To synthesise customer interviews, separate what participants said or did from your interpretation, code observations against the research questions, compare patterns across relevant participants, and write findings with evidence, scope and uncertainty attached. Do not count memorable quotations as conclusions. A useful synthesis helps a team make a named decision while allowing another person to trace the reasoning back to the source.
Prepare evidence before interpretation
Analyse soon after each round while context remains fresh. Standardise notes enough to compare sessions, but retain the participant’s situation, decision stage and recruitment source.
For each session, capture:
- participant code and relevant attributes;
- recent trigger or event discussed;
- previous approach and considered alternatives;
- decision criteria, objections and constraints;
- experienced or expected outcomes;
- exact language worth preserving;
- researcher observations;
- uncertainties and follow-up questions.
Remove unnecessary identifiers and apply your consent and retention rules. If the interviews have not happened yet, use customer interview questions for better positioning to collect event-based accounts instead of abstract opinions.
Use the observation–pattern–finding chain
Keep three levels distinct:
- Observation: something that occurred or was said in a session.
- Pattern: a meaningful similarity or difference across observations.
- Finding: your evidence-backed explanation and why it matters to the decision.
For example, “P04 exported reports to a spreadsheet before sending them to a director” is an observation. “Four finance leads reconstructed the output before internal review” is a pattern within that group. “Current reporting does not provide review-ready evidence for finance leaders” may be a finding, provided the notes support it and the scope is stated.
The GOV.UK analysis guide recommends extracting observations, grouping them into themes and then determining findings and actions. It also encourages involving observers, which can reduce reliance on one person’s memory or preferred interpretation.
Build an evidence matrix
Create rows for participants and columns for decision-relevant themes: trigger, alternative, desired progress, perceived difference, proof, anxiety and outcome. Place concise evidence in each cell, with a link to the underlying note or timestamp.
Then ask:
- Which patterns recur within the priority audience?
- Which accounts contradict the apparent pattern?
- Does the difference follow role, stage, company size or another condition?
- Are we seeing behaviour, recollection or speculation?
- What did we not recruit for?
- Which claim requires another method?
Do not turn every repeated phrase into a theme. “Easy” may refer to setup, day-to-day use, approval or procurement. Preserve the object and context.
Illustrative example
Suppose six recent buyers of an invoicing tool mention “control”. A shallow synthesis labels control as the primary value. A closer comparison shows three agency owners mean seeing overdue accounts, two finance managers mean preventing unauthorised changes, and one freelancer means knowing when to follow up.
The defensible finding is not that all customers want the same control. It is that the word masks several jobs that may require distinct messages. This example is illustrative; it is not a customer study.
Write findings teams can use
A strong finding contains:
- statement: what you learned in plain language;
- scope: who and which situation it applies to;
- evidence: linked observations, not only a quotation;
- meaning: why it affects the decision;
- confidence: what supports or limits the interpretation;
- action: decide, test or research next.
Label hypotheses separately from findings. Record negative evidence as well as confirming evidence. If one participant’s account changes the decision because it reveals a serious risk, say that explicitly rather than pretending frequency is the only basis for importance.
Use findings to improve the language choices in how to use customer language in messaging, but do not paste quotations into marketing without permission and contextual review.
Limitations and trade-offs
Affinity mapping is good for collective sense-making, but clusters can create false neatness. Automated transcription saves time but can mishear specialist language and names. AI-assisted coding may help organise a large corpus, but it can strip context, invent patterns or expose personal information if governance is weak. Human review and traceability remain essential.
Qualitative frequency is not market prevalence. A pattern across five interviews can guide a positioning hypothesis; it cannot establish a percentage. State the sample and recruitment limitations beside the finding.
Frequently asked questions
Should we transcribe every interview?
Not always. Detailed notes with timestamps may be enough for a focused round. Transcripts help when language is central or several analysts need access, but increase privacy, storage and review work.
How should we handle contradictory interviews?
Do not average them away. Check whether the difference follows a meaningful condition, exposes a missing segment or challenges the question. Record the contradiction and decide whether another round is needed.
Can an AI tool synthesise the interviews?
It can assist with retrieval or provisional coding if your consent, privacy and security arrangements allow it. It should not be the sole judge of meaning. Verify every cited passage and keep the human decision trail.
Your useful next step
Select three completed interviews and build a small evidence matrix before analysing the whole round. Write one finding using the six-part structure above, then ask a colleague to trace it back to the notes. Store the approved finding and its limitations in a customer research repository rather than leaving it in a presentation.