Testing and Monitoring
Once an agent is published, its page in the Admin app becomes the place to check how it behaves. You can talk to it in the playground before a change reaches customers, read back every conversation it has had, and follow its trends over time. This section covers those three jobs.
The Agent Page
Open Agents and select a published agent. Its page has four tabs:
| Tab | What it is for |
|---|---|
| Metrics | Totals, trends and scores for the last 7, 30 or 90 days. See Metrics. |
| Sessions | Every conversation the agent has had, with its transcript, score and summary. See Sessions. |
| Playground | A test conversation with the agent, by voice or text. See Playground. |
| Quick Config | Common settings you can change without opening the full editor. See Quick Config. |
The same tabs are offered as shortcuts in an agent's actions menu on the Agents page.
The agent page only opens for an agent that has been published at least once. Selecting an agent that has never been published opens it in the builder or the editor instead, where you test it in the builder's Test It Out or Try It Out step, or the editor's Test panel. See Testing Before You Publish. See Building with IQ and The Agent Editor.
Before You Start
- Role: the Admin or Organisation Admin role. Both can test agents and read their sessions and metrics. See Roles and Permissions.
- A published agent. See Publishing and Versions.
How Testing and Monitoring Fit Together
Every conversation an agent has is saved as a session. A session is either Live, meaning a real caller or system reached the agent, or Evaluate, meaning someone tested it in the playground. See Live and evaluate sessions in the key terms.
A typical loop looks like this:
- Change the agent and save the change as a draft.
- Test the draft in the Playground tab, and compare it with the live version.
- Publish the draft when you are happy with it.
- Read new sessions in the Sessions tab as callers and systems use the agent.
- Watch the Metrics tab for changes in volume, scores or tool failures, and go back to step 1 when something needs fixing.
Sessions and metrics belong to the environment the agent is in. An agent in Sandbox and an agent in Production have separate histories, even when they are configured the same way.
In This Section
- Playground. Test an agent by voice or text, and compare a draft with the live version
- Sessions. Filter sessions, read scores and summaries, and open the full transcript
- Metrics. What each total and chart means, and how to read them