Sample frameDay 293·Product · Usability

Morning. The instruction that makes this work is the one nobody can follow: don't help them.

Product.

The Five-User Test

5 min read·Apply by lunch

The question

Have you ever sat behind someone who'd never seen your product, handed them a real task and then stayed silent while they failed at it — or is everything you believe about how the thing gets used assembled entirely from your own hands on your own keyboard?

The idea

In 1993 Jakob Nielsen and Thomas Landauer presented a paper at the INTERCHI conference in Amsterdam under an unglamorous title — a mathematical model of the finding of usability problems — and it changed how software gets tested. Averaged across a large set of real projects, one person watched using an interface turns up roughly 31 per cent of its usability problems. Each extra person re-finds most of what the last one found, so the curve flattens fast: five people surface around 85 per cent of what's there, and the sixth and seventh mainly hand you things you've already written down. Nielsen made the practical case in a 2000 essay, and the argument was economic as much as statistical — several small tests through a build beat one big one at the end, because you're fixing while fixing is still cheap. Laura Faulkner stress-tested the claim in 2003, running sixty users on one interface and then drawing random sets of five from the pool: some sets found 99 per cent of the problems, others found 55. Five is a good average and a shaky single roll, which is an argument for running the test again rather than running it bigger. Sitting there while someone flounders feels like bad manners, so you explain what the button does — and destroy the only data the morning was going to produce. That is the trap, and it arrives dressed as politeness.

The framework

01
Five people from one group, not five people in total.
The number holds for one kind of user. If your product serves an administrator and an end customer who do genuinely different jobs inside it, they'll trip on different things, and five of each is closer to right than five between them. The same caveat applies to anything quantitative: if you want to know how long a task takes or what proportion of people complete it, five is nowhere near enough, and Nielsen never claimed otherwise. Five is a number for finding problems, not for measuring anything. Recruit by the job the person does, not by how easy they are to book.
02
Write tasks, not instructions.
A session is only as good as what you ask people to do, and most briefs quietly contain the answer. A task that reads open the app, go to Settings, then change your plan tests only whether someone can follow you reading aloud. A task that reads you're on the wrong plan and you want the cheaper one tests the product. Write each one as a situation with a goal and no route: a real reason, a real outcome, and none of the words that appear in your own navigation. If your task names a label that's visible on screen, you've handed over the answer.
03
Can you sit through ninety seconds of someone struggling?
This is the entire skill, and it's physically uncomfortable. When a person stalls, the instinct is to point, and the moment you point you've swapped a finding for a favour. Sit on your hands. If you have to speak, ask two things and nothing else: what are you trying to do, and what did you expect would happen. Ask people to narrate as they go — a habit borrowed from think-aloud protocols in cognitive psychology — so you hear the reasoning behind the wrong click rather than guessing at it afterwards. Count your own interventions when the session ends; a good one has almost none.
04
The output is a ranked list, not a report.
Nielsen's argument was never about one test; it was about running several small ones through a build. So the deliverable isn't a deck. It's a list of what broke, ordered by how badly it blocked the task and how many of the five hit it, with the top three assigned to a name and a date before the day ends. Then you fix those and watch five more people — a different five — a fortnight later. A team that runs four rounds of five learns more than a team that runs one round of twenty, and it learns it while changing things is still cheap.

Members only · 4 principles + template + AI mentor

364 more frameworks are waiting.

The Five-User Test is Day 1 of 365. One framework every morning for a year — across strategy, sales, negotiation, leadership, and more.

$1 /day

Billed $365/yr · cancel any time

365 frameworks

12 topics · templates · AI mentor

Reserve your place — $365/yr →

One payment a year · the whole library · no auto-renewal.

More Product frameworks

All 22 product frameworks →