Escalation to a human.
When a call needs a person (a real dispute, a frustrated customer, something out of scope) the teammate hands off with the full context and puts it in your open-items queue. Escalating is success, not failure.
Goal
Get the hard cases to a human quickly and cleanly, with everything they need already in hand, so nothing stalls and no customer is stuck arguing with an AI. Graceful escalation is the correct behaviour, logged as its own outcome.
Triggers
- The AI hits a genuine dispute, detects frustration, or is asked something out of scope
- An unrecoverable tool error, or a customer explicitly asking for a human
- A lapsed promise or an item that's exhausted its cadence without resolution
What the AI does
- Offers a human callback and closes the call warmly rather than looping
- Logs the outcome as ESCALATED and creates an open item in your queue
- Hands the human the full context: transcript, conversation memory, holds, and the live Xero data
- Flags the reason (dispute / frustration / out-of-scope / tool error) so you can triage at a glance
- Keeps the AI + recording disclosure and compliance intact right up to the handoff
What the AI won't do
- Argue an amount or a dispute to 'win' the call
- Pretend a problem is resolved when it isn't
- Bury the handoff: every escalation surfaces in the Briefing and open items
Tools available to the AI
transfer_to_staffcreate_open_itemschedule_callbackrecord_outcomeTools are typed and bounded: the AI calls them, it doesn't define them. Each tool call is logged with an idempotency key, so retries never produce duplicates.
Success looks like
Toggle
Always on across every workflow: escalation is a safety behaviour, not a togglable campaign. Choose which staff mobiles receive handoffs under Dashboard → Team.
Related
- All workflows
- Configuring the AI (what each workflow reads from your tenant config).
- Approval mode & graduation (how this workflow behaves in Approval Required vs. Monitored).
- Compliance & privacy (disclosure, DNCR, kill switches).