Live conversation outcome / predicted-CSAT scoring
Predictive scoring for a currently-open conversation: resolution_likelihood (0-1), escalation_risk (0-1), predicted_csat (1-5, before any survey), a confidence, the live drivers behind the score, and a single recommended_action (monitor | coach_agent | proactive_escalate | reassign | offer_resolution). Powered by the platform LLM over the redacted live transcript + the conversation’s operational signals + the contact’s deterministic scores. Short-cached (90s, busts on each new message); ?refresh=1 forces a recompute. Returns 200 with null scores + a note when the LLM is unavailable or the thread is too thin; 404 when the conversation does not exist.