Interviews

How to run customer interviews that tell you the truth

A customer interview is a conversation about the other person's life so far, not about your idea. It becomes reliable when you keep three rules: talk about their routine rather than your plan, ask about past behaviour rather than future intent, and speak for less than a third of the time. Five questions fill twenty minutes comfortably. What counts as evidence at the end is not praise but what somebody spends on you: time, their reputation, or money. Ten conversations run this way tell you more than a hundred survey responses.

The friendly yes proves nothing

Describe your idea and then ask whether it is any good, and you will almost always hear yes. The yes is sincerely meant and still worthless. Two things are working against you at once: hardly anyone will tell an enthusiastic person to their face that their plan is weak, and the question is about the future, where people estimate their own behaviour far too generously.

The way out is not to ask harder, it is to ask differently. Every question in this article points backwards, at something that has already happened and therefore cannot be talked up. That shift is the entire method, and it goes back to Rob Fitzpatrick's "The Mom Test": ask so that even your own mother could not give you a falsely encouraging answer.

Same subject, two questions

Would you pay for a mobile bike repair service?

What did you do the last time your bike was out of action?

The first question asks someone to guess, the second asks them to remember. What a person did can be checked. What a person would do is politeness in the future tense.

What would something like this be worth to you?

What did that breakdown cost you, in money and in time?

A wished-for price is invented, the cost of the problem was lived through. A price can be derived from the cost. It cannot be derived from a wish.

Do you think the idea is any good?

Who else you know has this problem? Would you introduce me?

Agreement costs nothing. An introduction costs the other person some of their reputation, and that is exactly why it means something.

The bars illustrate how much an answer proves, they are not measured data. Every pair moves in the same direction: away from an opinion about your idea, towards behaviour that already happened.

Three rules that carry the conversation

They sound modest and are hard to keep, because all three run against the impulse that brought you into the room.

1Talk about their life, not about your idea

As long as your idea is not in the room, there is nothing for anyone to be polite about. Ask about routines, irritations and workarounds from the past month. The moment you explain what you want to build, the conversation changes role: a witness turns into a guest who would rather not disappoint you.

The most common breach of this rule happens in minute two, out of nerves.

2Ask about what happened, not about what would happen

Past behaviour can be checked, future intent is an estimate, and people estimate their own future behaviour far too generously. Any question starting with "would you" can be rebuilt into one starting with "when did you last".

"I would use that straight away" is a compliment, not a commitment.

3Talk for less than a third of the time

A good interview sounds boring from the outside: the founder asks short questions and then goes quiet. The most useful sentences almost always arrive in the pause that opens up when you do not immediately speak after an answer. Two seconds of silence is enough.

Fill the silence and you get your own assumption back, confirmed by the other person.

Where the list of twenty names comes from: how to define your target audience in four steps →

The five questions

In this order, and worded roughly like this. Each one measures something the previous one leaves open, and none of them mentions your idea.

1Talk me through how you do this today.

Measures: The actual routine

The opener, and the only question deliberately left wide. It produces the routine in real steps rather than in the tidied-up version people tell once they know where you are heading. Listen for detours above all: a second spreadsheet, a call to a friend, a note on paper. Every detour marks a place where the current solution does not reach.

Do not interrupt and do not correct, however clumsy the routine sounds.

2When did that last go wrong? What happened next?

Measures: Frequency and consequences

This question puts a date into the conversation, and a date is where checking begins. If the answer is "last week", you are looking at a recurring problem. If it is "once, about three years ago", you have an irritation, not a business. Then ask what happened immediately afterwards, because that is where the follow-on costs sit.

"It happens constantly" is not an answer. Ask for the last specific instance.

3What did that cost you, in money and in time?

Measures: The size of the problem

Two numbers will do: the amount spent and the hours lost. Both are estimates and both are still useful, because they set an order of magnitude. A problem that costs thirty dollars and half an hour will not carry a forty-nine dollar price. One that costs a working day carries far more than most founders dare to ask for.

Do not ask what your offer is worth. Ask what the problem costs.

4What have you already tried about it?

Measures: Urgency

The sharpest of the five, because it is the only one that cannot be answered with words alone. Anyone who has already searched, asked, bought or rigged something up has proved the problem before you asked. Anyone who can name nothing here has described a problem they can live with. People who live with their problems rarely buy a solution to them.

"I did once look whether something like that exists" counts as a no, not half a yes.

5Who else you know deals with this? Would you introduce me?

Measures: Spread and seriousness

The close of every conversation, for two reasons. It shows whether the problem reaches beyond this one person, and it is the only test in the interview that costs anything: recommending you puts the other person's reputation on the line. As a side effect, it solves the problem of finding ten interviewees by itself after the second conversation.

A no is a result. Write it down instead of reading it as a scheduling issue.

Five is a ceiling, not a target. A conversation that only gets through three questions because the second one opened up a whole story is a good conversation, not an incomplete one.

In five steps to ten conversations

Two weeks of calendar time and around six hours of work, most of it spent on the parts that happen before and after the conversation.

Collect twenty names before you contact anyone

30 minutes

Write down twenty people who belong to your chosen audience, sorted by closeness: your own contact, a contact of a contact, a stranger. Twenty, because roughly half will not reply and you want to land at ten conversations. Friends and family do not belong on this list, they cannot take the role of a witness.

Result: a list of twenty names with one sentence each on why this person is affected.

Ask without mentioning the idea

1 hour

The request names the topic, not the venture: you are trying to understand how somebody handles a particular thing today, you need twenty minutes and you are selling nothing. That wording raises the acceptance rate and protects the conversation at the same time, because the other person walks in without expecting a solution.

