Topic discovery

Topic discovery uses AI to analyze messages across your inboxes and automatically identify the most common topics your team handles. This helps you understand what drives your support volume so you can prioritize automation.

How it works

Topic discovery samples messages from the inboxes you select and uses AI to detect recurring themes. It groups these into a hierarchy of topics and subtopics, ranked by volume. You can then review, refine, and promote the discovered topics into a classification model.

Starting a discovery

Step 1: Open topic discovery

Navigate to "Analytics" and click the "Discover topics" button. If this is your first time, you will see a banner inviting you to get started.

Step 2: Select your inboxes

In the dialog that opens, select one or more inboxes to analyze. You can choose specific inboxes or include all of them.

Step 3: Add instructions (optional)

Optionally, add custom instructions to guide the discovery. For example, you could write "Focus on billing and returns issues" to narrow the scope. If you leave this blank, Tekst generates a default instruction based on your organization and selected inboxes.

Step 4: Start the discovery

Click "Discover topics" to begin. The process typically takes a couple of minutes. You can navigate away and come back later, as the banner on the Analytics page will update with the progress and notify you when results are ready.

Reviewing results

Once the discovery completes, click "View topics" on the banner to see the results.

The results page shows a table of discovered topics and subtopics, sorted by message volume. For each topic you can see:

• The topic name
• The number of messages and its share of total volume
• Subtopics grouped under their parent topic

Click on any topic row to open a side panel with sample messages that were tagged with that topic.

Refining results

If the initial results need adjustment, click "Refine discovery" at the bottom of the results page. Enter additional instructions describing what to change, then submit. A new discovery run starts with your refined guidance.

Removing unwanted topics

To remove a topic you do not want, click the delete button on its row. This marks the topic as excluded and it will not be included when you promote the results.

Promoting to a classification model

When you are satisfied with the discovered topics, click "Use for classification" to create a classification model from the results. This turns the discovered topics into a working model that Tekst uses to automatically classify incoming messages.

After promotion, you are taken to the new model page where you can further configure and manage it.

Things to know

• Topic discovery analyzes a sample of recent messages (up to 6 months) from the selected inboxes.
• You cannot start a new discovery while any of your inboxes are still backfilling. Wait until at least one inbox has finished backfilling.
• Each discovery run is saved, so you can revisit previous results from the Analytics page.
• Refining a discovery creates a new run. Your previous results are preserved.

For a broader look at all analytics features, see the Analytics overview.

If you have any questions or need assistance, please contact your customer success manager.