How many customer interviews do you need to validate an idea?

The short answer

Five to ten interviews with people from one narrow segment is enough to see clear patterns; fifteen to twenty if the answers conflict. The real stop signal is saturation: when the last three interviews told you nothing new, more interviews will not either. Ten sharp interviews beat fifty polite ones.

Founders overestimate this number the same way they overestimate how long validation takes, and for the same reason: a bigger number postpones the verdict. The research on qualitative interviews is unusually consistent, and it says most of the insight arrives embarrassingly early.

Why the number is smaller than you think

Within one tightly defined segment, the same handful of pains, objections, and current workarounds start repeating by interview five or six. Researchers call it saturation, and you will feel it: you can predict the next answer before it comes. When that happens, stop. The exception is when answers genuinely conflict, which usually means you are interviewing two different segments and calling them one; split them and do five more in each half.

What actually decides it

Three things move the number. Segment tightness: 'freelance designers who invoice monthly' saturates in seven interviews, 'small businesses' never saturates at all. Problem severity: if nobody in the first five interviews describes the problem without being prompted, you do not need fifteen more interviews, you need a different problem. And question quality: leading questions produce agreeable noise at any sample size, which is why the customer interview questions matter more than the count.

Two warnings. Interviews are evidence about the problem, not the product: people describing a pain honestly is real signal, people saying they would use your solution is not, as covered in what counts as real validation. And ten interviews you actually do beat twenty you keep planning; in GRILLR, customer interviews are dated, graded tasks in the validation phase, and 'I posted in a few groups' does not pass as proof that they happened.