Best for
- Teams sitting on interview notes nobody has read twice
- Separating what users said from what the loudest user said
- Feeding real evidence into product decisions
What you give it
- The raw material: notes, transcripts, tickets, survey exports
- The decisions this research should inform
What you get back
- Findings ranked by strength of evidence, each with its supporting quotes and counts
- The contradictions and surprises called out, not averaged away
- A gap list: what the research cannot answer, so nobody pretends it does
How it works
- Reads everything first and codes statements to their source, so every claim stays traceable.
- Clusters observations into candidate patterns, counting evidence per pattern and per segment.
- Separates what people said, what they did, and what they were asked — these disagree, informatively.
- Promotes only supported patterns to findings; the rest are labelled hypotheses.
- Writes the result for the decision at hand, with quotes attached and limits stated.
Example
You: Here are 22 interview transcripts and three months of cancellation tickets. Why are people leaving?
Result: Five findings: the top one (present in 14 of 22 interviews and 61% of tickets) is not price but a setup step people fail silently; one surprise (long-tenure customers leave for a different reason than new ones); one contradiction flagged where interviews and tickets disagree — with the likely sampling cause named.
Limits — please read
- Synthesis cannot repair biased collection; sampling problems get named in the report.
- Counts of qualitative data indicate weight, not statistical proof — it says so.
- It informs decisions; choosing what to build remains yours.