Enscale Docs
AI agents

AI handover

When your AI agent should stop and fetch a human — and how to make that handoff invisible to the customer.

Handover is when an AI agent stops and gives the conversation to a person.

Getting this right matters more than getting the AI's answers right. An agent that handles 90% well and escalates the rest cleanly beats one that attempts everything and occasionally invents your refund policy.

When agents hand over

ReasonWhy it matters
Not confident enoughBetter to escalate than guess
Customer asked for a personAlways honour this, immediately
Sensitive topicPayments, account changes, refunds, complaints
Repeated failureIt's tried, the customer is still stuck
Your own limitA turn count or time limit you set

Configure these per agent under AI Agents.

Make "talk to a human" actually work

If a customer types "agent", "human", or "speak to someone", the agent should hand over on that message. No clarifying question. No one-more-attempt.

A customer who asks for a person and gets another AI reply becomes an angry customer.

The cost of an unnecessary handover is a few minutes of someone's time. The cost of trapping a frustrated person in a loop with a bot is a public review you can't delete.

You can reinforce this with a Keyword Received trigger on AGENT or HUMAN that assigns the conversation directly, as a backstop.

What the person picking it up sees

A handed-over conversation lands in the inbox with the whole AI exchange visible. Your team member reads what's already been asked and answered, and carries on.

Making the customer repeat themselves is the most common way a handover goes wrong — and it's entirely avoidable, because the context is right there on screen. Worth telling your team explicitly: read up before you reply.

Route handovers automatically

Handover fires the AI Handed to Human trigger, and tells you why it escalated. That's what makes handover operational rather than just a status change:

Trigger: AI Handed to Human
  → Condition: was the reason a refund?
      → Yes: tag "urgent", assign to senior support

Other patterns worth building:

  • Notify a specific person when handover happens out of hours, so an angry customer at 10pm doesn't wait until morning.
  • Tag by reason, so you can see at a glance what your AI consistently can't handle.
  • Route billing questions to whoever handles billing, rather than the general queue.

Handover is free

A handed-over conversation is not an AI resolution, and you are not charged for it.

You're only charged when the AI solved the problem and closed the conversation itself. Handover, human takeover and idle auto-close are all free.

Which means over-escalating costs you nothing. Tune towards handing over too readily rather than too rarely — the failure mode on the other side is much more expensive.

Use handover reasons to improve the agent

The reasons your agent escalates are a to-do list. Look for patterns:

What you seeWhat to do
The same question over and overAdd it to your knowledge base
Handing over right after the greetingInstructions are too cautious, or there's too little information to work from
Never handing over at allCheck it escalates on sensitive topics — silence here is a risk, not a win
Escalating on something you did documentThe source may not be marked Ready, or its wording doesn't match how customers actually phrase it

That last one is common. You wrote "Returns Policy"; customers type "can I send this back". Both need to be in there.

After a handover

A human-owned conversation stays human-owned. The AI doesn't jump back in mid-thread, which would be jarring for everyone.

If the customer starts a fresh conversation later, the AI picks that one up as normal.

Next steps

On this page