When to move from surveys to interviews

You wrote a survey, got two hundred responses, and still don't know what to build. This happens more than anyone likes to admit. Surveys are cheap, they scale, they produce tidy charts — and they cap out at the end of your own imagination, because every multiple-choice question is a list of answers you already thought of.

Interviews don't have that ceiling. The cost is volume: you'll never talk to two hundred people. But one or two hours of interviews reveal what two hundred survey responses can't.

What surveys can't do

Reveal language you haven't heard before. Your multiple-choice options are your language. If your participants use different words to describe the same problem, your survey turns them invisible.

Uncover behavior versus stated opinion. People answer "what do you do" with what they remember doing, which is correlated with what they actually do but isn't the same thing. Watching someone work tells you what the survey couldn't.

Surface edge cases. The power-user with a weird workflow won't fit into your answer choices; they'll pick the closest option and move on. You'll never know they exist.

Three signals you need interviews

1. Your answer-choice fields don't cover most responses. If the "Other, please specify" field contains twenty distinct themes, the structured part of your survey is hiding more than it reveals.

2. The open-ended answers contradict the structured ones. When "How satisfied are you?" skews positive but the free-text comments are full of frustration, people are being polite in the structured questions and honest in the open ones. Skip the politeness layer.

3. You can't predict what people will say. If you and a teammate disagree on what the most common answer will be — and both of you have watched these users work — you don't have enough context to write a survey yet. Talk to people first.

How many interviews

Five interviews will surface roughly 80% of the themes that matter for most product decisions. The diminishing-returns curve is steep: the sixth interview rarely tells you something the first five didn't hint at.

Bump to twelve to fifteen when your user base is heterogeneous — say, enterprise admins and individual contributors who use the same product for different reasons. You're effectively running two parallel studies; each needs five.

The hybrid: a survey that routes into interviews

The best version of this isn't a choice — it's a sequence. Send a short survey. On the last page, for respondents whose answers looked interesting, offer a calendar link to a fifteen-minute follow-up. The survey does the filtering; the interview does the learning.

The rate is usually between five and ten percent. That means 300 survey responses will net you 15–30 interview candidates, which is more than enough for a useful study.

The takeaway

Surveys and interviews aren't rivals; they're sequential. Surveys find the pattern. Interviews explain it. Skipping either one usually means shipping something that tests well and flops anyway.