Track incident value in Azure SRE Agent

You deployed an AI agent to handle incidents. Leadership wants to know: Is it reducing toil? Which incidents is it resolving on its own? Are we getting ROI?

Answering those questions today means querying telemetry, cross-referencing tickets, and guessing which response plans work. There's no single view showing what your agent did, how each response plan performed, or whether mitigation rates are improving.

Incident metrics dashboard

Navigate to Monitor > Incident metrics to see the dashboard.

Metric cards

Each card shows a count, its proportion of total incidents, and a sparkline with week-over-week change:

Metric	What it tells you
Incidents reviewed	Total incidents your agent investigated in the selected time range
Mitigated by agent	Incidents resolved autonomously — the core ROI metric
Assisted by agent	Incidents where the agent provided investigation data and a human completed resolution
Mitigated by user	Incidents resolved entirely by a human — potential automation opportunities
Pending user action	Incidents waiting for human input — your current backlog

The Incident summary chart plots all five metrics over time so you can spot trends.

Response plan breakdown

Below the chart, a Response plan grid breaks down performance per plan. Click any plan to drill into its incident history and root cause categories.

This is where decisions happen. You can see which plans run in Autonomous mode and resolve incidents without human involvement versus plans that still require approval. A plan with zero autonomous mitigations is a signal to adjust its instructions or increase its autonomy level.

Intent Met score

The Intent Met score measures how effectively your agent resolves work on a 1–5 scale. After each thread completes, an automated evaluation scores the outcome:

Score	Meaning
5	Exceptionally resolved — exceeded expectations
4	Well resolved — successfully completed
3	Partially resolved — made progress but didn't fully resolve
2	Poorly resolved — attempted but failed significantly
1	Completely unresolved — failed to address the core objective

The Intent Met card on the Overview dashboard shows the average score across all threads from the past 30 days with a daily trend sparkline. The score combines incident threads, scheduled task threads, and conversations into a single quality metric.

Intent Met scoring is fully automatic — no configuration needed.

Tip

If your Intent Met score is lower than expected, review individual threads in Session Insights to understand where the agent struggled.

Incident overview

Go to Incidents in the left sidebar for a real-time view of every incident your agent is handling. Each row links to the agent's investigation thread — review which tools it called, what evidence it found, and what it recommended.

Daily reports

Your agent generates automated daily reports at Daily reports in the left sidebar.

Select a date to view that day's report. Each report covers:

• Security findings — CVE vulnerabilities across connected repositories, grouped by severity
• Incidents — Active, mitigated, and resolved counts with per-incident details
• Health and performance — Per-resource health status with availability, CPU, and memory metrics
• Recommended actions — Prioritized action items with descriptions and estimated effort

Daily reports replace the "what happened overnight?" morning routine — the information is compiled and waiting.

Limits

Resource	Limit
Scorecard data	Retained in Application Insights (follows your workspace retention policy)
Daily reports	Generated once per day
Intent Met scoring	Applied automatically to incidents, scheduled tasks, and conversations

Get started

Incident tracking is built in — open Monitor > Incident metrics once your agent starts handling incidents.

Resource	What you'll learn
Set up a response plan	Configure response plans that generate tracking data

Related capabilities

Capability	What it adds
Automate incident response	Configure response plans for each incident type
Automate tasks on a schedule	Set up recurring tasks reflected in Intent Met scores
Monitor agent usage	Track AAU consumption alongside incident metrics
Review agent insights	Per-thread qualitative evaluation of agent performance
Audit agent actions	Review specific actions during incident investigations