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

# Diagnose Quality Issues

> Find and fix the root cause when your CX Score drops or escalation rates rise

Your CX Score has dropped, escalation rates are climbing, or negative feedback is increasing. This guide takes you from symptom to root cause to fix.

## Prerequisites

* Admin or Account Manager access
* [Thread Evaluator](/features/thread-evaluator) enabled with at least a few days of scored conversations

<Steps>
  <Step title="Identify the Problem Area">
    Open the **Dashboard** and check the **Overview** tab. Look at the four key metrics:

    * **Escalation Rate rising** — the agent can't handle certain questions
    * **Positive Evaluation Rate dropping** — response quality has declined
    * **Thread count spiking** — a new topic may be overwhelming the agent

    Next, switch to the **Evaluation** tab and check the **CX Score dimension averages**. The lowest dimension tells you where to focus:

    | Low Dimension      | Likely Cause                                                               |
    | ------------------ | -------------------------------------------------------------------------- |
    | Resolution Quality | Knowledge gaps — the agent doesn't have the information to answer          |
    | Response Accuracy  | Conflicting content in the knowledge base, or hallucinated answers         |
    | Tone & Empathy     | Agent instructions don't specify tone clearly, or tone guidelines conflict |
    | Efficiency         | Agent asks too many clarifying questions or repeats itself                 |
    | Customer Effort    | Users have to rephrase questions or navigate complex flows                 |
  </Step>

  <Step title="Find the Conversations">
    Go to **Chat History** in the sidebar and filter to isolate the problematic conversations:

    1. Set **CX Score Range** to **Poor** (0-49) or **Needs Improvement** (50-79)
    2. If a specific dimension is low, use the **CX Dimension Filter** — for example, set Resolution Quality \< 50
    3. Set the date range to match the period where you noticed the decline
  </Step>

  <Step title="Read and Identify Patterns">
    Open 5-10 of the filtered conversations. For each one, note:

    * **What the user asked** — is there a common topic?
    * **What the agent answered** — was the answer wrong, incomplete, or off-topic?
    * **Where it went wrong** — did the agent retrieve the wrong document? Did it hallucinate? Did it fail to use a tool?

    Look for repeating patterns. A single bad conversation is an outlier; the same failure across multiple conversations is a systemic issue.
  </Step>

  <Step title="Fix the Root Cause">
    Based on the patterns you found:

    | Pattern                                           | Fix                                                                              |
    | ------------------------------------------------- | -------------------------------------------------------------------------------- |
    | Agent gives wrong information on a specific topic | Update or correct the relevant knowledge base documents                          |
    | Agent can't find relevant content                 | Add new documentation covering the missing topic                                 |
    | Agent retrieves the wrong document                | Improve document titles, tags, or content to help retrieval                      |
    | Agent tone is inconsistent                        | Add explicit tone guidelines to the agent's instructions                         |
    | Agent asks too many questions before answering    | Adjust instructions to attempt an answer before asking clarifying questions      |
    | Agent doesn't use available tools                 | Check tool descriptions and update them so the agent knows when to use each tool |
  </Step>

  <Step title="Verify the Fix">
    After making changes:

    1. Run [Simulation Testing](/features/simulation-testing) with test cases that match the problem scenarios
    2. Monitor the **CX Score trend** on the Evaluation tab over the next few days
    3. Re-filter Chat History by the same criteria to confirm new conversations score higher
  </Step>
</Steps>

## Related Features

* [CX Score](/features/cx-score) - How scores are calculated and what each dimension measures
* [Thread Evaluator](/features/thread-evaluator) - Configure evaluation settings and thresholds
* [Evaluation Dashboard](/features/evaluation-dashboard) - View CX Score charts and dimension averages
* [Conversations](/features/conversations-list) - Filter conversations by CX Score and dimensions