How it works
The path from customer data to personas, simulations, and findings.
Fraser moves through four stages. Understanding them makes the output much easier to interpret, and much easier to improve.
1. Capture customer data#
Fraser starts from evidence about your real users: customer feedback, support tickets, call recordings, transcripts, and conversation notes you provide.
This step is optional (you can run Fraser with stock personas), but personas grounded in your own customer data behave far more like your actual users.
You are responsible for having a lawful basis and any required consent for the data you provide. For call recordings and transcripts specifically, you must have obtained all consents required to record and share them. See Authorization & limits.
2. Build personas#
Fraser analyzes that data for recurring topics, intents, pain points, objections, and feature requests, and turns them into reusable personas: a goal, a temperament, and a set of things the persona cares about and reacts to.
Personas persist. They're the stable unit you re-run against, which is what makes results comparable across builds. See Personas.
3. Run simulations#
A simulation puts one or more personas in front of a target, a URL, a flow, a prototype, or an Android build, with a goal to accomplish. Subagents drive the interaction and record what happened at each step: where the persona paused, backtracked, misread a label, or gave up.
See Simulations.
4. Get findings, and file them#
The run produces findings: specific, located observations rather than general opinions. Each names the screen, describes the behavior, and proposes a fix.
Findings roll up into a report, and each one can be filed to Jira or Linear as a ticket, with the suggested fix attached, without leaving Slack or Teams. See Findings & reports and Slack, Teams & tickets.
Why this order matters#
The quality of a run is bounded by the stage above it. Weak persona grounding produces generic findings. A vague goal produces a wandering simulation. If output feels shallow, move up the chain rather than re-running the same configuration.
Was this page helpful?