> ## Documentation Index
> Fetch the complete documentation index at: https://docs.cura.so/llms.txt
> Use this file to discover all available pages before exploring further.

# Escalation: when the AI hands off to a human

> Your AI is built to know its limits.

Your AI is built to know its limits. **Escalation** is the moment it steps back and hands a conversation to a person on your team — so customers always reach a human when they need one. This article explains when escalation happens and how to set it up.

The goal is simple: the AI handles the routine, and your team is there for everything else.

***

## What escalation is

**Escalation** is the AI flagging a conversation for a human. It's different from a **handoff** or **take over**, where *you* choose to step in. With escalation, the AI itself decides it shouldn't keep replying and passes the conversation on.

When a conversation is escalated, the AI stops replying so a person can take it from there.

## When the AI escalates

The AI hands off automatically when:

* **A customer asks for a human** — "Can I speak to someone?"
* **It can't help** — the question is outside what it knows or can do.
* **Something goes wrong** — an error means it can't complete the task reliably.

You can add your *own* triggers on top of these in the **Escalation & handoff** field of [Advanced AI instructions](/ai-agent/advanced-ai-instructions) — for example, escalating when a customer is upset or asks about a refund.

***

## Set up who receives escalations

So escalations reach the right person, set a designated team member to handle them.

<Steps>
  <Step title="Open your workspace settings">
    * Open the **workspace** and find its **Escalation** settings.
  </Step>

  <Step title="Choose who handles escalations">
    * Pick the **team member** who should receive escalated conversations. They'll be the point of contact when the AI hands off.

    <Frame>
      <img src="https://mintcdn.com/cura-2d9288ab/pPBYeI1qPGlrN2Ia/images/settings/organization-settings.png?fit=max&auto=format&n=pPBYeI1qPGlrN2Ia&q=85&s=fb073aeb63bcf688d2074667ff1a20cd" alt="Choosing the escalation contact in Cura's organization settings" width="2880" height="1560" data-path="images/settings/organization-settings.png" />
    </Frame>

    <Check>Escalations now route to the right person, so nothing slips through the cracks.</Check>
  </Step>

  <Step title="Add your own escalation triggers">
    * In [Advanced AI instructions](/ai-agent/advanced-ai-instructions), use the **Escalation & handoff** field to describe extra situations that should always go to a human.

    <Tip>
      Be specific about *when* to escalate. "Hand off if the customer mentions a complication or asks for a refund" works far better than "escalate difficult cases."
    </Tip>

    <Note>
      A multi-location business may need the customer to choose a branch before the AI can escalate — it'll ask which location they want first, then hand off.
    </Note>
  </Step>
</Steps>

***

## Troubleshooting

* **No one is receiving escalations.** Confirm a team member is set as the escalation contact in the workspace settings.
* **The AI escalates too often.** Tighten your Escalation & handoff instructions, and make sure your Knowledge Base, FAQs, and Services cover the questions it's giving up on.
* **The AI keeps replying when it should stop.** Once escalated, the AI stops; if it didn't escalate, check whether the trigger you expected is described clearly in your instructions.

***

### Related articles

* [Advanced AI instructions](/ai-agent/advanced-ai-instructions)
* [The Master AI switch and per-conversation control](/ai-agent/master-ai-switch)
* [How your AI agent works](/ai-agent/how-your-ai-agent-works)