Agents
Escalation to a human
Letting the agent hand off to a teammate or close out a conversation.
Turning it on
In an agent's Settings tab, under Escalation & ticketing, check "Let the AI escalate to a human or mark a conversation resolved." This gives the model two abilities during a conversation: escalate to a human, and mark a conversation resolved, which it uses at its own judgment (e.g. when a customer is frustrated, asks for a person explicitly, or the request is outside what it can safely handle).
What happens when the agent escalates
An escalated conversation shows up in the Ops tab under "Needs attention" — the same triage queue used for conversations a teammate takes over manually. See Conversations & Ops for how to work that queue.
If escalation is off
The model will never escalate on its own, but nothing stops a teammate from manually taking over any open conversation from Ops at any time — escalation being off only controls whether the AI decides to hand off, not whether a human can step in.
Where this shows up in reporting
Whether a conversation was escalated or resolved feeds the escalation-rate and resolution-rate figures on the organization-wide Analytics page (not yet on the per-agent Analytics page, where those are still listed as upcoming metrics) — see Organization analytics.