Project Management

Analytics Suite

What is the Analytics Suite for?

The Analytics Suite is where you see how your agents are actually performing: not just whether conversations are happening, but whether they’re going well. It combines usage statistics with evaluation results, so you can judge both how much your agents are used and the outcomes they generate.

You can view data across the whole project or drill down to a specific agent, agent version or user.

You find it under Performance → Analytics Suite in the left sidebar.

Filters

The filter bar at the top of the Analytics Suite controls the data shown in every card below it. By default it shows the last 30 days. Filters are reset when you leave the page.

From / To

Set a custom start and end point for the data, including the time of day. Pick the day and time in the picker, or type the value straight into the field in the Aug 18, 2026 @ 13:58 format. Times use your local timezone.

Time Period

Choose a preset range, from Last 15 minutes up to All time, instead of setting the dates by hand. Adjusting From or To switches this to Custom.

Select Agent

View data for all agents, or narrow it down to specific agents and versions. See Agent and Version Selection.

Filter by User

View data across all users, or filter by a specific one. Use the search field to find a user.

Export data

Download the currently filtered data as Excel (.xlsx), CSV (.zip), or JSON (.json). See Exporting Data.

Agent and Version Selection

Select Agent opens the Agents version selection panel: agents on the left, the versions of the selected agent on the right. The number of selected versions is shown in the top-right corner of the panel, and each agent in the list shows how many of its versions are selected.

By default All Agents is selected, which includes every version of every agent.

To analyse specific versions, first click All Agents to clear the selection. Then choose an agent on the left and toggle the versions you want on the right. Otherwise, everything stays selected.

Use All drafts to select every draft version of the chosen agent at once, and Clear drafts to remove them again. Use the Search agents field to find an agent in a long list.

The button itself reflects the current selection: All Agents, a single agent version, or the number of selected agent versions.

All Agents includes draft versions, so the conversations you have while testing an agent in Studio are counted too. To see production only, choose the agent and click Clear drafts.

What Is Counted

  • Conversations from test suites are never counted.

  • Conversations with only one turn are not counted.

  • Lifetime figures ignore From / To, but follow the agent and user filters. On large projects they can show a dash for a moment while they are calculated.

A green or red chip compares the selected period with the period of the same length right before it. Hover over the chip to see that period. +0% can also mean that the previous period had no data.

Usage Metrics

Conversations

Conversations in the period, split into Short (1-2 turns), Medium (3-5 turns) and Long (6+ turns), plus Lifetime conversations.

Unique Users

Distinct users in the period, a chart of unique users over time and Lifetime users. The chart interval adapts to the range, from 5 minutes up to 30 days.

Total conversation length

Combined duration of all conversations.

Average conversation length

Average duration of one conversation.

Average turns per session

Average number of turns per conversation.

Conversations per Agent

Conversations per agent in the period, with all versions of an agent counted together.

Conversations per User

How many users had one, two, three or more conversations in the period. Shows how deeply users adopt your agents, not just how many are active.

Conversation lengths in analytics are not the minutes you are billed for. See Billing and Best Practices.

One card for each evaluation that has a featured output and Admin Facing on. Each card shows the evaluation name, the agents it is attached to, Users evaluated and Attempts, and:

  • Pass / Fail: the percentage of users who passed, and on which attempt (passed 1st try, passed 2nd try, passed 3rd+ try, not passed).

  • Number: avg best, the average of each user’s best score, and the average per attempt (1st attempt, 2nd attempt, 3rd+ attempts).

The figures cover the users active in the selected period, across all of their conversations, including those before the period.

Evaluations Metrics

One card for each output data element of the evaluations with Post-Conversation Analysis and Admin Facing on. Use Filter by type and Select evaluation to narrow the cards. With both filters on "All", you can drag cards to change their order for this visit.

Pass/Fail

Percentage of users with at least one passing conversation, counts per label, the number of users and the number of sessions evaluated.

Number

The average of each user’s best value, on the element’s scale (Min / Max).

Text

The three most recent answers. Export the data to see all of them.

List

How often each value was picked, as a percentage.

Pass rates and averages are per user, not per conversation. A user who passed once counts as passed.

If a card you expect is missing, see Evaluations troubleshooting.

Conversations

A list of the conversations in the selected period. Select one to see its summary, Date & time, User, Agent (with version), Turns and the transcript.

This section is hidden when the account setting Display full conversation data is off.

Interactive Content

A table of your interactive content items (the Type column shows the item’s reference) and how often each was shown (Total Views). Select Show no views entries to also list items that were never shown.

Exporting Data

Export data downloads the data that matches the current filters:

  • Summary: the filters and the usage metrics.

  • Sessions: one row per conversation (user, agent and version, start and end in UTC, turns, duration). The transcript is included only when Display full conversation data is on.

  • Evaluations: one row per result of an Admin Facing evaluation, with the value, reasoning and evidence.

Excel (.xlsx) puts these on three sheets, CSV (.zip) contains three CSV files, and JSON (.json) contains all of it in one file. One export can contain at most 100,000 conversations; for more, narrow From / To.

Troubleshooting

Evaluations Metrics or Featured Evaluations is empty?

Most often Admin Facing is off in the evaluation, or the evaluation isn’t attached to the published agent. See Evaluations troubleshooting.

Fewer conversations than expected?

Single-turn conversations and test-suite runs aren’t counted. Also check Select Agent and Filter by User.

The numbers include my own tests?

Open Select Agent, choose the agent and click Clear drafts.

The Conversations section is missing?

The account setting Display full conversation data is off.

The export fails?

The range contains more than 100,000 conversations. Narrow From / To.

The chip shows +0%?

The previous period may have had no data.