Customer discovery interviews

0:00 Customer discovery interviews

Your mum loves your idea. Your friends say they'd definitely use it. And yet, three months after launch, nobody buys. What happened? You asked people to predict their own future behaviour, and people are terrible at that, especially when they like you. In this lecture, you'll learn how to run customer discovery interviews that get the truth instead of compliments. You'll learn which questions to ask, which to avoid, how to recruit the right people, and how to turn a pile of notes, and increasingly AI transcripts, into decisions you can trust.

0:40 The core principle

Why does this matter? Because interviews are the cheapest learning tool you'll ever have. A few conversations cost you nothing but time, and they can save you from building the wrong product entirely. But only if you run them properly. The core principle comes from Rob Fitzpatrick's book The Mom Test. Talk about their life, not your idea. Ask about specifics in the past, not generalities or opinions about the future. And talk less, listen more. Here's the key idea. Past behaviour is evidence. Future promises are noise.

1:18 Detective questions

Think of yourself as a detective, not a salesperson. A detective doesn't ask the suspect, would you ever commit a crime? They ask, where were you on Tuesday at eight? So swap your questions. Instead of, would you use an app that does this, ask, tell me about the last time this happened. Instead of, how much would you pay, ask, what have you tried so far, and what did it cost? Instead of, is this a problem, ask, how are you dealing with it now? Good follow ups are simple. Why was that hard? What happened next? Can you show me? That last one is gold. If someone opens the actual spreadsheet they use, you learn more in two minutes than in twenty of opinions.

2:13 A 30-minute structure

Next, structure. A thirty-minute interview has five parts. First, two minutes of context: thank them, say you're learning, not selling, and ask permission to take notes or record. Second, about five minutes on their role and routine. Third, the heart of it, about fifteen minutes on the problem: the last time it happened, what they did, the workarounds, and what those cost. Fourth, a few minutes on priorities: where does this rank against everything else on their plate? And fifth, the close: who else should I speak to, and may I follow up? That last question tests commitment. People who care introduce you to colleagues.

2:59 Worked example 1: meal planning

A simple worked example. Tom in Manchester wants to build a meal-planning app for new parents. Bad interview: he describes his app and asks, would you use this? Everyone says yes. Better interview: he asks, tell me about dinner last Wednesday. One parent says, honestly, we ordered takeaway three times last week because we were too tired to think. What did that cost? Probably sixty or seventy pounds. Have you tried anything to fix it? We bought a recipe book. Never opened it. Now Tom has real evidence. The problem is tiredness and decision fatigue, not a lack of recipes. That points to a very different product, maybe ready-to-cook kits, not another app.

3:48 Worked example 2: clinics in Dubai (illustrative)

Now a realistic scenario, with illustrative numbers. Rania, in Dubai, planned software to stop double bookings at small clinics. She interviewed twelve clinic managers. Here's what her synthesis showed. Nine of twelve raised no-shows unprompted, especially on Sunday mornings, and a receptionist spent one to two hours a day on reminder calls. Seven mentioned sending reminders manually on WhatsApp in both Arabic and English. Only four mentioned double bookings, and none had paid to fix it. And the commitment signals? Three asked to trial something for no-shows, and one introduced her to a group manager with five branches. So she pivoted before writing a line of code. The real problem was bilingual reminders and no-shows.

4:38 Watch me: AI-assisted synthesis

Watch me do it. I have twelve transcripts. First, I anonymise them, replacing names with Interviewee A to L and removing clinic names. Then I paste them into an AI assistant on a business plan that doesn't train on my data. My prompt asks it to extract every problem with the exact quote and the interviewee letter, cluster the problems into themes, count how many raised each one unprompted, and list evidence of commitment, like asking for a trial. I add: only use what's in the transcripts, and say when something is unclear. In seconds I get a draft table. Then I check it. I open three transcripts at random and confirm the quotes are real. The AI saves hours. It doesn't get the final word.

5:33 Recording, privacy, recruiting

A quick word on recording and privacy, because it's easy to get wrong. Always ask permission before recording or using an AI note-taker, explain what you'll do with it, and offer notes only. Data protection laws differ by country, including the UK GDPR and the personal data protection laws in the UAE and Saudi Arabia. And recruitment? Start with people who have the problem, not people who like you. Post in relevant communities, ask for introductions, message people on LinkedIn with a short, honest note: I'm researching how clinic managers handle reminders, not selling anything, could I ask you about it for twenty minutes? Offer to share a summary of what you learn.

6:22 When have you done enough?

So how do you know when you've done enough interviews? There's no magic number, but there is a clear signal. When the last three or four conversations stop surprising you, and the same problems come up in the same words, you've reached saturation for that segment. For most founders that's somewhere between ten and twenty conversations per segment. And watch for what I call the commitment ladder. At the bottom, a compliment, which is worth nothing. Then time: they agree to a second meeting. Then reputation: they introduce you to a colleague or their boss. Then money: a deposit or a paid pilot. Every step up the ladder costs them something, which is exactly why it's evidence. Track where each interviewee lands, and you'll see which segment genuinely cares.

7:18 Common mistakes

Let's cover the common mistakes. Pitching instead of listening. The moment you describe your solution, people start being polite. Asking leading questions like, wouldn't it be great if? Interviewing only friends and family. Stopping at five interviews when answers still vary. Counting compliments as validation. Trusting an AI summary without reading the raw quotes. And finally, forgetting to ask for a commitment at the end. If nobody will introduce you to a colleague or join a trial, the problem may not be painful enough.

7:55 Recap and try this now

Let's recap. Ask about specific past behaviour, not future opinions. Follow a thirty-minute structure with the problem at the centre. Recruit people who have the problem, ask for consent, and protect their data. Use AI to speed up synthesis, but check it against the transcripts. And measure commitment: referrals, trial requests, deposits. Here's your try this now. Write your interview guide using the structure from the lesson, send the recruiting message to ten people today, and book at least five interviews for this week. Next up, we'll turn what you learn into cheap experiments and MVPs.