# Autonomous AI performance reports

Review autonomous AI outcomes, handoffs, customer satisfaction and deflection over time.

*Release 3.3 · admin, supervisor*

**Reports › AI › Autonomous AI Performance** · **Service Overview › Performance › Ai Autonomous** · **Service Overview › Performance › Email Autopilot** · **Service Overview › Performance › Voice Autopilot** · **Settings › AI › Features**

## What it is

**Autonomous AI Performance** shows how autonomous AI conversations ended during a selected period. It includes resolution outcomes, escalation, spam, **CSAT** and **Deflection**. Separate views in **Service Overview** help you review AI activity for chat, email and voice.

## Who uses it and why

- Supervisors monitor AI handling and identify conversations that need human follow-up.
- Administrators compare outcomes over time and check the effect of an Autopilot configuration.

## Where to find it

Open **Reports › AI › Autonomous AI Performance** with the **Reports** role.

In **Service Overview › Performance**, the available AI views are **Ai Autonomous**, **Email Autopilot** and **Voice Autopilot**. Their availability depends on your website and account access.

## How it works

1. Select your website, **Date From** and **Date To**.
2. Choose the grouping and any displayed scope filters. The historical table supports location groups, subsites, tags, ending sentiment, resolution status, interaction type and **Feature**.
3. Select **Apply Filters** and compare the date rows with the total.
4. Use a channel view in **Service Overview** for period comparisons. Where a metric is linked, open the matching conversations to investigate the result.

Read the main measures as follows:

- **Resolved**, **Unresolved** and **Undetermined** describe recorded AI resolution outcomes. **Spam** identifies conversations classified as spam.
- **Escalated** counts handoffs to a person, including relevant linked-ticket handoffs. It can overlap a resolution outcome.
- **CSAT** is the percentage of qualifying recorded survey results with a score of at least 4. When several answers are recorded for a conversation, their average determines whether it meets that threshold.
- **Deflection** divides conversations resolved by AI without human handoff by all resolved conversations in the selection.

## Configure it

Set your period and filters in the report. **Feature** lets you select Autopilot configurations and includes their historical versions, so earlier activity remains available when a feature changes.

Manage the underlying AI behaviour through **Settings › AI › Features** with the **Settings** or **Developer** role. Reporting reads the recorded outcomes; changing a report filter does not change how Autopilot handles visitors.

### Permissions

| Permission | Lets you |
|---|---|
| Reports | Open Autonomous AI Performance in the reporting workspace. |

## Limits and good to know

- Hourly grouping is available for ranges up to seven days. Longer ranges automatically use daily grouping.
- Deflection measures resolved conversations. It is not the percentage of all incoming demand avoided.
- Do not add escalation to the resolution counts to reconstruct a total.
- A zero CSAT result can reflect a lack of qualifying survey answers. Check survey participation before interpreting it.
- Align dates, filters and data availability when comparing the historical table with a channel dashboard.

## Depends on and works with

### This feature uses

- **Requires** [Report selection and filters](REP-001.md): The historical table uses the reporting workspace and its filters.
- **Configured in** [Roles and permissions](ADM-008.md): Roles control reporting and service-overview access.
- **Configured in** [AI feature settings and change history](AI-001.md): Autopilot feature selection includes the historical versions that handled conversations.
- **Configured in** [AI resolution and first-contact metrics](AI-030.md): AI metrics instructions and outcome rules determine the resolution classifications used in autonomous performance reports.
- **Configured in** [Visitor survey questions and scores](VX-013.md): Survey question types and scores determine the feedback available for autonomous CSAT calculations.
- **Works better with** [Autopilot for live chat](AI-003.md): Live conversation Autopilot supplies autonomous handling activity.
- **Works better with** [Autopilot handover and error handling](AI-004.md): Handoffs and recovery affect escalation outcomes.
- **Works better with** [Email Autopilot for tickets](AI-012.md): Email Autopilot has a separate channel performance view.
- **Works better with** [Voice Autopilot call handling](AI-013.md): Voice Autopilot has a separate channel performance view.
- **Works better with** [Sentiment and emotion](CHAT-022.md): Ending sentiment can narrow the historical table.
- **Works better with** [Service Overview performance comparisons](REP-015.md): Service Overview adds period comparisons for the available Autopilot channels.

### Used by

- **Enhances** [Service Overview performance comparisons](REP-015.md): Separate AI views analyse autonomous handling where enabled.
- **Enhances** [Service Overview dashboard](REP-016.md): Enabled AI performance views add autonomous and Autopilot measures.
- **Related** [AI resolution and first-contact metrics](AI-030.md): Autonomous AI Performance reports the recorded resolution outcomes.

## FAQ and troubleshooting

### Why has hourly grouping changed to daily?

The selected range exceeds seven days. Shorten it to use hourly rows.

### Why is deflection higher than the resolved percentage?

They use different denominators. The resolved percentage uses total chats; deflection uses resolved chats.

### Why do outcome columns not add up?

Escalation can overlap a recorded resolution outcome. Treat it as a separate handling measure.

### Why is an older Autopilot version included?

The feature selection groups historical versions so that conversations handled by earlier versions remain part of the selected feature's history.