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

# Evaluate conversations and extract structured insights

> Configure success criteria to automatically score conversation outcomes, and extract structured data fields from transcripts to track what users ask about.

Conversation analysis lets you define what a successful conversation looks like and automatically score every call against that definition. You can also configure data collection to pull structured fields — like user questions or intent — directly from transcripts. Both features are available under the **Analysis** tab in your agent's settings.

## Set up success evaluation

Success evaluation scores each conversation as `success`, `failure`, or `unknown` based on criteria you define. The agent's LLM applies the criteria to the transcript after the call ends, so there is no runtime cost to latency.

<Steps>
  <Step title="Open the Analysis tab">
    Go to [xuna.ai/app/agents](https://xuna.ai/app/agents), select your agent, and click **Analysis**.
  </Step>

  <Step title="Add an evaluation criterion">
    Under **Success evaluation**, click **Add criterion**. Give it a short machine-readable name — for example, `solved_user_inquiry`.
  </Step>

  <Step title="Write the evaluation prompt">
    Describe what the evaluator should look for. Be specific about what constitutes success and what constitutes failure.

    ```text Example criterion theme={null}
    Name: solved_user_inquiry
    Prompt: The assistant was able to answer all of the queries or redirect them
            to a relevant support channel.
    Success criteria:
    - All user queries were answered satisfactorily.
    - The user was redirected to a relevant support channel if needed.
    ```
  </Step>

  <Step title="Save and test">
    Click **Save**. Run a test conversation (or use an existing one from Call History) and open the conversation detail to see the evaluation result and rationale.
  </Step>
</Steps>

### Outcome values

Each evaluation criterion produces one of three outcomes:

| Outcome   | Meaning                                                                                                                          |
| --------- | -------------------------------------------------------------------------------------------------------------------------------- |
| `success` | All success criteria were met                                                                                                    |
| `failure` | One or more criteria were not met                                                                                                |
| `unknown` | The evaluator could not determine an outcome — for example, the conversation was too short or the user disconnected unexpectedly |

The rationale field explains which criteria passed or failed. Use it to debug prompt phrasing when you see unexpected `unknown` results.

<Note>
  You can define multiple evaluation criteria per agent. Each criterion is scored independently, so you can track success across different dimensions — for example, both `solved_user_inquiry` and `maintained_brand_tone`.
</Note>

## Set up data collection

Data collection extracts specific information from the transcript and stores it as structured fields on each conversation record. Use this to track trends, feed downstream systems, or filter conversations in Call History.

<Steps>
  <Step title="Add a data field">
    Under **Data collection**, click **Add field**.
  </Step>

  <Step title="Configure the field">
    Set the data type, identifier, and a description that tells the extractor what to look for.

    ```text Example field theme={null}
    Data type: string
    Identifier: user_question
    Description: Extract the user's questions and inquiries from the conversation.
    ```
  </Step>

  <Step title="Choose a data type">
    Select the type that matches the value you want to capture:

    | Type      | Use for                                                |
    | --------- | ------------------------------------------------------ |
    | `string`  | Free-text values like questions, complaints, or topics |
    | `number`  | Numeric values like order amounts or queue positions   |
    | `boolean` | Yes/no flags like whether the user agreed to terms     |
  </Step>

  <Step title="Save and verify">
    Save the field and run a conversation. Open the conversation detail view and check the **Collected data** section to confirm the field is populated correctly.
  </Step>
</Steps>

## Search conversation history

The Call History tab supports two search modes:

* **Keyword search** — matches exact words or phrases in transcripts.
* **Semantic search** — finds conversations by meaning, not exact wording. Searching "billing problem" surfaces conversations where users said "I was charged twice" or "my invoice is wrong".

Use semantic search when you want to understand how users describe a topic, not just how often they use a specific term. This is especially useful for identifying gaps in your knowledge base or discovering emerging support themes.

## Next steps

<CardGroup cols={2}>
  <Card title="Analytics dashboard" icon="chart-line" href="/monitor/analytics">
    View aggregate metrics and filter conversations by evaluation outcome.
  </Card>

  <Card title="Testing" icon="flask-vial" href="/monitor/testing">
    Run automated tests to validate your evaluation criteria before deploying.
  </Card>
</CardGroup>