35.4 Escalation logic and confidence thresholds

Overview and links for this section of the guide.

The Confidence Score

When you ask for classification, ask for a `confidence` score (0.0 to 1.0).

  • Confidence > 0.9: Auto-reply (if safe).
  • Confidence > 0.7: Draft a reply, but put it in "Drafts" for a human to approve.
  • Confidence < 0.7: Do nothing. Route to human inbox.

Escalation Triggers

Hard-code certain triggers to ALWAYS escalate:

  • Keywords: "lawyer", "sue", "illegal", "injury".
  • Sentiment: "Furious", "Angry".
  • Topic: "Data Breach", "Security Vulnerability".

The AI is a triage nurse. It should handle the scrapes and bruises, but call the doctor for the heart attacks.

Where to go next