Two founders walk out of the same coffee shop. One is elated: "She said she'd definitely use it!" The other is frustrated: "He said it was a nice idea, but he wouldn't pay for it." Both talked to real people. Both got honest-sounding answers. Only one of them actually learned something.
The difference wasn't luck. It was the questions. Most early founders treat customer interviews as a sales pitch in disguise—they're hunting for a yes, not for the truth. And a yes means nothing when you asked for it.
Validating a startup idea with customer interviews is a skill, not a formality. Done right, it can save you months of building the wrong thing. Done wrong, it just gives you false confidence with a paper trail. Here's how to do the first version.
Key Takeaways
- Ask about the past and the problem, never about the future and your solution.
- Recruit strangers with a real stake, not friends who want to be nice.
- Aim for 15–25 conversations per segment before you trust a pattern.
- Score each interview on specificity: did they describe a problem they already spend time or money on?
- Your job is to try to kill the idea, not confirm it.
Why most customer interviews fail before the first question
The interview itself is rarely the problem. The setup is.
If you walk in already convinced, you'll hear what you want. If you're talking to your college roommate, he'll say the idea sounds cool because he likes you. And if your opening line is "I'm building an app that…", you've already lost—everything after that becomes a politeness exercise.
I watched a founder burn five months on a scheduling tool for freelancers. He ran 30 interviews. He got enthusiastic nods in almost every one. He shipped, launched, and got 11 signups in 6 weeks—and 2 of those were his own test accounts. When we replayed the interview recordings, the tell was everywhere: he'd asked "would you use something that does X?" People were answering the question he asked. The problem was the question.
How many customer interviews do you actually need?
There's no magic number, but there's a useful range. For a single, well-defined segment, 15 to 25 interviews is usually where patterns stop surprising you. If you're testing two distinct segments, treat them as two separate studies—25 each, not 50 mixed together.
Stop when the last five interviews tell you nothing new. If every conversation still surprises you at number 20, you don't have a segment yet—you have a vague target. Narrow it.
Where do you find interviewees outside your network?
Your first-degree network is a trap. It's the lazy option, and it gives you the worst data. Here's what actually works:
- Post in niche communities where your target already complains out loud—subreddits, Discord servers, Slack groups, industry forums. Frame it as "researching how X handles Y," not "I built something."
- Reach out cold, in small batches, with a specific ask. "Can I get 20 minutes of your time about how you handle invoicing?" converts far better than a wall of text about your vision.
- Offer a small incentive. A $20 gift card is a rounding error next to the cost of building the wrong feature. It also filters for people who'll show up.
- Ask each interviewee for one referral. This is how you escape your own network fast.
Don't pay for a survey panel to replace interviews. Surveys give you volume; interviews give you the sentences behind the answers. You're early—you need the sentences.
The questions that actually tell you the truth
Switch your entire interview from "future tense" to "past tense." People are terrible at predicting what they'll do and remarkably accurate at describing what they already did.
What should you never ask in a customer interview?
Three questions ruin more interviews than any other:
- "Would you buy this?" — It's hypothetical. The answer is almost always a polite yes, and polite yeses are worthless.
- "Do you like this idea?" — You're asking for approval, not information. Nobody wants to crush you.
- "How much would you pay for this?" — People invent numbers under social pressure. Real pricing comes from what they've already paid or the workarounds they've built.
Every one of those questions invites a guess. Cut them.
What questions should you ask instead?
Anchor everything in what already happened. A few that do real work:
- Walk me through the last time you ran into this problem. What happened, step by step?
- What did you do about it? What did that cost you—time, money, sanity?
- Have you tried any tools or workarounds for this? Which ones, and why did you stop or keep them?
- Who else is affected when this goes wrong?
- What have you already tried to fix it yourself?
Notice none of these mention your idea. That's the point. You're reconstructing a real past event, not pitching a fictional future. If a founder had asked the freelancers those questions, he'd have found out fast that most of them had already duct-taped a solution together—and weren't about to switch.
How to read the signals without fooling yourself
After the interview, before the coffee wears off, write down two things: what problem they described, and what they're already doing about it. Then score it.
| Signal | Weak | Strong |
|---|---|---|
| Problem specificity | Vague complaint, no example | A concrete recent event, with details |
| Current spending | "I just live with it" | Pays for a tool, hires someone, or builds a workaround |
| Emotional charge | Flat, indifferent tone | Audible frustration—they bring it up unprompted |
| Follow-up | Answers once, moves on | Asks you a question, offers to intro someone |
One glowing interview means nothing. Five interviews in the "strong" column for the same problem, across different people who don't know each other, is a signal. That's when you build the smallest possible thing.
Here's the honest part: I once collected 18 interviews that all scored "strong" on frustration, and I still built the wrong product—because I'd misread what they were frustrated about. The emotion was real. My interpretation wasn't. Interviews give you raw material; they don't hand you the answer.
Questions readers keep asking
What does a good customer interview sample look like?
You want people who genuinely match your target segment and have something at stake—a real workflow, a budget, a deadline. A sample of 20 friends, colleagues, or random people who "find the space interesting" is not a sample; it's a focus group of goodwill. Aim for 15–25 per segment, all with firsthand experience of the problem, and mix cold outreach with referrals so you're not just interviewing your echo chamber.
How do you avoid friends saying yes just to be nice?
You can't fully avoid it—so reduce it. Don't pitch them. Don't tell them what you're building until the very end, if at all. Ask about their past, not your idea. And watch for the tell: if someone's enthusiasm comes with no concrete example of the problem, discount it entirely. Enthusiasm without specifics is politeness wearing a costume.
What to do with all of this
Validation isn't a moment where the market gives you permission. It's the slow accumulation of specific, boring, undeniable facts about how people already behave—and whether that behavior points toward something you can build and sell.
So before you write a line of code or a single line of pitch deck: go find 15 people who've lived the problem, ask them what they did last time it happened, and listen for the details. If you can't find 15 people, that's your answer.
And if you can? You've learned more than most founders learn in a year of building.
