Glossary

Agent Escalation

The process and triggers by which an AI voice agent hands off to a human agent — defined by rules around complexity, sentiment, or explicit caller request.

Agent escalation is the critical design boundary between "calls AI handles" and "calls humans handle". Trigger conditions typically include: explicit caller request ("can I speak with a person?"), emotional escalation (anger, distress, panic detected from vocal cues + language patterns), defined business rules (always escalate cancellations, always escalate $X+ refunds, always escalate VIP callers), intent ambiguity beyond a confidence threshold, or topic outside AI scope (legal advice, medical advice, complex troubleshooting). Done well, escalation feels seamless to the caller; done poorly, it feels like the AI is "passing the buck".

Why it matters

  • Escalation design is the difference between "AI you trust your customers with" and "AI you don’t".
  • Right escalation rules protect customer trust and brand reputation.
  • Sentiment-based escalation catches the moments before frustration becomes a public complaint.
  • Rule-based escalation enforces business policy (cancellations, refunds, VIPs) consistently.
  • Warm-transfer escalation with full context cuts AHT on escalated calls by 30–50%.