Result: six to twelve confirmed slots, spread over two weeks.

Run twenty minutes, no longer

20 minutes per conversation

Five questions fit comfortably into twenty minutes if you keep to the one-third rule. The short duration is not a compromise, it is the reason people say yes. If time is left over, go back into the last specific incident in more detail rather than inventing a sixth question.

Result: a conversation in which you did not explain your idea.

Write it up within the hour, word for word

10 minutes per conversation

Record what was said, not what you conclude from it. Verbatim sentences in quotation marks, plus the numbers and the date of the last incident. Your interpretation goes separately underneath. After ten conversations that separation is the whole difference between an analysis and a confirmation of what you already believed.

Result: half a page per conversation, quotes above, interpretation below.

Only analyse after all ten

1 hour

Lay the ten write-ups side by side and look for sentences that repeat, for the range of the costs named, and for how many people had already tried something. A pattern needs at least three independent mentions. Anything appearing once is one person's view and stays that way until it shows up a second time.

Result: three to five patterns, each backed by a quote, and a decision on what you test next.

What counts as evidence

A signal is worth exactly as much as it costs the person giving it. There are three currencies, and they rank in a fixed order: time, reputation, money. Anything that costs none of the three is conversation, however warm it sounded.

Signal	What it costs them	What it is worth
"Great idea, the market needs this."	Nothing	No evidence. A compliment ends the conversation instead of moving it on.
"Let me know when it's ready."	Nothing	No evidence. A polite goodbye that feels like an order.
Shows you their current workaround unprompted	Time and some candour	Weak evidence. The problem is real enough to talk about.
Books a second meeting, in the calendar	Time	Weak evidence. Time is the cheapest of the three currencies, but not free.
Introduces you to another affected person	Reputation	Strong evidence. Nobody passes on something they consider irrelevant.
Pays a deposit, pre-orders or signs a letter of intent	Money	Strongest evidence. The only commitment that cannot be withdrawn unnoticed.

Where the quotes from ten conversations go next: the Value Proposition Canvas, making offer and customer meet →

Notes, recordings and what you may keep

Written notes are the low-risk default and usually the better one. Keep what was said, not who said it: a first name and the role are enough to make sense of a quote later. Anything that identifies a person beyond that adds no analytical value and does add an obligation.

Recording is a different matter. An audio or video recording processes personal data, so it needs the other person's prior and freely given consent, and they have to know what it is for and how long you keep it. In practice the recording also costs you something in the conversation itself, because people become more careful the moment a red dot is on. If you record anyway, delete the file once the write-up is done.

General orientation, not legal advice. Rules differ by country and change; check what applies where you conduct the conversations.

Five mistakes that waste the conversation

Each of them produces a conversation that felt good. None of them produces one you can act on.

Showing the idea in the first few minutes. From that moment on the other person is assessing you rather than their own problem. If you must show it, do it at the end, after the fifth question, and explicitly as a separate part of the conversation.

Asking about the future and treating the answer as a commitment. "I would", "I could see myself" and "definitely" are declarations of intent grammatically and courtesies in practice. Nothing is evidenced until somebody spends time, reputation or money.

Only speaking to people who wish you well. Friends, family and anyone who likes you will reliably give the wrong answer, with the best of intentions. They are useful for rehearsing the questions, not for testing the idea.

Summarising instead of quoting. Writing "loved the idea" after a conversation destroys the information and keeps the interpretation. A verbatim quote can be reassessed in four weeks, a summary cannot.

Stopping after the first enthusiastic conversation. A single good conversation is the most seductive form of too little data. Ten conversations are not statistical proof, but they are the point at which repetition becomes distinguishable from coincidence.

Frequently asked questions

How many customer interviews do I need?

Ten per audience is the usual guide for a first round. Ten, because a pattern becomes distinguishable from coincidence at roughly three independent mentions, and because answers typically start repeating between the seventh and tenth conversation. If nothing repeats after ten, that is a result too: the group is cut too wide and contains people with different problems.

How do I find interviewees when I know nobody?

Three routes that need no budget. First, wherever the audience already is: clubs, specialist forums, events, waiting areas, workplaces. Second, through the fifth question of every conversation, which hands you the next contact, so the chain sustains itself after two conversations. Third, a short public request that names the topic and explicitly sells nothing. The wording decides: asking for twenty minutes about somebody's daily routine gets far more acceptances than offering to present your venture.

Am I allowed to show my idea in the interview at all?

At the end yes, at the start no. The five questions aim at the other person's past behaviour, and that behaviour changes as soon as a solution is in the room. If you need feedback on the idea, split the conversation visibly: the five questions first, then a pause, then say that you would now like to show something. Everything said after that goes into its own section of your notes and counts for less, because it was produced under social pressure.

Should I record the conversation?

Only with prior consent, and usually it is not necessary. A recording processes personal data and makes people more guarded, precisely on the questions about cost and failure. Notes are enough: verbatim quotes during the conversation, the rest within an hour afterwards. If you do record, ask first, say what for, and delete the file once you have analysed it. General orientation, not legal advice.

Everyone likes the idea but nobody buys. What does that mean?

That you collected opinions rather than evidence. Agreement is free, which is why there is so much of it. Check your ten write-ups for three things: did anyone name a specific recent incident with a date? Could anyone put a number on the cost in money or time? Had anyone already tried something to get rid of the problem? If all three are missing, the problem is real but not urgent, and a problem that is not urgent does not get paid for. At that point changing the segment is worth more than changing the offer.