How is my AI Agent doing - Agent Analytics

Last updated: July 13, 2026

To see how well your agent is handling issues go to the Analytics tab under an agent's settings page.

You can also use this dashboard to monitor your AI Agent usage against your plan limits by tracking how many issues have been assigned to the agent.

The following metrics are plotted by default for all agents:

To understand how specific metrics are calculated, hover over any chart to see a tooltip explanation, or click the three‑dot menu on a chart and select Edit to view the exact formula used for that metric.

  • Assign rate

  • Assist rate

  • Resolution rate

  • Escalation rate

  • Issues Assigned vs Not Assigned

  • Breakdown of Issues Assisted

  • Issues Escalated vs Resolved

  • Breakdown of Agent Escalations

  • Satisfaction Rate

  • CSAT Score Distribution

Use these metrics to set internal KPIs on AI Agent performance and work to achieve them by improving your knowledge base content, writing more runbooks, and tweaking your AI Agent configuration.

Note: Your target metrics will vary based on your use case. If your agent focuses on deflecting commonly asked questions, achieving high resolution rates (approaching 100%) is realistic, though outliers will always exist. However, if your agent primarily gathers information before routing to human agents, high assignment and resolution rates may not be meaningful KPIs for your workflow.

Assign Rate

This is the percentage of issues the AI agent has ever been assigned to. An issue that was reassigned will still be included in the number of issues assigned to the agent.

# Assigned Issues / # Total Issues

Assist Rate

This is the percentage of assigned issues the AI agent has assisted with. An agent assist is when it responds to an issue or runs a runbook. Any response visible to the customer is included. This may include issues where the agent only ran a runbook, or where it escalated after sending a response.

# Assisted Issues / # Assigned Issues

Note: For metric calculations, a "response" is the substantive AI‑generated response, not escalation messages outlined in escalation rules. The "I'm sorry, I don't have the information" message is not counted as an agent response.

Resolution Rate

  1. Marked as closed without human assistance (no human messages involved)

  2. Closed after the customer didn't respond for 24 hours following the AI agent's last message

# Resolved Issues / # Assisted Issues

Note: When an issue goes 24 hours without customer response, it is considered resolved for analytics purposes only. The issue status is not automatically changed to "closed."

Escalation Rate

# Escalated Issues / # Assisted Issues

Note: This metric only includes issues where the agent assisted (responded or ran a runbook) before escalating. See "Breakdown of Agent Escalations" below for a different view that includes all assigned issues that were escalated, even without agent assistance.

Issues Assigned vs Not Assigned

This shows the weekly count of issues that were ever assigned to this AI agent against issues that were never assigned to any AI agent. Issues that were only ever assigned to another AI agent are not included.

Breakdown of Issues Assisted

This shows the weekly count of issues that were assisted with by the AI agent broken down by how the AI agent assisted. By running a runbook, responding, both, or no assist. Only issues assigned to the AI agent are included in this view.

Issues Escalated vs Resolved

This shows the weekly count of issues that were escalated by the AI agent against the issues that were resolved by the AI agent. Issues can only be resolved OR escalated but not both. Escalated issues can be either reassigned or unassigned. Only issues that were assisted by the AI agent are included in this view.

Breakdown of Agent Escalations

The types of escalation are as follows:

  • Customer Escalated - The customer explicitly requested a human agent

  • Interrupted - A human agent specifically reassigned the issue to someone else or commented on an issue. Reassignment triggers are included as well.

  • Agent Error - The AI Agent encountered an error

  • Unknown - Issue was escalated prior to detailed tracking

Satisfaction Rate

This shows the percentage of CSAT scores that are a 4 or 5 attributed to the AI Agent

# CSAT Scores > 4 / # CSAT Responses

CSAT Score Distribution

This shows the distribution of CSAT scores attributed to the AI Agent.