Question 1

What happens to our conversation data?

Accepted Answer

Voiceloupe stores bot records, transcript records, and golden-script review records in the app’s persistence layer so your team can compare behavior over time. When you ask for a fix, the proposed-fix flow sends the relevant bot context and divergence to the Polsia-managed AI proxy for drafting; this repository does not make a broader retention, training, or compliance promise.

Question 2

How much do we need to change?

Accepted Answer

You do not need to rewrite the bot. Setup centers on connecting the existing bot and channel records, defining its brand-voice style guide and tone examples, capturing golden scripts, and comparing those expectations with real transcripts in the review workflow.

Question 3

Are we locked to one model?

Accepted Answer

In the current review workflow, the AI-assisted proposed-fix step uses the Polsia-managed AI proxy and defaults to `gpt-4o-mini`; there is no app-level model picker or fallback configured in this repository. The transcript-vs-golden comparison is deterministic, so model choice affects the suggested fix, not whether the underlying divergence is flagged.

Question 4

Can it change a bot without us?

Accepted Answer

No. Divergences land in the review inbox, proposed fixes create draft revisions, and `approved` is a distinct lifecycle state; Voiceloupe does not promise an automatic production deployment from that review step.

Question 5

How do we know it pays off?

Accepted Answer

The value is visible in the issues your team can catch earlier: dead ends, voice mismatches, broken fallback paths, and drift across the bot portfolio. Voiceloupe gives reviewers a repeatable golden-path baseline and a focused queue, so you can evaluate the payoff through review time saved and coverage that stays current rather than an invented benchmark or guaranteed savings figure.

The questions behind a better bot review loop.

Review the signal. Keep the call.