Setting up escalation to a human

The most important thing you configure. Where the conversation goes, when, and what the customer is told.

The most important thing you configure. Where the conversation goes, when, and what the customer is told.

Set it up

1. Open the AI agent's settings.
2. Set what triggers a handover.
3. Choose where the conversation goes: a group, or routing that decides.
4. Write what the customer is told at that moment.
5. Test it from several points in a conversation.

Configure this before anything else

An AI agent that cannot hand over is a trap. Every other setting can be imperfect at launch; this one cannot.

What should trigger it

• The customer asks. Immediately, in whatever words.
• Two failed attempts at understanding. Not five.
• Subjects you have excluded, such as complaints or cancellations.
• Signs of distress or anger in what the customer writes.

Hand over early

The instinct is to let the AI try a little longer, because each attempt might succeed. From the customer's side, each attempt is another minute spent not getting help, and the resentment is cumulative.

An early handover costs one automated resolution. A late one costs the relationship.

Send the transcript with it

The agent picking it up must see what was already said. Making the customer repeat themselves after a failed automated exchange is the single most complained-about moment in any deployment.

Say what is happening

That they are being passed to a person, and roughly how long. Silence after a handover feels like being dropped.

Decide what happens out of hours

Where nobody is available, say so and take a message. A handover into an empty queue at midnight is worse than the AI saying it cannot help.

See also

• Viewing and editing system replies
• Can I stop my AI agent handing over to a human
• Conversation design: how to build a good exchange