How to run customer interviews that give you real signal, not false confidence

Customer interviews have a strange property: it's entirely possible to run ten of them, feel great about how they went, and still learn nothing true about whether your idea will work. The interviews aren't the hard part - showing up, asking questions, and taking notes is mechanical. Getting real signal out of them is the hard part, and it depends on decisions you make before, during, and after the conversation, not on how charismatic you are on the call.

Before the interview: who you talk to matters more than what you ask

The single most common validation mistake is interviewing friends, family, or your own network of well-wishers. It isn't that they're being dishonest - it's that agreement from people who want you to succeed isn't signal at all. They're primed to be encouraging regardless of whether they'd ever actually use the thing, and "that sounds like a great idea" costs them nothing to say.

A genuinely useful interviewee clears several bars at once. They currently have the problem, not just relate to it abstractly - someone who can describe the last specific time it happened to them, unprompted, is far stronger signal than someone who nods along in agreement. They match your idea's real ICP signals (a job title, a tool they already use, a public complaint about the exact problem, a relevant company milestone) rather than just the broad demographic persona. If there's any B2B or paid component to the idea, they have - or recently had - the actual budget or authority to solve the problem, because agreement from someone with no path to adopting a solution produces a pleasant conversation and zero signal. And they're reachable within one or two degrees - a warm intro or an existing community, not a theoretical ideal candidate you have no actual path to.

Where to actually find these people

Roughly in order of effort-to-yield for a solo founder starting from zero: your existing professional network, filtered against your ICP signals rather than blasted broadly. Then communities where the persona is already discussing this specific problem - not just members of a general community, but people actively posting about the pain point itself. Then targeted cold outreach to people who match your signals and show public activity suggesting the problem - a complaint, a related job change, a tool switch. Adjacent professional events and meetups are a reasonable last resort, but they're the lowest-yield-per-hour option on this list, worth trying once the first three are genuinely exhausted, not before.

During the interview: ask about the past, not the future

The single biggest lever you have during the conversation is the shape of your questions. "Would you use a tool that did X?" is a hypothetical, and hypotheticals are cheap to say yes to - almost everyone will, out of politeness or genuine but untested optimism about their own future behavior. "Tell me about the last time this happened to you" asks for a specific memory, and specific memories are much harder to fake or exaggerate. If someone can't produce a concrete recent instance of the problem, that's itself useful information, even though it doesn't feel as satisfying in the moment as an enthusiastic hypothetical yes.

Push past the first comfortable answer. "That's annoying" is a start, not an endpoint - ask what they did about it, how often it happens, whether they've tried anything else, and what stopped them from fixing it already. The gap between "this is annoying" and "this is annoying enough that I've already tried three things to fix it and none worked" is the entire difference between weak and strong signal.

The two habits that quietly ruin an interview round

Two mistakes account for most misleading validation results, and both are easy to fall into without noticing. The first is only interviewing people who already said yes to a beta or waitlist - self-selected enthusiasm skews everything toward false positives before you've asked a single real question, because you've pre-filtered for exactly the people most likely to tell you what you want to hear. The second is stopping at the first one or two conversations that confirm your hypothesis. Confirmation bias is strongest exactly when you're eager to be done; the actual goal of an interview round is to find a pattern across a genuinely diverse set of five or so conversations, not to get an early yes and call it validated.

After the interview: judge the pattern, not the vibe

Resist the urge to walk away from any single great conversation feeling validated. One enthusiastic interview is an anecdote. What you're actually after is whether a genuinely diverse slice of your persona - different seniority levels, company sizes, industries, not five near-identical people - independently describes the same real, specific problem in their own words. If you set a threshold before you started (say, six of ten interviews confirming a specific behavior), hold yourself to it, even when the honest count falls short of what you were hoping for. A short count isn't a failed interview round - it's the interview round working exactly as intended, telling you something true before you've spent months building around a wrong assumption.

Bringing it together

Good customer interviews aren't about being a good interviewer in the conversational sense. They're about talking to the right people, asking about real past behavior instead of comfortable hypotheticals, and being honest with yourself about the pattern across a diverse set of conversations rather than the warm feeling from any single one. Nucleus builds this discipline into the process directly - it helps identify who's actually worth talking to from your own network based on real ICP signals, and reviews what you actually heard against the hypothesis you set out to test, so the pattern is judged on the evidence, not on how the best conversation felt.