When an AI agent should hand off
The common triggers are: the customer asks for a person; the question is not covered by the knowledge base or connected systems; the request needs an action the AI is not allowed to take, or goes past a limit such as a maximum discount; and the issue is sensitive enough that a person should decide, such as a complaint or a refund.
Handing off too rarely leaves customers stuck with an AI that cannot help. Handing off too often means the AI is only answering the easy questions. The useful question is not how many conversations were handed off, but which kinds, and why.
What a good handoff includes
The person who picks up should see the whole conversation, a short summary of what the customer wants and what the AI already did, and enough context about the customer (orders, past conversations) to act straight away. The handoff should stay on the channel the customer chose, so they are not asked to call or email instead.
Routing matters as much as context. A handoff that lands in a queue nobody watches is worse than no AI at all, so most teams assign handoffs to an available agent or a dedicated team, and alert someone when one sits unclaimed.
How to measure handoffs
Track the share of conversations the AI hands off, and group handoffs by reason or topic. A topic that keeps growing usually points to a gap in the knowledge base or an action worth automating. Track the time from handoff to the first human reply as well, since that is the wait the customer actually feels.