ServiceNow AI Agent Analytics: metrics that lead to decisions

What the ServiceNow dashboard measures

ServiceNow's current AI Agent Analytics dashboard is part of AI Agent Studio. It tracks agent use and efficiency gain on an instance. Its overview visualizations include execution tasks by status, agent errors over time, tool type and execution mode, and an inferred CSAT score for recent interactions. The Studio overview also identifies usage, time-to-resolution efficiency gain, and number of executions as vital statistics.

The dashboard is not a universal scorecard. It measures the agents and workflows configured in that ServiceNow instance. Access requires the documented sn_aia.viewer or sn_aia.admin role, and what appears depends on the products and configuration available to the instance.

This is the separate AI Control Tower dashboard, illustrating broader AI-value reporting rather than the AI Agent Analytics screen. Source: ServiceNow AI Control Tower.

Turn a metric into an investigation

A useful review starts narrow. Suppose an incident-triage workflow's failed executions rise after a tool update. First filter to the affected workflow and time window. Then open a representative set of execution logs and records. Separate a genuine tool failure from a missing permission, an unclear input, or a task the agent should have handed to a person.

Use the metric to decide what to test next, not to declare victory. ServiceNow's run results can score selected executions and export a CSV report. Its evaluation score is based on execution logs, so pair it with a review of the actual records and the user-facing outcome. Inferred CSAT is likewise an estimate, not a substitute for survey feedback or a resolved case.

For a release decision, keep a small before-and-after record: baseline volume and error rate, the exact change, representative tests, the owner who approved it, and the external result to inspect after rollout. ServiceNow also documents monitoring for AI Guardian detections such as prompt-injection attempts and sensitive-topic detections. A lower error count is not enough if a new configuration weakens a needed guardrail.

Use eesel CLI to close a support analytics loop

ServiceNow analytics tells a ServiceNow owner where to investigate. It does not manage an eesel teammate. For an eesel helpdesk teammate, people can use the eesel CLI from a terminal; a scheduled script can consume its JSON; and Claude Code, Codex, or Cursor can turn the inspected output into a review note. They all operate the same eesel workspace and teammate available in the dashboard.

For example, imagine support reports that a teammate is handing off too many order-status questions. Before changing anything, an analyst scopes the inspection to the teammate under review and checks its current status, connected sources, instructions, and recent activity. That separates a knowledge gap from an action or instruction problem. It is a different job from reading ServiceNow's analytics dashboard, but it lets the team apply the same evidence-first habit to its support teammate.

With Node.js 18.17 or newer installed, inspect the selected teammate:

npx @eesel/cli login
npx @eesel/cli whoami
TARGET_EESEL_AGENT="paste-agent-id-or-name-here"
npx @eesel/cli status --agent "$TARGET_EESEL_AGENT"
npx @eesel/cli integrations --agent "$TARGET_EESEL_AGENT"
npx @eesel/cli instructions --agent "$TARGET_EESEL_AGENT"
npx @eesel/cli activity --agent "$TARGET_EESEL_AGENT"

The support owner reviews the report first. Only after approval should a person or coding agent make a supported, scoped change; use --dry-run where that write supports it. Before any chat test, review the named teammate's actions and permissions, then disable or contain actions outside the test. A queue for held actions does not stop an action that is configured to run without a hold.

Use a representative order-status question in a test chat, examine the reply and activity, then check the real helpdesk ticket or connected system after a controlled release. CLI chat is billed work; dashboard-management commands are free. The CLI makes repeatable inspection easier. The support owner still decides whether a change is justified and checks whether the customer-facing outcome was correct.

Build an operating rhythm, not a dashboard ritual

Review ServiceNow AI Agent Analytics on a cadence that matches the workflow's risk and traffic. Assign a metric owner, define an investigation threshold, and record which change was made. For high-impact workflows, include a security or governance owner when evaluating tool permissions and guardrails.

The same discipline applies to an eesel teammate: identify the exact teammate, inspect before editing, keep the test case representative, and verify the system where the work lands. That is how analytics becomes a controlled improvement loop instead of a collection of attractive charts.

Review support-teammate evidence with eesel

The eesel Activity page lists work with filters for approved, rejected, and pending items.

Try eesel to inspect a scoped support teammate before making a customer-facing change. Start with one question, one owner, and a result that someone can verify outside the model response.