> ## 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.

# Set Up Topic Analysis

> Create topics, enable the Topic Analyzer, and start categorizing conversations automatically

Topic analysis automatically tags conversations by subject so you can track what users ask about, spot trending topics, and see CX Score and escalation rates per topic. Here's how to set it up.

## Prerequisites

* Admin access
* At least one agent handling conversations

<Steps>
  <Step title="Create Your Topics">
    Navigate to **Settings** > **Conversation Labels** and select the **Topics** tab (or **Tags** tab — the first tab shown).

    1. Click **Add Topic**
    2. Enter a **name** (e.g., "billing", "technical-support", "account-access")
    3. Add a **description** explaining what the topic covers — the AI reads this when deciding whether to assign the topic
    4. Optionally add **keywords** — words or phrases that trigger automatic keyword matching (e.g., "invoice", "payment", "subscription" for a billing topic)
    5. Click **Save**

    Repeat for each topic. Start with 10-20 topics covering your most common conversation subjects.

    #### Use Hierarchy for Structure

    If you have broad categories with subtopics, set a **parent topic**:

    * **Support** (parent)
      * **Billing Support** (child)
      * **Technical Support** (child)

    Child topics inherit their parent's context in the analytics treemap, and you can drill down from parent to children.

    <Tip>
      Don't create too many topics upfront. Start with the obvious categories and add more after reviewing the first week of results.
    </Tip>

    See [Conversation Labels](/features/conversation-labels) for all fields and options.
  </Step>

  <Step title="Enable the Topic Analyzer">
    Navigate to **Settings** > **Topic Analyzer** (under Thread Evaluation in the sidebar).

    1. Toggle the Topic Analyzer **on**
    2. Switch to the **Custom Instructions** tab and add guidance for how the AI should categorize conversations. For example:

    ```markdown theme={null}
    Assign topics based on the user's primary question, not side mentions.
    If a conversation covers multiple subjects, assign the most relevant 2-3 topics.
    Use parent topics only when no specific child topic fits.
    ```

    3. Switch to the **Settings** tab and configure:
       * **Maximum Topics per Conversation** — how many topics can be assigned (default: 5, start with 3-5)
       * **Always assign at least one topic** — toggle on if you want every conversation tagged

    4. Click **Save**

    See [Topic Analyzer](/features/thread-tagger) for all configuration options.
  </Step>

  <Step title="Run Analysis on Existing Conversations">
    The Topic Analyzer processes new conversations automatically. To analyze existing ones:

    1. Go to **Chat History** in the sidebar
    2. Open a conversation
    3. In the sidebar, find the **Topics** section and click **Run Topic Analysis**
    4. Check the assigned topics — do they match what the conversation is about?

    Run this on 5-10 conversations to verify your topics and instructions produce good results. If topics are being assigned incorrectly, adjust your custom instructions or topic descriptions.
  </Step>

  <Step title="Review Results in Topics Analytics">
    After a day or two of automatic analysis, check the results:

    1. Navigate to **Dashboard** and select the **Topics** tab
    2. Review the **treemap** — which topics have the most conversations?
    3. Check **Topics Over Time** — are any topics trending up or down?
    4. Look at the **Topic / Sub-topic Breakdown** for CX Score and escalation rate per topic — topics with low CX Scores or high escalation rates may indicate knowledge gaps

    See [Topics Analytics](/features/topics-analytics) for all available charts and metrics.
  </Step>

  <Step title="Refine Over Time">
    After the first week:

    * **Merge low-volume topics** that cover similar subjects
    * **Split high-volume topics** that are too broad (e.g., split "technical" into "technical-bugs" and "technical-how-to")
    * **Update keywords** based on terms you see in actual conversations
    * **Adjust custom instructions** if the AI misclassifies certain conversation types

    Review your topic structure quarterly to keep it useful.
  </Step>
</Steps>

## Related Features

* [Conversation Labels](/features/conversation-labels) - Create and manage topics
* [Topic Analyzer](/features/thread-tagger) - Configure automatic topic assignment
* [Topics Analytics](/features/topics-analytics) - View topic distribution and trends
* [Conversations](/features/conversations-list) - Filter conversations by topic