Open AI Supervisor
The illustrated scorecard shows how one reply can pass factual and policy checks while still needing a missing customer question answered. Open the real scorecard to see its exact evidence.
Workspace managers can open AI Supervisor from the main navigation. The page includes a workspace scorecard, quality by human or AI agent, and reply-level review evidence.
klaxify automatically reviews a new public human reply when it is saved. Select Review recent replies to backfill the scorecard from recent tenant conversations. Internal notes are never treated as customer replies.
Understand the five scores
- Factual accuracy checks completed commerce claims against verified order context and events.
- Policy adherence detects unsafe guidance and prohibited claims.
- Tone and empathy identifies blaming, dismissive, or needlessly cold language.
- Resolution completeness checks whether the response addresses the detected customer intent and gives a useful next step.
- Safe promises flags unverified arrival guarantees and other promises outside current evidence.
The first pass is deterministic and costs no external AI credits. A score is a review aid, not an automatic disciplinary decision.
Review evidence and add coaching
Open a reply scorecard to see the original response, dimension scores, and exact findings. Add a manager coaching note that explains the approved approach for future replies, then select Save coaching. If the reply is acceptable after review, select Mark reviewed.
Coaching resolution, manager identity, time, and agent name are saved in the tenant audit history. AI Supervisor does not send a message, edit a ticket, or execute a commerce action.
Treat high-risk findings as a review queue
Replies with an unsupported refund, cancellation, shipment claim, risky medical language, or an unverified customer promise require manager review. klaxify keeps the underlying evidence visible so the manager can decide whether to correct the customer conversation, update a policy, improve knowledge, or coach the agent.