Home / Platform / Personas & custom tests

Average callers are not the problem

An agent tested on cooperative, fluent, quiet-room callers will pass. The calls that cost you a customer are the other ones, and they are not rare in aggregate. Prologue builds a suite per persona so the difficult cases are measured separately instead of being averaged away by the easy ones.

Personas that map to real callers

• Hard of hearing. Repeats, asks for repeats, needs pace and volume handled rather than ignored.
• Code-switching. Changes language mid-sentence, which routinely breaks transcription before the model ever sees the turn.
• Escalating. Starts annoyed and gets worse, testing whether the agent holds tone under pressure.
• Silent line. Says nothing, so the agent has to decide whether it is being ignored, whether the call dropped, or whether something is wrong.
• Emergency. The call where getting it wrong is the whole risk.

Custom tests for your situations

The personas above are a starting point, not the list. The suites that matter are the ones written for your product: the intents you actually handle, the policies you have to follow, the phrases your callers use, the edge cases support already knows about. Every failure found in production becomes a test, so the same regression cannot ship twice.

Each suite gets its own bar

A single aggregate score hides exactly the failures you are looking for. Personas are scored and gated independently, so an agent that is excellent with easy callers and poor with an escalating one reports as what it is.

How they run

Suites are graded with the same rubrics used for evals and run automatically through CI/CD testing on web and mobile, on every deploy.