Human handoff escalation rules
The moment an AI assistant should say 'let me get someone to help' is exactly the moment most poorly-designed bots keep trying anyway, frustrating the customer further. This playbook sets clear, specific rules for when to hand off - tone signals, repeated questions, complaint severity - so escalation happens reliably.
Define your escalation signals
Common ones: the same question asked twice without resolution, negative sentiment detected in the guest's message, or specific complaint keywords.
Set the handoff threshold
Decide how many failed attempts or how strong a negative signal triggers an automatic handoff rather than another AI attempt.
Make the handoff itself smooth
Pass full conversation context to the human agent so the guest never has to repeat themselves after escalation - that's often the more frustrating part.
Worth knowing
A bot that escalates too late frustrates customers; one that escalates too early defeats the purpose of automating in the first place. Tune thresholds based on real conversation transcripts, not guesses.