How to Run Customer Discovery Interviews That Actually Tell You Something

You sit down with a potential customer. You ask, “Would you use this if it existed?” They smile. They say yes. You walk out feeling validated. Six months later you ship the product. Twelve people sign up. Three actually use it. Two of them are your friends.

That’s not a product problem. That’s an interview problem.

Most founders run startup customer discovery interviews that produce garbage data. The questions are wrong. The audience is wrong. The signal a “yes” gives you is meaningless. You spend three months building a thing nobody buys, then wonder why.

The myth: customer discovery interviews are about asking what people want

They’re not. People don’t know what they want. People are nice. People will tell you anything sounds like a great idea when there’s nothing at stake. The kindest, most enthusiastic interview subjects are usually the worst signal.

Discovery is about getting people to show you what they already do. Past behavior is the only thing that survives contact with their wallet. Every question that gets you a story about last Tuesday is worth a hundred questions that get you a “great idea.” Receipts compound. Compliments fund nothing.

What discovery is actually for

Discovery is the phase before you have a hypothesis. Validation is the phase after. Sales is the phase after that. Most founders blur all three. They show up to a “discovery” call already pitching, because they’re nervous and want to hear something good. The interview turns into a sales call. The sales call gets a polite no. The founder concludes the idea is broken when the conversation was the actual broken thing.

The one question that beats every script is short: “When was the last time you tried to solve this?” If the person can answer with specifics, you have something. If they pause and say “I haven’t thought about it,” you have nothing. The pause is the data.

The 4-question framework that gets you real signal

The hard work in a discovery interview is staying away from your own product. The framework below keeps you out of the pitch and inside the person’s actual workflow. The classic reference is Rob Fitzpatrick’s The Mom Test. Read it. The four moves below are the operational shape I use with coaching clients on top of it.

1. Ask about last week. Skip the hypothetical

“How did you solve this last week?” Not “would you use this if it existed?” The first question gets you receipts. The second one gets you compliments. Specifics about last week tell you what tools they reached for, how much time it cost, who else was involved, and whether they were frustrated enough to try anything new. Hypothetical questions about next week tell you they’re polite.

If they can’t name a recent time they tried to solve the problem, it’s not a real problem yet. Move on.

2. Map the workflow surface

Ask what tools they already pay for in this space. What budget line a fix would come out of. What they tried before and why it didn’t stick. You’re not validating an idea. You’re surveying a battlefield.

The point is to find out what the current stack looks like and what’s already failing them. If they’re using six tools and a spreadsheet to hold the workflow together, you have a real wedge. If they’re getting by fine with one tool they barely use, you don’t.

3. Find the wedge with a money question

“What would you defund to pay for a fix?” This is the question that separates a nice-to-have from a budget line. If they can’t name something they’d cut, the problem isn’t urgent enough yet. They might agree it’s annoying. They might tell you they wish someone would solve it. Neither of those is a check.

This question feels rude the first time you ask it. Ask it anyway. The discomfort is the point. Polite questions get polite answers.

4. Reverse the pitch and make them defend the problem

Say: “I’m honestly not sure this is a real problem. Talk me out of it.” If they shrug and agree it’s probably not that big a deal, kill the hypothesis. If they push back with specific stakes, real money, named consequences, and what they’ve already tried, you’ve found something. You’ve forced them to be the seller instead of the buyer.

This is the move I see founders skip most often. They’re so scared of hearing no that they soften every question. Reversing the pitch breaks the pattern.

Where to find 10 strangers when you don’t have a network

The Mom Test doesn’t cover this part. Most founders skip past it and interview their friends first, which is the fastest way to confirm a bad idea. Friends are not your audience. Strangers are.

Four channels work for early discovery:

LinkedIn search. Search by title plus an industry term. Send a short DM that names the specific problem you’re researching. Don’t ask for “20 minutes of your time.” Ask one specific question and offer to share back what you learn. People answer questions. They don’t show up for vague calls.

