Conversations
A conversation is a thread of question-answer interactions between a user and the AI assistant. Kapa persists all conversations across your deployments for you to review and analyze.
The Kapa platform offers two different modes of exploring conversations:
- Review Mode: shows you the most relevant conversations to analyze.
- All Conversations: lets you search and filter all of your conversations.
Auto Labels
Kapa automatically assigns labels to conversations by analyzing the intent of the questions in the thread.
The available Auto Label categories are:
| Category | Description |
|---|---|
| Uncertain | Conversations where Kapa expresses low confidence or cannot find sufficient information in your knowledge sources to provide a complete answer. |
| Troubleshooting | Conversations where users describe errors, unexpected behavior, or implementation issues they're experiencing with your product. |
| Unsupported Feature | Conversations where users inquire about functionality that isn't currently available in your product offering. |
| Competitor | Conversations that mention or compare your product with competing solutions in the market. |
| Off Topic | Conversations unrelated to your product, documentation, or support scope. |
| Deflection Success | Conversations where a user initially intended to submit a support ticket, but the Support Form Deflector provided a satisfactory answer, eliminating the need for human intervention. |
| Deflection Failure | Conversations where a user proceeded to submit a support ticket after receiving an answer from the Support Form Deflector, indicating the automated response didn't resolve their issue. |
Review Mode
Review Mode is the default tab when you open the Conversations screen in the Kapa platform. It shows you only the most relevant conversations your Kapa instances have had in the past week. This helps reduce the noise and lets you focus on the conversations that matter the most. It also tracks your review progress so you can pick up where you have left off.
For guidelines on how to use the review mode, see Conversation review best practices
How does Review Mode filter conversations?
By default, Review Mode only shows conversations whose Auto Labels include at least one of the following categories:
- Competitor
- Troubleshooting
- Unsupported Feature
Conversations tagged with the Off Topic Auto Label are automatically excluded from the Review Mode. You can configure which categories to display in the Review Mode using the Filter by auto-labels drop-down menu.
How does it track my progress?
At the top of the page you can see how many conversations are left to review. When you're done reviewing a conversation, it's removed from the Review Mode. This lets you pick up again where you have left off.
All Conversations
The All Conversations tab contains all raw conversations across all deployment integrations for your project. By default, it lists the conversations that occurred in the past 30 days in reverse chronological order.
Filters
The conversation list can be filtered using several criteria to help you find specific conversations:
- Text contents: Search for specific words or phrases in questions and answers
- Integration: Filter by specific Kapa deployment (e.g., Website Widget, Slack bot)
- Auto Labels: Filter by automatically assigned conversation categories
- Status tags: Find conversations with specific status indicators
- Feedback ratings: Filter by user-provided feedback scores
- Date: Limit results to a specific time period
Each Kapa deployment must use a unique integration ID for proper analytics tracking. If multiple deployments (such as your documentation website and marketing website) share the same integration ID for their Website Widgets, you won't be able to distinguish between them in the analytics.
Export conversations
You can export conversation data to a CSV file for further analysis or reporting purposes.
To export conversation data to a CSV file:
- Set the desired date range using the date picker
- Apply any filters to narrow down the data set
- Click the Export to CSV button next to the date picker
The exported CSV file contains the initial question and answer (the complete thread is not exported), timestamp, thread ID, URL, and other metadata for all conversations matching the selected filters.