York, 2009
In 2009 co-founders Brian Chesky and Joe Gebbia traveled to New York to stay with hosts and interview them; observing how hosts struggled to create listings led to the team personally photographing apartments, and revenue in New York is reported to have roughly doubled within a month.
Lean Foundry: “Steve” reframes 50-interview VR research
A founder ran 50 interviews and 200+ surveys about a VR rendering tool with no clarity; a single 30-minute past-behavior conversation about a recent client presentation surfaced the real problem — $400, three-day revision cycles — and pivoted the product from better rendering to instant self-serve revisions.
Further reading
- The Mom Test
- The Customer Discovery Handbook
- How I Interview Customers
- Greg Guest, Arwen Bunce, Laura Johnson — How Many Interviews Are Enough? An Experiment with Data Saturation and Variability (Field Methods, 2006)
- Great Lakes I-Corps Hub — How to Use AI for Customer Discovery
- What are your favorite methods for doing problem interviews during customer discovery?
Got something to add? Share with the community.
