Five well-chosen people will find most of the severe usability problems in a single flow. Not all of them, and not the rare ones, but the ones that are costing you money this quarter. It takes an afternoon and it does not need a lab, a research platform, or a budget line. It needs a real task, a working thing, and the discipline to keep your mouth shut.
Why five
Because severity clusters. The problems that matter are the ones most people hit, which means the first two or three participants find them and participants four and five confirm them. Beyond that you are paying to hear the same thing again. If you have two genuinely different audiences, run five each; that is a different question from running fifteen of the same person.
What you need before anyone arrives
- One task, phrased as an outcome the participant wants, not a feature you want to show. “Get a refund for the second item in this order”, not “try the refunds page”.
- Something clickable. A rough prototype is fine. A slide of screenshots is not, because people cannot get stuck on a slide, and getting stuck is the data.
- Five participants who are plausibly your users and do not work at your company. Colleagues know too much and are too kind.
- A note-taker who is not the facilitator, and a recording if you have consent. One person cannot run the session and capture it.
- 01Thanks for doing this. We are testing the product, not you. If if something is confusing, that is us getting it wrong, and it is exactly what we need to hear.
- 02Please think aloud as you go. Tell me what you are looking at and what you expect to happen. I will mostly stay quiet.
- 03Before we start: last time you did something like this, how did it go? (Two minutes, no more.)
- 04Here is the task: [outcome the participant wants]. Take as long as you need. Start wherever you would normally start.
- 05When they pause: what are you thinking? What did you expect that to do? What would you do next if I were not here?
- 06When they ask you a question: I will answer that at the end. What what would you do if you had to guess?
- 07Afterwards: what was the most annoying part? If you could change one thing, what would it be? Anything you expected to find and did not?
- 08Thank them. Then, and only then, answer the questions they asked during the task.
How to run it without spoiling it
The facilitator has one job: do not help. Not by pointing, not by narrating, not by saying “so as you can see”. The moment you explain the interface, the session stops being a test of the interface and becomes a test of your explanation. Silence is uncomfortable and it is where the findings are. When someone gets stuck, let them stay stuck long enough to try something. What what they try tells you what they expected.
If you find yourself demoing, the session is over. You are now the interface, and you do not ship with the product.
What to do with what you got
Sort every observation by severity, not by frequency: did it stop them, slow them, or merely annoy them. Anything that stopped two or more people is not feedback, it is a defect, and it goes into the build. Anything one person disliked is an opinion until a second person hits it. Write the outcome as a ranked list with the decision each item forces. That that is the artefact people act on, as we found on Fren.
Where it usually goes wrong
- Recruiting people who already know the product. They cannot un-know it, and their fluency reads as your clarity.
- Asking whether people like it. Opinions on aesthetics are cheap and unreliable; watch behaviour and ask about expectation instead.
- Testing five flows in one session. One flow, five people. Tired participants are polite participants.
- Filing the results. If nothing enters the backlog within a week, you paid for entertainment.