Customer discovery interviews: questions that produce usable evidence

Most founders run customer discovery interviews and walk away with a stack of notes that say, in one form or another, "they seemed really interested." That's not evidence. It's a feeling, and endorsing body assessors have read thousands of applications built on feelings dressed up as market research.

The difference between an interview that produces usable evidence and one that produces noise isn't effort or interview count. It's the questions. A structured set of questions aimed at behaviour and history produces facts you can cite. A loose conversation aimed at your product produces opinions you can't.

Why "would you use this?" is a worthless question

Ask anyone a hypothetical purchase question and watch what happens: they picture themselves as the kind of person who would use a useful-sounding tool, and they say yes. It costs nothing to agree. There's no wallet involved, no calendar commitment, no real trade-off against the other things they could spend money or attention on.

Rob Fitzpatrick's framing (widely cited across the customer-discovery literature) is that people are polite liars when a stranger pitches them an idea in a room. They want to be encouraging. They want the conversation to end well. None of that generates evidence — it generates a social outcome that happens to look like validation.

An endorsing body assessor reading "80% of interviewees said they would use the product" learns nothing about demand. They learn that you asked a leading question and 80% of people were polite. See the market research standards endorsing bodies apply for how this distinction plays out in the actual assessment.

The four categories of question that actually work

Structured customer discovery interviews work through four categories, roughly in this order:

1. Problem frequency. "How often does [the problem] come up for you?" Daily, weekly, monthly, rarely. A problem someone encounters daily and has never solved is a different opportunity from one they hit twice a year and shrug off.

2. Current workaround. "What do you do about it today?" Nobody has zero response to a real problem — they've built some workaround, however clumsy. If the honest answer is "nothing, I just live with it," the problem may not be painful enough to pay to fix.

3. Cost of the workaround. "How much time (or money) does that cost you?" This converts a vague annoyance into a number. Ten hours a month of manual reconciliation work has a defensible cost; "it's a bit annoying" does not.

4. Willingness to pay, anchored to a number. Not "would you pay for this?" but "if this saved you those ten hours a month, what would that be worth to you?" — and then let them name a figure, or react to one you propose.

Interview for the problem before you mention the solution

The single biggest structural mistake in founder-run discovery interviews is leading with the pitch. "I'm building a tool that does X — would this help you?" primes the interviewee to evaluate your idea rather than describe their actual experience.

The better sequence: spend the first two-thirds of the interview entirely on the problem, using the four question categories above, without mentioning that you're building anything. Only in the last third, if at all, describe what you're building and ask for reactions — treating that reaction as commentary, not as the primary evidence.

This ordering matters because waitlist signups and early interest signals are weak on their own precisely because they're collected after the pitch, not before the problem is understood. An interview run in problem-first order produces a cleaner separation between "this problem is real" and "this specific solution appeals to them" — two different claims that endorsing bodies assess differently.

Questions that expose a founder-market-fit gap

A secondary use of discovery interviews is testing whether you, specifically, are positioned to solve this problem. Ask interviewees: "Who else have you talked to about this? Who in the industry do people trust on this topic?" If the names that come back are people you already know, or people in a network you can credibly join, that's a signal worth documenting. If the names are unfamiliar and the industry is opaque to you, that's worth knowing before you go further — see what a credible founder-market-fit case looks like for how this connects to the innovation and viability pillars.

How to document interviews so they survive scrutiny

An interview is only evidence if it's recorded in a form an assessor can trust. That means:

• Name and role of the interviewee (with consent to reference them, even if anonymised in the public-facing document).
• Date of the interview.
• Direct quotes, not paraphrases, for the load-bearing findings.
• A consistent question set across interviews, so patterns across the sample are real patterns and not artefacts of asking different people different things.

Recruiting the right interviewees in the first place

Good questions asked of the wrong people still produce weak evidence. A common shortcut is interviewing friends, family, or people in your existing network who resemble your target customer in name only — a former colleague who now works in a vaguely adjacent industry, a friend who runs a small business but not the kind you're targeting. These interviews are easy to arrange and almost worthless as evidence, because the interviewee's politeness bias compounds with a personal relationship bias.

The stronger approach is recruiting interviewees who have no reason to be kind to you: cold outreach to people matching your actual target profile, introductions through a trade association or industry group you're not personally embedded in, or paid participant-recruitment services. It's slower and less comfortable than emailing your network, but the resulting evidence survives scrutiny in a way that friends-and-family interviews don't.

Common failure modes beyond the leading question

A few other patterns quietly weaken interview evidence even when the core questions are well-designed:

• Sample bias toward the enthusiastic. If your interview pool is entirely people who responded to a call for "anyone interested in talking about [problem]," you've pre-filtered for people already primed to be positive. Balance this with some interviewees recruited without reference to the problem at all — screened only on demographic or firmographic fit.
• Stopping after the first few confirm what you hoped to hear. Confirmation bias is easy to fall into once two or three interviews go well. Keep the full planned sample size even when early results look good — later interviews sometimes surface disconfirming detail that changes the picture.
• Treating every interview as equally weighted. A ten-minute rushed call and a forty-minute detailed conversation are not equivalent data points, even though they'll look the same in a simple tally. Note interview length and depth, and weight your synthesis accordingly.

What a credible interview summary looks like in an application

Rather than "we spoke to 20 potential customers who were very positive," a stronger version reads: "We interviewed 22 operations managers at UK logistics SMEs between March and May. 18 of 22 described manually reconciling delivery data weekly, at a self-reported cost of 6-14 hours per month. 14 of 22 named a specific tool or spreadsheet workaround currently in use. When asked what they would pay for an automated version, responses ranged from £150-£400/month, with a median of £220." That level of specificity is what separates evidence from anecdote — the same standard applies whether you're running interviews or testing demand through a landing page.

Sources and further reading

• GOV.UK: Innovator Founder visa guidance
• GOV.UK: Appendix Innovator Founder — Immigration Rules
• Davidson Morris: Innovator Founder visa

Key takeaways

• Never lead with "would you use this?" — it produces near-universal false-positive agreement.
• Interview for the problem first: frequency, current workaround, and the cost of that workaround.
• Anchor willingness-to-pay questions to a specific number, not a yes/no.
• Record interviews (with consent) and cite direct quotes with names, roles, and dates.
• 15-30 structured, consistently-run interviews outperform a much larger pool of loose conversations.
• Interview before you build, or at minimum before you pitch, to avoid biasing the interviewee toward your existing solution.

• customer-discovery
• market-validation
• interviews
• evidence
• validation