> For the complete documentation index, see [llms.txt](https://doc.aissist.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://doc.aissist.io/capabilities/escalation-and-handover.md).

# Escalation and Handover

How Aissist detects when a conversation needs a person, hands it to the right team with the context attached, and steps back.

*Last updated: October 25, 2026*

Aissist knows when a conversation needs a person. This page explains how escalation is detected, what happens in your help desk when it is, and how to set the rules.

### Why it matters for CSAT

A good handover is part of good service:

* **Customers reach a person at the right moment** instead of going round in circles with an AI that cannot help.
* **The AI knows when to step back.** It does not keep answering once your team owns the conversation.
* **Your agent gets the context.** A summary and next step are waiting in the ticket, so the customer does not repeat themselves.
* **It goes to the right team.** Billing goes to billing, returns to returns.

### How escalation is detected

Escalation detection is built in and runs on every AI reply. It checks for:

1. the customer asking for a person or to be transferred
2. the AI saying it will transfer, escalate, notify someone, check with someone, or that someone will follow up, investigate or send something later
3. the AI saying it does not know, lacks the information or access, or cannot do what was asked (out of context)

A transfer that is still **conditional** does not escalate. If the AI is still collecting an order number, or offers to transfer "if the steps do not work", the conversation stays open until the condition is met.

You can also escalate after a set number of interactions (see below).

### What you configure

#### In the workspace: **Workspace → Escalation**

* **Customized Escalation Instruction (Optional)** - narrows or widens what counts as an escalation, for example "Only escalate when the user has a payment issue." It only decides whether a reply is an escalation; it does not change what the AI says. Keep it aligned with your instructions and sub-agents.
* **Escalation Summary (Optional)** - a JSON template for the note your team receives, so it captures what they need:

```json
{
  "order_number": "the order number, if provided",
  "issue": "what the customer needs, in one sentence",
  "already_tried": "steps the customer already took"
}
```

* **Suspend Afterwards** - the AI stops replying once it has escalated, so your team fully owns the conversation.
* **Escalate by Max # of Interactions** - **Maximum Interactions (default 20)** escalates a conversation that runs too long, with an optional **Message to User Before Escalation**.

Decide **when** to hand over in the instructions of the workspace or, better, each sub-agent: for example, "collect the order number and the reason, then tell the customer the returns team will follow up". See [Sub Agents](https://doc.aissist.io/capabilities/sub-agents.md).

#### In the gateway: where it goes

Open the gateway in **Deploy → Gateways**. In the **Escalation** section:

* **Escalation** - pick the default team, group, inbox or queue (depending on the platform) that receives escalated conversations. If you leave it empty, the AI unassigns itself and leaves the routing to you.
* **Handover Rules** - map one or more sub-agents to their own target, so each workflow reaches the right team. On some platforms, such as Intercom, a rule can also create a ticket.
* **Handover Message** - sent to the customer when the AI hands over, or when it has no reply to give, for example "Connecting you with someone now." It is rephrased in the customer's language.

See [Deploy](https://doc.aissist.io/tutorial/deploy.md) for each platform.

### What happens in the help desk

When a conversation escalates:

* the `sys_escalation` tag is added
* an internal note with the summary and next step is left for your team
* the conversation is assigned to the matched handover rule's target, or to the escalation target
* the handover message is sent, if enabled

By default the AI does not process conversations assigned to someone else, so it stays out of the way while a person handles it. You can also build your own help desk rules on the `sys_escalation` tag, such as notifying a team, moving tagged conversations to a shared inbox, or adding them to a shared view.

<figure><img src="https://301624521-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FyKHM836KGDAfAWvw6XL6%2Fuploads%2FXQwBE4OTvHW0IB2weK9r%2FScreenshot%202026-05-14%20at%2010.38.09%E2%80%AFAM.png?alt=media&amp;token=c08d79a4-19d2-42cd-a426-aa2e071ab35e" alt=""><figcaption><p>A help desk rule that assigns conversations tagged sys_escalation to a team</p></figcaption></figure>

### With Ask Engine

[Ask Engine](https://doc.aissist.io/capabilities/ask.md) tells you why conversations were handed over and what would let the AI finish the job:

> Why were conversations escalated last week, and which sub-agents escalate most?

> For the Refund sub-agent, what is missing that would let the AI finish instead of handing over?

Ask Engine counts every escalation in the period, reads the recorded reasons, and suggests what to add: an asset, an action, or an instruction.

### Best practice

* Define handover in each sub-agent: what to collect first, and what to tell the customer.
* Avoid vague triggers such as "may need review". Say exactly when to hand over.
* Use **Escalation Summary** for the fields your team always asks for.
* Turn on **Suspend Afterwards** when people should fully own escalated conversations.
* Review escalation reasons weekly. Many handovers point to a missing asset or action.

### Related

* [Gateways](https://doc.aissist.io/capabilities/gateways.md)
* [Sub Agents](https://doc.aissist.io/capabilities/sub-agents.md)
* [Success Metrics](https://doc.aissist.io/success-metrics.md)
