How do I find people to interview before launch?

Search for the problem rather than your solution — Reddit, niche forums, app-store reviews and G2 complaints about incumbent tools. People already complaining in public have pre-qualified themselves and reply more often than cold contacts, because someone finally engaged with their frustration. After that: communities around competing tools, second-degree introductions to a specific job title, then cold outreach.

How many customer interviews do I need before building?

Ten to fifteen usually reveals a pattern. Question design matters far more than count — five interviews asking about past behaviour beat thirty asking whether someone would use your product. If two-thirds describe the same specific pain and can say what it cost them, you have a signal.

What questions should I never ask in a customer interview?

Anything asking someone to predict their own future behaviour: would you use this, would you pay for this, do you like this idea. These reliably produce polite agreement because they cost the respondent nothing. Ask instead about the last time the problem occurred, what they did, and what it cost them.

Can AI replace real customer interviews?

No. AI can rehearse your questions, surface leading phrasing and generate the objections you should be ready for — genuinely useful before you spend scarce interview slots. It cannot produce revealed preference, because no model can decide to give you money. Treat simulated panels as rehearsal and real conversations as evidence.

How do I tell real interest from politeness?

Ask what they have already spent. People with real pain have paid for a partial fix, built a workaround, or can name what the problem cost them with a number or a date. Politeness sounds like 'that sounds useful' and 'send me a link' — enthusiasm that never references the person's own situation.

Customer interviews when you have no users yet

Pre-launch interviews fail in a specific and predictable way. The founder describes the idea, the participant says it sounds useful, everyone leaves encouraged, and nothing has been learned. Repeat ten times and you have a folder of enthusiasm and no evidence.

The cause is not bad interviewing technique. It is that you asked people to predict their own future behaviour, which humans are reliably bad at.

Finding people to talk to with no user base

The usual objection to pre-launch interviews is that there is nobody to interview. In practice the constraint is willingness to ask, not availability. In rough order of signal quality per hour spent:

People already complaining in public. Search the problem — not your solution — on Reddit, niche forums, app-store reviews, G2, X. Message the people describing the pain. They have pre-qualified themselves and often reply, because someone finally cared.
Adjacent-tool communities. Wherever users of the incumbent product gather to complain about it. Highest-intent group available and the least contested.
Your second-degree network. Not friends — friends are polite. Ask contacts for introductions to people with the job title, which converts a favour into a real conversation.
Cold outreach to the exact role. Low response rate, unbeatable precision. Fifteen well-targeted messages typically yields two or three conversations.
Paid panels. Fast and expensive. Save this for when you know exactly which question you are buying an answer to.

The question set

Every question below is designed to elicit memory of past behaviour rather than prediction of future behaviour. Past behaviour actually happened, so it cannot be flattered into existence.

Ask this	Not this	Why
When did you last run into this? Walk me through it.	Would you use a tool that solved this?	The first recalls an event. The second invites a courtesy.
What did you do about it?	How would you want it solved?	Action taken is the strongest available signal of real pain.
What did that cost you — time, money, or a customer?	How much would you pay for this?	Cost of the status quo is verifiable. Hypothetical price is not.
What have you already tried?	Do you like this approach?	Prior attempts prove budget and urgency existed.
Who else on your team deals with this?	Are you the decision-maker?	Reveals the buying unit without prompting a title claim.
What would have to be true for you to switch?	Would you switch to us?	Surfaces the real blocker instead of a yes.

If a participant has never taken any action about the problem, you have learned something important: it is not painful enough to pay to fix.

Reading the transcript honestly

Politeness looks like signal until you learn to sort it. Two lists, and the difference between them is whether the person has already spent something.

Real signal

They have already paid for a partial solution, or built an internal workaround.
They can name what the problem cost, with a number or a date.
They interrupt you to add detail — a sign you touched something live.
They ask when they can have it, unprompted.
They offer to introduce you to a colleague with the same problem.

Politeness

“That sounds really useful.”
“I'd definitely try it.”
“Send me a link when it's ready.”
Broad agreement with everything you say.
Enthusiasm that never references their own situation.

Rehearsing before you spend real conversations

Interview slots are your scarcest resource and the first two are usually wasted on discovering that the questions were wrong. That is what a simulated panel is for: Truewick runs your questions past a panel of AI respondents built from your target persona and distils the answers into themes and objections.

Be clear about what that is worth. It will not tell you whether real people will pay — a model cannot generate revealed preference, and treating simulated answers as validation reproduces exactly the error this guide is about. What it does is expose leading questions, vague phrasing and unstated assumptions before you burn a real conversation on them. Rehearsal, not evidence.

How to run customer interviews when you have no users yet