How to run a usability test with five users
This is for anyone who owns a flow that is quietly underperforming. An account opening form. A claims submission. A checkout. You suspect something is wrong, but the only evidence you have is a drop in the funnel and a few complaints.
Five users, one task, two weeks. That is usually enough to find what is breaking.
Here is how I run it, and where five people stop being enough.
Pick one task, not the whole product
On a normal working day this starts as "can we test the app". Someone books a room, invites eight stakeholders, and the scope grows until the test covers onboarding, the dashboard, payments and the help centre. Then the sessions run long, nobody finishes anything, and the findings are a list of forty small things with no order to them.
Narrow it before you write a single line of script. Whole product, then one flow, then one task. Pick the task the business actually needs to work. Open the account. File the claim. Pay the bill.
One task means every one of your five people does the same thing, so you can compare them. If four out of five stall at the same screen, that is a finding with weight behind it. If everyone did something different, all you have is five opinions.
A good test of scope: can one person do it in under fifteen minutes when it works properly? If not, cut it down further.
Recruit people who match
The fastest way to waste two weeks is to test with whoever is available. A colleague from another floor. Someone's cousin. A friendly customer who has used the product for six years and could fill the form in with their eyes shut.
Five of the wrong people tell you nothing. They will be polite, they will find the buttons, and you will walk away thinking the flow is fine.
Screen for the real user. The person who has the problem today and has not solved it yet. If you are testing account opening, you want people who do not have an account. If you are testing a claim, you want people who have had a reason to claim.
Recruit ten to land five. People cancel. Some fail the screener once you talk to them. Write three or four screening questions, ask them by phone or by short form, and be strict. The extra hour you spend here decides whether the whole study is worth anything.
Write a script that does not lead
Here is what leading sounds like in a live session. "Now tap Continue." "You'll see the menu at the top right." "Most people go to Settings here." The facilitator is nervous about silence, so they help. The recording then shows five people successfully following instructions, which proves nothing.
Give them the goal. Let them find the way. "Pay your bill" instead of "tap Continue now". Then stop talking.
Ask what they expected before they tap. Ask what they think will happen next. Then stay quiet while they work it out. Silence is the tool. Most people will start narrating on their own after a few seconds, and that narration is the data.
Write the script out in full, including your opening lines and your debrief questions. Read it aloud once before the first session. If any line contains a button name, a menu name or a step number, rewrite it.
Run the sessions and record
Thirty to forty minutes each. Screen recorded, face recorded if the person agrees. One facilitator, one note taker, nobody else in the room unless they can keep quiet.
The structure is simple. Warm up, task, debrief. The warm up is two or three questions about how they handle this today, which also settles their nerves. The task is the middle twenty minutes. The debrief is where you ask what they would tell a friend about it.
Watch where they pause. That is the whole skill. Someone will tell you the screen is perfectly clear and still sit there for a full minute doing nothing. What people say and what people do come apart constantly, and the recording is how you catch it.
Take timestamps as you go. Ten seconds of note, not a paragraph. You want to be able to find the moment again without scrubbing through forty minutes of video.
Cut a reel and a number
Most studies die as a slide deck nobody opens. Pull the clips instead. Find the places where all five people got stuck in the same spot, and cut those moments together into a few minutes. A room full of executives watching three real customers fail the same field will move faster than any summary you could write.
Then write down one baseline number. Task completion, four out of five. Or time on task, an average in minutes and seconds. Pick one and keep it simple.
The number matters because of what happens next. You fix the thing, you run the same test with five new people, and you see whether it moved. Without a baseline, "did it work?" gets answered with a shrug and a feeling.
Be honest about what five people can and cannot do. Five users find the obvious breaks, and they find them fast. They will not give you percentages you can put in front of a board. If you serve very different user groups, five is too few and you need a bigger study. Say so out loud rather than overclaiming.
What to do this week
- Choose the one task your business most needs to work, and write it in a single sentence.
- Draft four screening questions that separate real users from people who already know the product.
- Write the script with no button names in it, and read it aloud once.
- Book five sessions of forty minutes, and recruit ten people to fill them.
- Decide your baseline number now, before you run anything, so you have something to beat later.
