How many customer interviews do you need before building?

How many customer interviews do you need? Fewer than you fear if they're the right people, and more than one round if the first one is messy. Most founders should plan for 10 to 20 well-screened conversations per hypothesis-testing round. You don't need 100.

Four things drive that number.

Why the number is smaller than you expect

Customer discovery isn't a statistical exercise. You aren't estimating a population parameter to three decimal places; you're checking whether a clear pattern exists.

Say you talk to 12 people who really fit your segment, and 9 of them describe the same problem unprompted, spend money on the same workaround, and light up at the same moment. That's a signal, and a 13th interview won't change it. Qualitative research hits diminishing returns quickly because you're listening for themes, and themes repeat.

What determines how many you need

Segment tightness

A narrow, well-defined segment ("freelance designers who invoice 5+ clients a month") reaches saturation in 8 to 12 interviews. A broad one ("small business owners") might take 30 and still feel noisy, because you're really interviewing several segments at once. If every conversation heads somewhere different, tighten the segment before you add interviews.

Signal strength

A strong signal is obvious by interview 8. A borderline one, where some people care and some don't, needs a second round with more respondents so you can tell whether it's a real sub-segment or noise. Weak, mixed signal across 20 good interviews is a result in its own right. See when to kill a startup idea.

Screener quality

Ten screened interviews beat fifty unscreened ones. A screener that removes people who don't have the problem makes every remaining conversation more informative. Without one, you spend interviews 15 through 40 relearning that some people just aren't your customer.

Number of hypotheses

You're testing problem, willingness to pay, behavior, solution fit, and market. Each one doesn't need its own 15-person round, since a well-built interview covers several, but the riskiest one or two deserve their own focus. Plan 10 to 20 per round, and expect two or three rounds over the life of the idea.

A practical plan

Round	Focus	Interviews	Outcome
1	Problem + current workaround	10-15	Is the problem real and painful?
2	Willingness to pay + behavior	12-18	Would they pay and switch?
3 (if needed)	Solution fit on a concept	8-12	Does your approach beat the status quo?

If round 1 comes back weak, you stop. That's 10 to 15 interviews to save six months, which is why you validate before you build.

Signs you've done enough

• The last three interviews taught you nothing new.
• You can predict what the next respondent will say, and you're right.
• Each hypothesis is clearly confirmed or clearly broken.

Signs you need more

• Every interview contradicts the last one.
• The pattern is 50/50 and you can't tell why.
• You realize halfway through that you've been interviewing the wrong segment.

When you're in the second list, fix the targeting and run another focused round. Grinding out 40 more of the same interviews won't help.

Common questions

Is 5 interviews ever enough?

With a very tight segment and an obvious signal, five can justify a second round. Five is rarely enough to make a build-or-kill call on its own.

Do surveys change the number?

Yes. Once you can write precise questions, a screened survey of 30 to 50 checks whether the interview pattern holds at scale. Interviews find the pattern and surveys tell you how big it is.

What if I can't reach my target customers?

That's a finding. A segment that's hard to reach for research is usually expensive to reach for sales too, so factor it into the verdict.

Should the founder do the interviews personally?

For the first rounds, yes. You'll hear nuance a hired interviewer would miss, and it recalibrates your instincts faster than reading someone else's notes.

Run a round without scheduling 15 calls

Hypothis builds a screened research round from your hypotheses and collects responses asynchronously from people you recruit, with adaptive follow-ups standing in for the live probe. You watch the pattern form and get a scored verdict for each hypothesis. It's free during early access. See how it works.