How do you do customer interviews that are actually useful?

Most customer interviews fail because founders ask people what they want instead of studying what they do. Useful interviews treat the customer as a source of behavioral evidence, not a focus group. The goal is to stress-test your user model, not validate it.

Stop seeking confirmation — seek disconfirmation

The instinct in a customer interview is to present your idea and watch for enthusiasm. This is exactly backwards. Enthusiasm is cheap; behavior is expensive. A customer who says 'I'd totally pay for that' and a customer who has already duct-taped together a workaround using spreadsheets and Zapier are in completely different categories. The second one has revealed a real problem through revealed preference. Your job in the interview is to find the second type of customer.

Paul Graham's point about user models is directly relevant here: your mental model of the customer's needs is almost certainly wrong in at least one important way, because needs shift as circumstances change. That means the interview shouldn't feel like a confirmation hearing — it should feel like an investigation. Come in with your strongest hypothesis about what problem this person has, and then design every question to try to break that hypothesis. If the hypothesis survives the attempt, it's stronger. If it breaks, you've learned something that most competitors will take months longer to discover.

Practically, this means replacing opinion questions with history questions. Not 'would you use a tool that did X?' but 'walk me through the last time you had to deal with X — what did you actually do?' The specificity of a real incident surfaces details no hypothetical question can reach: what tool they reached for first, who they called, where they got stuck, what they settled for, and how much time or money it cost them.

The size illusion: why small companies have the advantage here

Garry Tan has observed that early-stage founders often make the mistake of mimicking big-company behavior to appear credible — including big-company indifference to individual users. This is a costly error in customer interviews. Your actual advantage over any established competitor is that you can sit with one customer for two hours and care intensely about every word they say. A product team at a large company runs surveys at scale precisely because they can't do this. You can, and you should.

This means calling people directly, not sending a Typeform. It means asking if you can watch them do the task you're trying to improve, rather than asking them to describe it from memory. Memory compresses and distorts; observation doesn't. If you're building a tool for, say, operations managers, ask if you can shadow one of their Monday morning standup sessions. The fifteen minutes of observation will be worth more than a ninety-minute interview based on recall.

Being small also means you can respond in real time. If an interviewee mentions something unexpected — a constraint you'd never considered, a competitor you hadn't heard of, a workaround that is oddly specific — you can follow that thread on the spot. You're not bound to a script written by a research team six weeks ago. Stay curious when something surprises you. Surprise is usually signal.

Who to interview and how to get them to talk honestly

The easiest interviews to get are with your friends, warm connections, and people already inclined to be supportive. These are also the least useful. Polite people give polite answers. The interviews you need are with people who have no social stake in making you feel good — ideally, people who are currently using a competitor's product or have abandoned a solution in your space entirely. The abandoned user is particularly valuable: they tried to solve the problem, failed, and have already worked through their own post-mortem on why.

Cold outreach for interviews works better than most founders expect if the ask is framed correctly. Don't ask for 'feedback on your startup.' Ask for fifteen minutes to learn about their experience with a specific problem. People who've struggled with something genuinely want to be heard about it — their frustration is a motivator. Keep the ask short, make it clear there's no sales pitch, and offer to share what you learn in aggregate. Response rates on well-targeted cold outreach for customer research are consistently higher than for sales outreach, because the power dynamic is different: you're asking for expertise, not money.

During the interview, the most powerful move is to stop talking after you ask a question and let silence do the work. Silence is uncomfortable. Most interviewers fill it. If you can tolerate ten seconds of quiet after a question, the other person will fill it — usually with the thing they were hesitating to say, which is almost always more useful than their first answer.

What to do with what you learn

A customer interview is only useful if it changes something. That means you need a process for turning raw conversations into decisions. The minimum viable version: within an hour of each interview, write one paragraph summarizing what surprised you. Not what the person said — what surprised you. If nothing surprised you, the interview probably confirmed what you already believed, which is weak evidence. If something surprised you, that's where to look.

After five to ten interviews, look for the patterns that cut across conversations — but pay close attention to the outliers too. The customer who described a completely different workflow than everyone else, or who mentioned a constraint no one else raised, might be representing a segment you've overlooked rather than noise you can discard. Early markets are often won by founders who served the 'weird' segment that established players had written off.

Finally, use interviews to sharpen your next prototype or test, not to justify your current one. The worst outcome of a customer interview is walking away with a quote you can put in a pitch deck. The best outcome is a specific change to what you're building, how you're pricing it, or who you're targeting. If your interviews aren't producing decisions, redesign the questions until they do.

“Your user model almost couldn't be perfectly accurate, because users' needs often change in response to what you build for them.”

— Paul Graham, source

The one thing to do

After every interview, write down the one thing that surprised you — if nothing did, your questions aren't probing deep enough.

Frequently asked questions

How many customer interviews do you need before you can trust the pattern?

There's no universal number, but five to eight interviews with people who match your target user profile will usually surface the two or three most important themes. Run more until new conversations stop producing surprises — that's the practical signal that you've reached saturation on your current hypotheses.

Should you record customer interviews?

Yes, with permission — but don't let recording become a substitute for active listening. The goal is to be fully present during the conversation and use the recording to catch what you missed. Transcripts are useful for pulling exact language customers use to describe their problem, which you can often use verbatim in your own copy.

What's the biggest mistake founders make in customer interviews?

Pitching. The moment you shift from asking questions to explaining your solution, you've lost the interview. Even well-intentioned founders slip into pitch mode when they hear a problem they think they've solved. Save the pitch for a separate, explicit conversation.

How do you interview customers when you don't have any yet?

Target people who have the problem your product will solve, not people who would use your specific product. Find them in communities, subreddits, Slack groups, or LinkedIn searches. Frame the ask around understanding their experience with the problem — you'll get higher response rates and more candid answers than if you lead with what you're building.

Sources

More playbook answers · Growth Prophet home