A specific script, the questions that work, and the ones that quietly destroy your data.
15 min read · Level 1
The moment you describe your solution, the conversation stops being about their life and starts being about your feelings — and people are kind, so the data quietly becomes worthless. The single most valuable question is not about your idea at all: it is "have you tried to fix this, and what happened?". Effort already spent is evidence that words cannot fake, and it is the only reliable proof that a problem is worth money.
A 25–30 minute conversation about how the person's work actually goes today. You do not describe your product. You ask about specific past events.
The structure: context → last time it happened → what it cost → what they tried → what happens next.
Because unstructured conversations produce encouragement, and encouragement feels like evidence. People are kind, especially to someone visibly excited.
A structured interview about past behaviour is much harder to distort. 'What did you do last Tuesday' has an answer that exists independently of how the person feels about you.
The script:
Question 5 is the one that matters most. If they have tried to fix it — searched for a tool, built a spreadsheet, hired someone, paid for something and abandoned it — that is real evidence of pain. If they have never tried anything, it is an annoyance, not a problem worth a company.
Questions that destroy your data:
First-time founders pitch within the first three minutes. It is very hard not to — you are excited and it feels rude to withhold. But the moment they know what you are building, they start being helpful about your idea instead of honest about their life.
Second: only talking to people who already like you. Friends and supportive contacts will confirm anything. You need people with no stake in your feelings.
Third: not asking for introductions. Question 7 is how ten interviews become forty without any additional prospecting.
'I'm researching how people handle X. I'm not selling anything and I'd love to hear how it actually works for you.' This gives them permission to be blunt.
'The last time', 'that specific occasion'. Never 'usually' or 'would you'.
Aim to speak less than a quarter of the time. Silence after an answer usually produces the most useful sentence in the conversation.
The single strongest signal. Effort already spent is evidence that words cannot fake.
If they ask what you are building, say you are still figuring it out and offer to show them later. Then actually do.
At the end, always. This is how the sample grows without cold outreach.
Throughout validation, and periodically for ever. Aim for a floor of three a week early on.
This is not the script for a sales call or a usability test. Once you have a product and are trying to sell it, use a discovery call instead — different goal, different questions.
The best-known guidance on customer interviews converges on one rule: anchor every question in something that already happened. "Walk me through the last time this occurred" has an answer that exists independently of how the person feels about you; "would you use this?" does not.
Rewrite every hypothetical question as a historical one before the call. It is a mechanical change that transforms the quality of what you hear.
"That sounds really useful" is the most common outcome of a badly run interview and the least informative sentence a founder can collect. It costs the speaker nothing, it is socially expected, and founders reliably record it as validation because it is what they hoped to hear.
Count only what people have already done or already paid for. Enthusiasm is not a signal; it is politeness with good manners.
Create a free account to read the rest and apply it to your own startup.