Insights tells you how your agents perform: how conversations end, how well the agent handles them, how customers react, and what it all costs.
How Insights works
Insights turns each conversation into metrics in four steps:
Conversations happen. Every chat, API call, and phone call with your agents starts the flow.
The platform records them. It stores messages, tool calls, handoffs, errors, timings, tokens, and cost, plus any custom tags you add.
Pipelines analyze them. Built-in and custom pipelines read each finished conversation and score it with an LLM.
Dashboards show the results. The prebuilt pages, custom dashboards, and Analytics Explorer read from one analytics store.
Three more sources feed Insights: your settings (costs, thresholds, and billing policy), customer surveys, and telephony call reports. To learn how each metric is calculated and how to compare numbers across pages, see How Insights metrics work.
Before you begin
Enable the built-in pipelines in Insights → Pipelines. The quality, sentiment, and outcome metrics and the other dashboards depend on them.
Platform counts update in real time, and LLM-scored metrics appear after the conversation ends. A dash means no data yet, not zero.
Never run the sample data generator in a project with real traffic. Its sample conversations appear in every dashboard and can’t be removed there.
Insights dashboards
In the product, the Insights sidebar groups the dashboards as follows:
Use when prebuilt pages don’t answer your question. Shows near real-time event, model, token, and cost data, with session and trace drill-down and custom queries.
Use when no prebuilt page shows the metric you need. Build charts from SQL or visual queries over the project’s analytics data, and share them with the project.
Use to see which evaluations run, or to add your own LLM analysis, such as whether the customer’s goal was met. See Custom pipelines. Fields a custom pipeline stores are available to custom dashboards.