Interview synthesis
Synthesis is the part of user research that takes a week and gets skipped under deadline. An agent can do the first pass, and the discipline is that every theme carries the quotes it came from so a researcher can disagree with it properly.
The job on one screen
| Runs when | A research round finishes. |
|---|---|
| Reads | Transcripts, the research questions, participant segments. |
| Decides | Which statements group together, and how strong each theme is. |
| Produces | Themes with counts, supporting quotes, and the segments they came from. |
| Stops when | Themes produced. Interpretation is the researcher's job. |
| Tools it needs | Transcript store, clustering, quote index. |
| Autonomy to start at | Assist. It prepares, a person decides and acts. |
Optional: load a real model
Run it
The routing rules
These are the exact buckets the demo above scores against. Change the wording of a ticket and you can watch the confidence move.
| Bucket | Signals it looks for | What happens next |
|---|---|---|
| Reliability | slow, failed, crash, timeout, unreliable, gave up, takes so long, error | Theme: people stop using features that fail unpredictably, even when the feature works most of the time. |
| Discoverability | did not know, did not realise, found out, showed me, hidden, could not find, existed | Theme: capability exists but is invisible. Usually the cheapest problem to fix and the most commonly ignored. |
| Workarounds | spreadsheet, manually, workaround, copy, export to, our own, next to, instead | Theme: users route around the product. Every workaround is a specification written by someone who needed it. |
| Pricing confusion | pricing, plan, cost, confusing, which plan, billing, upgrade, expensive | Theme: commercial friction. Confusion about what someone is paying for shows up in churn long before it shows up in support. |
| Below the bar | no clear signal | A quote that fits no theme is kept as an outlier rather than forced into the nearest bucket. Outliers are where the surprising findings live. |
Where this one goes wrong
Themes without quotes
A theme you cannot evidence is the model's prior, not your users' opinion.
Counting participants wrong
Ten quotes from one talkative participant is one participant. Count people, not statements.
Losing the outliers
The one participant who described a completely different workflow is often the most valuable interview in the set.
How you would know it is working
| Measure | Why that one |
|---|---|
| Themes a researcher keeps | The quality signal, and it is qualitative on purpose. |
| Quotes per theme | Under three and it is not a theme yet. |
| Time to first synthesis | The saving, typically a week to an afternoon. |
Earning more rope
| Assist | First pass themes for a researcher to rework. Correct home. |
| Approve | Themes plus a draft findings document with quotes embedded. |
| Auto | Never present synthesis as finished research. The interpretation needs a person who talked to the participants. |
Keep the outlier pile. An agent optimises for the biggest clusters, and the finding that changes a roadmap is usually sitting in the pile of quotes that fit nowhere.
Related: Knowledge base gaps · Exit interview themes · Literature sweep · all agent jobs · Agent Lab home
Related: Knowledge base gaps · Exit interview themes · Literature sweep · all agent jobs · Agent Lab home
Free from AI School - no signup, everything runs in your browser.