Five Users Is Not a Discovery Plan

Choose interview participants around the decision you need to make. Five users is advice about usability testing, with specific assumptions behind it.

Suppose you have five interviews booked. Great. Who are those people? What can they help you decide? I want those answers before calling the research plan ready.

How many user interviews do you need for discovery?

It depends on the question's breadth and the people involved. A narrow study of similar work needs different coverage from a service spanning several roles. Start small, analyse as you go, and leave room for further recruitment. If you need a population percentage, use an appropriate quantitative study alongside the interviews.

I check who we are recruiting before agreeing to the interview count. Who does the work? Who handles the exceptions? Who might show us that our explanation is wrong?

That gives each interview a job to do. It also makes the budget easier to explain.

Where did the five-user advice come from?

Jakob Nielsen's usability-testing explanation describes a mathematical model for finding problems in an interface.

The model uses an average individual discovery probability of about 31%. It estimates roughly 85% of usability problems after five comparable participants. Those are model assumptions and estimates, rather than a guaranteed result for each test.

You improve the design between small rounds of testing. The participants need to be comparable and doing similar tasks. Different groups can require additional testing.

In discovery, I am often still figuring out the problem. There may be little to put in front of someone. An interview helps explain their experience. The usability model measures something else.

Carry the number across and you lose the reason it was useful in the first place.

What does interview research actually suggest?

As of October 2026, NN/g suggests starting with about five or six relevant participants, then analysing and adapting. This is a suggested starting batch. Further recruitment depends on what the analysis reveals.

Hennink and Kaiser's 2022 systematic review included 23 articles examining saturation. The empirical studies commonly reached it within 9 to 17 interviews, particularly with narrow objectives and relatively homogeneous populations.

That finding also has boundaries. Broader questions or more varied populations can need more research. Recognising a theme and understanding its different meanings are separate ambitions.

I use those findings to budget in stages. Swapping five for seventeen still leaves the same question: whose experience do we need to understand?

Who should you recruit first?

I start with differences that could change the decision. Take a hypothetical purchasing workflow. The person requesting a purchase and the person approving it see different parts of the work.

A regular requester may know the shortcuts. Someone buying an unusual service may hit an entirely different approval path. I want the recruitment plan to make room for those differences.

Write a simple recruitment matrix before sending invitations:

Relevant difference	What it could reveal
Frequent versus occasional use	Familiar shortcuts or confusing instructions
Routine versus unusual cases	Exceptions hidden by the standard process
Requester versus approver	Different needs at a handoff
Continued use versus abandonment	Reasons someone chose another route

These are example recruitment choices. Your question may call for different ones, including accessibility needs or differences in digital skills.

The sponsor's closest contacts are an easy place to start. I check who that leaves out. If recruitment is difficult, I record the gap so we know which views are missing.

How do you adapt the plan after interviews?

I review the material while recruitment is still open. Here is the sequence:

1. Record the themes supported by each session. Keep participant references beside the supporting material.
2. Note contradictions and missing situations. Separate a new theme from another example of an existing theme.
3. Identify the uncertainty that matters to the decision. Decide whether an interview is the best way to resolve it.
4. Recruit for the gap. Update the question if the original scope proves too broad.

Suppose regular requesters describe a clear process, but an occasional requester gets stuck. I speak with another occasional requester next. That gives us a way to investigate the difference.

If everyone reports waiting for approval, I look at the queue and speak with the approver. More accounts of waiting can add detail. We still need to understand the cause.

I keep a short decision log alongside the recruitment plan. When I change who we recruit, I write down why. Later, someone can follow the reasoning directly from the log.

What can a small interview study tell you?

A small study can show you where to look next. It can explain why someone struggles and expose an assumption to test before development. That is useful progress.

It cannot establish how common that problem is across your market. Avoid turning the share of interviewees mentioning something into a market percentage.

The distinction matters when the build depends on demand. You may need a quantitative study or a commercial experiment. Choose the method around the remaining question.

In Before the PRD, I explain how those findings connect to a product decision. Sampling is part of that evidence base.

When have you heard enough?

Ask whether further interviews are likely to change the decision in front of you. Check that the important circumstances are covered and that disagreements have an explanation or a follow-up plan.

I write the limit beside the decision. We might have enough evidence to prototype for one role, while demand across the market remains unresolved. Keep both conclusions visible.

Turning interviews into decisions also requires synthesis. Repeated conversations only become useful when the team compares them and decides what to do next.

We keep that connection through Discovery and Requirements in the process. The available offers describe ways to continue. If the research plan needs another view, book a call with the decision and proposed participant groups.

Choose the next interview because it can change what you do. That is the standard I use.

What else should you know about interview sample sizes?

Are five interviews enough for discovery?

Sometimes a narrow question needs little additional research. Five is a possible starting batch, rather than evidence that every important perspective is covered.

What does the 85% usability figure mean?

It is a model-based estimate of usability problems found with comparable test participants. It concerns usability problems, rather than market demand or interview coverage.

Should I recruit the same number for every persona?

Recruit around differences that could change the decision. A fixed quota for every persona can waste sessions while leaving important circumstances unexplored.

How do I know when to stop interviewing?

Review what new sessions add and check missing perspectives. Document unresolved uncertainty. Stop when further interviews are unlikely to change the current decision.