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

# The Escalations tab

> Group recurring escalations into issues, read the AI write up of what is going wrong, and close the gap rather than the conversation.

Individual escalations tell you that one guest question went unanswered. The [Escalations tab](https://base.conduit.ai/ws/~/escalations) tells you *why it keeps happening*.

Rather than listing flagged conversations one by one, it groups them into **Issues**: recurring gaps behind many escalations at once. Each issue is written up for you, with an explanation of the pattern and a set of concrete recommendations for closing it.

This is the surface to work when your goal is raising automation rather than clearing today's queue.

## Issues

Open **Escalations** from the left hand navigation panel. The **Issues** list sits on the left, split across three tabs:

| Tab | What it holds |
| - | - |
| **Pending** | Issues that have not been dealt with yet |
| **Dismissed** | Issues you decided not to act on |
| **Resolved** | Issues you have closed |

Each issue in the list shows its title, a one line summary, its category, its status, and the number of conversations behind it. Filter controls at the top of the list narrow it down.

<Tip>
  The conversation count is the priority signal. An issue sitting behind 28 conversations is worth more of your time than four separate issues behind 9 each, because closing it removes the most future escalations for the same effort.
</Tip>

## Reading an issue

Selecting an issue opens it with two tabs, **Details** and **Conversations**, plus the issue reference, its age, and the conversation count in the header.

The **Details** tab is written by the AI from the escalations it grouped, and has three parts:

**Overview** states the pattern in plain terms. For example, guests want to stay past checkout, but staff have no reliable way to confirm whether cleaning schedules and same day turnover allow it.

**How to fix** gives specific recommendations rather than a generic prompt to add knowledge. For the example above, that might mean documenting a late checkout approval process tied to turnover status, creating a standard response covering when to confirm with operations and when to decline for back to back bookings, and setting an escalation path for same day turnover conflicts.

**Examples from escalated conversations** shows real cases behind the issue, each with a link straight to the conversation so you can check the summary against what actually happened.

The **Conversations** tab lists every conversation grouped under the issue.

<Note>
  Read a couple of the examples before acting on the recommendations. The grouping and the write up are AI generated, so they are a strong starting point rather than a verdict, and the underlying conversations are the ground truth.
</Note>

## Categories

Issues are categorized by the kind of gap behind them, for example **Missing Info** where the agent had no information covering the question. The category tells you where the fix belongs before you read anything else. See [What Goes Where](/build-your-agent/what-goes-where/overview).

## Acting on an issue

Three actions sit at the bottom of an open issue:

| Action | Shortcut | When to use it |
| - | - | - |
| **Dismiss** | `X` | The pattern is real but you do not want the agent handling it. Some things should always reach a person |
| **Mark as Resolved** | `E` | You have made the change, in your knowledge base, a skill, or a guardrail |
| **Fix with Operator** | `R` | Hand the issue to the Operator to build the fix with you. |

**Fix with Operator** is the fastest route on most issues, since the Operator already has the write up and the example conversations as context. See [Operator](/build-your-agent/operator).

<Warning>
  Marking an issue resolved records your decision. It does not verify that the change works. Test the fix before you close the issue, or you will resolve the same pattern again next month. See [Testing Agents](/build-your-agent/testing-agents).
</Warning>

## A working rhythm

A practical loop for the first weeks of an account:

1. Sort **Pending** issues by conversation count and take the largest first
2. Read the Overview and two or three example conversations
3. Decide whether the fix is knowledge, a skill, or a guardrail
4. Make the change, or hand it to the Operator
5. Test it, then mark the issue resolved

<Tip>
  Working issues moves automation in a way that answering escalations one at a time never will. Answering a conversation helps one guest. Closing an issue stops the next twenty-eight conversations from escalating at all.
</Tip>

## Frequently asked questions

<AccordionGroup>
  <Accordion title="How is this different from the escalations in my inbox?">
    The inbox shows you individual conversations that need a person right now. The Escalations tab groups those same conversations into the recurring gaps behind them, so you can fix the cause rather than the instance. See [Escalations and AI tags](/inbox/agent/escalations).
  </Accordion>

  <Accordion title="Do I have to answer the underlying conversations separately?">
    Yes. Resolving an issue is about closing the gap for the future. The guests in those conversations still need replies.
  </Accordion>

  <Accordion title="What is the difference between dismissing and resolving?">
    Dismiss means you looked and decided no change is needed, usually because the situation should always reach a human. Resolve means you made a change. Keeping them distinct is what stops dismissed issues from looking like completed work.
  </Accordion>

  <Accordion title="Can a dismissed issue come back?">
    New conversations matching the same pattern keep arriving whether or not you dismissed the issue. If a dismissed pattern is generating volume you did not expect, it is worth revisiting rather than dismissing again.
  </Accordion>
</AccordionGroup>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.