Prepare the sources and question
Choose a research question before summarizing. Record each source, account, participant role, date and permission to process. Mark missing transcript sections. For notes, say which parts are direct quotations and which are paraphrases.
Keep participants from the same account distinguishable without counting them as independent buying decisions. Remove unnecessary personal details before the source goes to a model, while preserving the context needed for attribution.
Extract observations before themes
In Customer Discovery, ingest the interview and inspect evidence attribution. Ask whether each extracted observation answers the research question and whether its wording is stronger than the source.
Keep the situation, action and consequence together. Requested features belong in the record, but should not erase the problem the participant was trying to solve.
- Synthetic note
- I check the export before I send it, because the last report had an old figure.
- Observation
- The participant checks exported figures before sharing.
- Consequence
- An older figure previously reached a report.
- Unproven hypothesis
- All customers need an automated approval workflow.
Compare explanations and contradictions
If one participant checks accuracy and another asks who may send the report, keep those jobs separate. A recurring word such as review is not enough to establish a shared requirement.
- Select the evidence and interviews to include, with their segment and time scope.
- Group observations by the problem or situation, not only by a matching keyword.
- List supporting sources and distinct accounts for each candidate theme.
- Find an observation that does not fit and consider an alternative explanation.
- Keep unanswered questions visible in the synthesis.
Review an AI analysis against the inputs
Inspect a source for every material claim. Check that a paraphrase has not become a quotation, a guess has not become a fact and an absent customer has not appeared in the account count. Resolve contradictions explicitly; do not ask for a cleaner summary that quietly removes them.
When using jobs-to-be-done or pain analysis, treat the output as an interpretation to test. The next customer conversation can show whether the explanation fits the person's actual behavior.
Copyable working template
Candidate theme:
Supporting source references:
Distinct accounts and segment:
Contradictory source:
Alternative explanation:
Next question or observation:Choose the next interview from the gap
A useful research plan targets what the current evidence cannot distinguish. In the synthetic report example, ask the participant to describe the last incorrect report and how review happened. Look for the sequence of work rather than asking whether an approval feature sounds useful.
Save the extraction, the synthesis and your review separately. In Oqera, choose the next manual research action and the sources it may use. This method illustrates how to review the work; it does not claim a live run or a validated customer outcome.
Access and review
Closed Beta requires an application, manual approval, invitation and verified login. Applying does not grant immediate access. Six ready-made teams, guarded Custom teams and the Apps catalog are included. No Oqera payment or subscription is required. Bring your own AI provider key and pay the provider directly. External actions require specific permission and review; unattended schedules and triggers are off.