Plan

To decide what a test is for before somebody sits down in front of the prototype. Sessions run without this produce a lot of observations and no agreement about what they meant.

Use it for every round, including the informal ones. The plan takes twenty minutes and it is what makes five sessions conclusive rather than five sessions interesting.

Write the question first, then the task that would answer it. A task is something the person tries to do, not a screen you show them. "Book a table for four on Friday" rather than "have a look at the booking screen".

Decide who you need, how many, and what counts as a pass, before the first session. Then write down what you will change if they fail and what you will change if they succeed. Sessions where both answers are the same are sessions you did not need to run.

The task leads. "Use the filter to narrow the results" tells them the filter exists and where it is, and the session then proves only that a person can follow an instruction. Say what they are trying to achieve and stop talking.

The second failure is testing with people who already know the product, which measures familiarity rather than design. The third is the debrief in the corridor, where the loudest observation wins and the quiet one is never written down. Agree the read-out format before the first session and fill it in after each one, while it is still specific.