The Performance tab gives you an overview of how your chatbot is handling conversations — resolution rates, activity trends, and the ability to review individual conversations in detail.
How to get there: Go to Setup → AI agent in the top menu → click your chatbot → Training → Performance in the sidebar.
Select a time period from the dropdown in the top right:
A donut chart showing the percentage of conversations resolved by the chatbot without human intervention. This is a key indicator of how well your chatbot handles questions on its own.
A summary card showing:
A line chart below shows the daily trend for chats and AI replies over the selected period. It counts complete days only, so it stops at yesterday and says so underneath ("Complete days through ..."). Today is still running, so a point for it would always look like a sudden drop. The averages above do count today.
Below the overview, a table lists all conversations for the selected period. Each row shows:
Filter the conversation list by outcome:
Click on any conversation to see its full details.
The full conversation is displayed as message bubbles showing:
A sidebar showing the search queries the chatbot made against your training data, along with match scores:
This helps you understand which training content was used and how well it matched the visitor's questions.
To see the exact sources used in a specific reply and edit them on the spot, use Review sources in the Inbox or the chatbot preview instead.
An AI-generated summary of the conversation along with the overall health score.
A table showing:
Details about the visitor — name, email, channel, and conversation date.
From the conversation detail view, you can create or update exact answers directly:
If the message was already a revised answer, it updates the existing one instead of creating a duplicate. This is the fastest way to improve your chatbot's responses based on real conversations.
The Performance tab and the Answer quality tab serve different purposes:
| Performance | Answer quality | |
|---|---|---|
| Focus | Conversation outcomes | Training content effectiveness |
| Key metrics | Resolution rate, activity trends | Knowledge coverage, average score |
| Shows | Individual conversations with full context | Knowledge gaps and top-performing topics |
| Use for | Monitoring day-to-day operations | Improving training data |
Use Performance to see how conversations are going. Use Answer quality to identify where your training content needs improvement.
The health score is an AI-generated rating (0-100) of how well the chatbot handled each conversation. It is used both for performance monitoring and as a check you can add to a notification rule (the AI couldn't fully answer check).
| Score | Signal |
|---|---|
| 0 – 39 | Chatbot failed — no useful answers, customer likely frustrated |
| 40 – 59 | Poor experience — partial answers at best, needs attention |
| 60 – 69 | Mediocre — answered but not well |
| 70 – 79 | Decent — normal conversations |
| 80+ | Good — handled well |
Conversations with a health score below 60 are flagged as "needs help" and can be routed via a notification rule using the AI couldn't fully answer check. This covers roughly 6% of conversations, the ones where escalation rates are 2-3x higher and customers are typically frustrated or left without answers.