01 / Understand what you need to find out
Start with the choice ahead: what needs to be decided, by whom and when? Agree which audience and task matter, what is already known and what would change the team’s mind. A study should reduce a particular uncertainty rather than attempt to answer every possible product question.
02 / Choose the method
An expert review examines an interface. Interviews investigate needs and experience. Usability testing observes people attempting tasks. Accessibility evaluation examines barriers within an agreed sample and method. We select the approach for the evidence gap and explain what it will not establish.
03 / Agree access and participant arrangements
Before research begins, the plan identifies the people involved, recruitment criteria, access needs, participant information, consent, recording choices and who may see the material. Use safe test data and environments. Retention and deletion arrangements are agreed for that engagement rather than implied by a generic website promise.
04 / Record what happens
Capture the task, page or prototype version, relevant state and observation. Keep source material separate from interpretation. A screenshot may establish what an interface displayed; it does not establish how every user understands it. AI-assisted organisation or synthetic perspectives are labelled and never counted as human research.
05 / Explain the finding
- What happened: the observation and the material that supports it.
- Why it matters: how the problem could affect someone using the service.
- Priority: how serious the problem is and how confident we are in the finding.
- Action: the proposed correction or next research question.
- Follow-up: how to check whether a change has helped.
06 / Help your team decide what to do next
The report should help a product, content or development team decide what to do. We provide the agreed evidence and recommendations, explain their limits and identify questions still open. Your team chooses the implementer. Retesting is a separately defined activity against the changed service, not a promise that a recommendation has already worked.