When should you use a survey instead of interviews?

The two methods answer different questions

Interviews answer what and why. Where do people get stuck, what were they actually trying to do, what did they try before you. Small samples work because obstacles are shared — if four of six people cannot find the export button, the seventh will not change your mind.

Surveys answer how many and how much. What share of users hit this problem, which of these three options is most preferred, how does this differ by segment. They need enough respondents for the differences to exceed the noise.

Why the order is nearly always discover, then measure

A closed survey question can only return the options you wrote. If you ask why did you cancel and list price, missing features, switched to a competitor and other, you will get a tidy distribution — even when the real answer is that the onboarding email went to spam and half of them never activated. That reason was not on the list, so it appears nowhere, and other is the box people tick on the way out.

Interviews first tell you what the options are. The survey then tells you how big each one is. Reversed, the survey manufactures a finding out of your own assumptions and hands it back to you with a percentage attached, which is far more persuasive than it deserves to be.

When a survey genuinely is the right first move

• You need to size something you already understand well.
• You need to segment — the same known question across enough respondents to compare groups.
• You need a tracked measure over time, asked identically each period.
• The population is large and geographically spread and you need coverage, not depth.

Superhuman's product-market-fit survey is a good example of the format used correctly: one question, asked at scale, tracked over time, on a topic the team already understood qualitatively. The number was a thermometer, not a discovery tool.

Writing a survey that does not lie to you

Ask about behaviour, not intent. How many times did you do X last week beats how likely are you to do X.

One idea per question. Was the setup fast and easy is two questions and gets one confused answer.

Avoid leading framings. How much did our new dashboard improve your workflow assumes the answer.

Always include an open-text box. It is the only part of the instrument that can surprise you.

Watch who answers. Surveys are answered disproportionately by your most engaged and most annoyed users. The quiet middle, which is usually the majority, is systematically underrepresented, and a survey that ignores this will overstate both enthusiasm and outrage.

The combination that works

Interview six to eight people in a segment until the same obstacles repeat. Turn those obstacles into survey options. Field the survey to size them. Then go back and interview the group whose answer surprised you.

That loop is what continuous discovery looks like in practice, and neither method does the job alone.