Niche communities. Slack groups, Discord servers, Reddit subreddits, IndieHackers threads, specific Twitter circles. Don’t post your idea. Lurk first. Read what people complain about in their own words. The language you find there is the language you’ll write your eventual landing page in.

Conference speaker lists. Public, easy to find, and the speakers are already self-identified as people who care about the problem space. Cold outreach to a conference speaker has a much higher response rate than cold outreach to a random title at a random company.

Customer reviews of adjacent tools. Search G2, Capterra, and Reddit for the tools your customer already uses. The complaints in the one-star reviews are your interview script. A coaching client of mine, Josh, was building a CFO analytics tool. He spent a weekend reading competitor reviews on G2 and saw that every single complaint was about complexity. He shipped a simpler interface and adoption moved faster than every other validation tactic he’d tried.

The rule: ten strangers before you talk to anyone you know.

Read the behavior. Words are noise

Verbal signals are noise. Behavioral signals are signal. A founder I work with had thirty “great idea” interviews and zero customers when she came to me. We re-ran her notes and split everything into two columns: things the interviewees said and things the interviewees did. The “said” column was full of enthusiasm. The “did” column was almost empty. Nobody had tried to solve the problem before. Nobody had paid for anything adjacent. Nobody had built a workaround. The hypothesis died that afternoon.

Here are the behavioral signals I look for, in order of weight:

• Money already spent. They’ve paid for something else trying to solve this.
• Workaround already built. A spreadsheet, a script, a Zap, a contractor.
• Forwarding behavior. They forward your email to a colleague unprompted.
• Specific time-cost claims. “I lose two hours a week on this” with the calculation already done in their head before you asked.
• Pre-existing rage. They start telling you stories before you finish asking.

Anything that requires their imagination is noise. Anything that already happened is signal.

A real example: Flightfox, $9M to $25M

A founder I work with ran Flightfox, a corporate travel platform that had been a YC 2012 company. By 2023, COVID had wrecked the business. Churn was out of control. Quick fixes weren’t working. His team had written off validation work as a waste of time, because they’d done it once at the start and moved on.

I pushed him back to first principles. He re-interviewed enterprise buyers from scratch. Stripped away years of assumptions about who the customer was and why they were buying. Asked about last week. Mapped what the buyers were actually doing inside their travel programs.

Fourteen months later the company went from $9M to $25M in revenue. From zero enterprise leads to over a hundred. One new hire. Zero extra marketing spend. The interviews weren’t a one-time phase. They were the foundation he came back to every time growth stalled.

Most founders treat discovery as the thing you do once before you build. The teams that keep growing treat it as the thing you do every time you’re stuck.

Where this advice breaks

Three honest edge cases.

If the product category doesn’t exist yet in any form, past-behavior questions stall. People can’t tell you how they handled a problem they didn’t know to call a problem. In that case, ask about the adjacent workflow they’re cobbling together to work around it.

If the buyer and the user are different people, ten interviews with users tells you nothing about whether the buyer will sign the check. Talk to both. Especially in B2B, the budget owner has a completely different problem than the person who’ll actually use the thing.

If you’re a second-time founder building for a domain you’ve lived in for a decade, the rule about strangers loosens. Your network in that domain is your audience. Don’t manufacture distance you don’t need.

What to do with ten interviews

Pattern-match. Look for the same complaint phrased five different ways. Look for the same workaround invented independently. Look for the same budget line getting mentioned. When you can write a single sentence that captures what every interview circled around, you have a hypothesis worth testing.

That hypothesis is what the validate your startup idea framework picks up. Discovery feeds validation. Validation feeds the build. The startup guide walks the full five-stage sequence end to end. If you want the deeper read on what good product-market fit signal looks like, product market fit validation covers the lagging indicators. When you’re ready to scope a first version, build your MVP is the playbook.

If you want a second pair of eyes on the questions you’re about to ask, book a call.

Nice people are the worst interview subjects. Find ten strangers. Ask about last week.