From Observation to a Testable Problem

Research leaves a board full of observations. The temptation is to choose the most moving quotation and generalize it, or turn the first complaint into a feature request. Synthesis instead organizes specific facts and states what they do and do not support. Put one unit of evidence on each card: source, date, channel, precise action or wording, and point in the journey. Do not merge distinct accounts into “users are confused.” If you saw a participant reopen a confirmation three times before a visit, that is behaviour. Claiming that the person feared rejection is an interpretation that needs a question or another source. Precision lets a colleague challenge the inference without having attended the session.

Cluster observations by the mechanism of friction, not only by participant type. Information can be lost between the booking page and the message, while a separate problem occurs at reception. Use a table with observation, possible interpretation, supporting evidence, competing evidence and next test. Four people asking about a document before their appointment may point to vague wording or to an office rule change. Comparing the page and message versions with the document list on each visit date distinguishes these explanations. Four interviews do not establish a service-wide rate. In the Design Council's exploration and definition cycle, defining is a narrowing of the question based on learning, with permission to revisit the frame when evidence changes.

A need is not a wish for a particular tool. A useful formulation is: “A person in this situation needs to achieve this goal because this evidenced obstacle affects the outcome.” State where the evidence ends and the hypothesis starts. For example: “A visitor who booked on mobile needs a clear description of documents before travelling; some people we interviewed brought a version the office could not accept. We need to find whether the term was unclear or the checklist was not updated.” That wording leaves open a visual definition of the document, booking assistance, or a repair to the list-update process. “The visitor needs a chatbot” would make research a justification of a chosen solution.

Test the quality of a problem statement with two questions. Could a team that has not selected a solution suggest two different responses? “We need an app” fails that test. What new observation would change the statement itself? If entrance directions were present and understood in all three cases but a door closure forced a detour, an operational access issue would precede message copy. This is a real counterexample because a checkable fact changes the decision. Not every uncertainty needs an open-ended research project; a team can make a bounded decision and record what remains unknown.

Before leaving definition, write four sentences: user scope, supported observation, provisional interpretation and a discriminating test. Add the decision owner and their control boundary. The message team can change copy; building operations controls entrance signs; testing both routes requires coordination. If the log counts enquiries rather than people, retain that denominator when comparing later observations. This work reduces the risk of prototyping something that was never the main problem. The frame remains provisional. A revisable statement is more useful than a reassuring sentence that no evidence could disprove.