Glossary term
Glossary term
Governance and Compliance
Transfer of a case or decision from AI to a human when defined thresholds are breached.
Zendesk's AI triage agent uses sentiment-score escalation: if customer sentiment drops below -0.7 or a message contains escalation keywords, the conversation is transferred to a human agent within 30 seconds.
An insurance claims AI escalates to a human adjuster when the claim amount exceeds $25,000, the damage description contains disputed-liability language, or the claimant has filed 3+ prior claims.
Anthropic's Claude for Enterprise has a configurable 'human review required' flag - deployments set topics (medical diagnosis, legal advice) that trigger an automatic escalation message redirecting users to a licensed professional.