Get startedGet started for free

Monitoring and analytics

1. Monitoring and analytics

You've published the agent and shared it with a pilot group. Every conversation they have is recorded, and reading them shows which questions the agent handles and which ones it drops.

2. Meet the Monitor tab

Let's ground this in our HR help agent a week after it was published, fielding time-off requests, portal sign-in problems, and the odd question nobody planned for. Open the Monitor tab across the top of the agent, behind the overflow button if your window is narrow. You get a date range, a few summary cards, and charts underneath.

3. Conversation sessions

Start with the Overview card and its first number, conversation sessions: how many conversations the agent took on. Next to it sits engagement, the share that went somewhere. Together they show whether anyone's using it and whether that's climbing. On the Billing card, Copilot credits should read zero for the standard agents you've built. These cards only fill in once real people use the agent, because Test panel chats don't count, so a new agent shows zeros.

4. How conversations ended

Next, scroll to Effectiveness and the conversation outcomes chart, which splits every chat four ways: resolved, escalated, abandoned, and unengaged. Resolved means the agent handled it, and abandoned means the person left first. A few abandoned chats are normal, but if that share keeps rising, look into it. See sessions opens the conversations behind the chart, so you can read what people said.

5. From numbers to fixes

The dashboard doesn't give your agent a grade. Each of these numbers points to something you can fix, like improving a description or adding a new topic.

6. Turning a spike into a fix

Say more of our HR agent's conversations ended abandoned this week. In See sessions, you find people typing "I can't log in from home" and "my portal password expired while I was away." Those belong with your HR portal password topic, but they never reached it. The agent picks a topic by reading its description, so add people's real wording to that topic's example queries.

7. When there's no topic at all

Sometimes people ask about something your agent has no topic for, like a wave of questions about a new parental-leave policy. Editing a description won't cover that. It's your cue to build a new topic, and the sessions you read show which one.

8. Close the loop and watch

Then you republish with the new examples or topic and come back a week later. Done right, fewer of those conversations end abandoned, because the agent now has a topic for them. Keep checking and fixing for as long as the agent is live.

9. Let's practice!

Now it's your turn to open your own agent's Monitor tab and find where sessions, credits, and conversation outcomes live.

Create Your Free Account

or

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.