Start with the decision
Research should be commissioned because a decision is uncertain. Name that decision, who will make it and what evidence could alter the choice. ‘Understand our audience’ is too broad. ‘Decide which onboarding explanation to test first’ provides a boundary. List current assumptions and their risk if wrong. This creates a research plan that can finish and helps participants understand why particular questions matter without steering their answers.
Recruit for relevant experience
Demographics alone rarely identify the right participants. Recruit people who recently attempted the task, abandoned it, chose an alternative or support others through it. Define inclusion and exclusion criteria before outreach. Avoid relying entirely on current enthusiastic customers, who may have learned around problems newcomers still face. A small varied sample cannot estimate population prevalence, but it can reveal mechanisms, language and failure points worth testing at scale.
Interview for episodes, not opinions
Ask participants to reconstruct a recent event: what triggered it, what they did first, what information they noticed and where they hesitated. Concrete episodes produce better evidence than predictions about what someone ‘would probably’ do. Follow the sequence and ask for artefacts when appropriate, such as a redacted note or screenshot. Do not reward a preferred answer with praise or jump in to explain the product. Silence often gives the participant room to remember.
Synthesize without flattening
After each session, capture observations, quotes, interpretations and open questions separately. Group evidence by behaviour or obstacle rather than by participant type alone. A theme should point back to multiple specific observations. Preserve contradictions; they may indicate different contexts rather than bad data. Write findings as conditional statements describing when and why something occurs, not as universal personas that claim every member of a segment behaves the same way.
Translate findings into options
Research is useful when it changes what the team can do. For each finding, list the affected decision, plausible response and risk of acting too quickly. Convert uncertain responses into experiments. If people overlook a critical condition, test a clearer placement rather than assuming the entire flow needs rebuilding. Rank opportunities by consequence, confidence and effort. The research report should make traceability easy from recommendation back to evidence.
Close the loop
Share what was decided, what remains uncertain and when the evidence should be revisited. Participants’ contexts change, and research from a different market or product stage may expire. Store consent, recordings and notes according to the promised retention policy. Remove unnecessary identifying data. A lightweight research repository is valuable only if entries include dates, scope and limitations; otherwise old findings can acquire more authority than they deserve.
Worked example: abandonment at a comparison step
Assume analytics show people leaving a comparison step, but the reason is unclear. Recruit recent completers, abandoners and support staff with direct exposure to the journey. Ask participants to reconstruct what they expected, which conditions they compared and the moment they changed course. Pair interviews with a usability observation of the existing step. Synthesis may reveal two mechanisms: some people cannot distinguish options, while others need a condition disclosed earlier. These findings support separate tests. They do not justify a single persona called the confused user or a claim that every abandonment has the same cause.
Research failure modes
Leading questions and selective notes can make assumptions look like evidence. Avoid asking whether a proposed feature is helpful before understanding the underlying episode. Do not count repeated wording from a discussion guide as an emergent theme. Separate direct observation from interpretation and invite another reviewer to trace major findings back to notes. Recruitment through one enthusiastic channel can conceal important contexts. Report who was not represented, which sessions were incomplete and where moderator behaviour may have influenced answers. Limitations make research more usable because decision makers know where not to generalise.
Maintenance rhythm
Give findings an evidence date, product context and review trigger. Revisit high-consequence assumptions after major audience, policy or workflow changes. Keep the repository searchable by decision and behaviour, not only project name. Remove identifying details that are unnecessary for reuse, and honour the promised retention period for recordings. When a later study contradicts an earlier one, preserve both contexts rather than silently overwriting history. A healthy repository does not accumulate timeless truths; it shows how evidence, conditions and decisions changed.
Focused workshop
Use this short working session to turn the guide into a decision-ready artefact. Keep the scope to one live example and record assumptions beside the output.
- Write one decision statement and three risky assumptions.
- Create a six-question episode-based interview guide.
- Turn one hypothetical finding into two testable product or content options.
Definition of ready: another practitioner can inspect the evidence, understand the proposed change and name the result that would stop or reverse it.
Review questions
What evidence supports the problem? Which audience and context does it describe? What is the smallest change that can test the mechanism? Which signal is close enough to the decision to be useful? What harm could a single success metric hide? Record answers before implementation, then return to them when the review window closes.