## Overview

The Run Comparison page lets you overlay metrics across 2 to 5 runs of the same test. This is a powerful way to track changes and catch regressions over time.

**AI Run Comparison** takes it further. Alongside the chart, it generates a structured breakdown of what changed, what regressed, and where to investigate, so you spend less time cross-referencing metrics and more time acting on results.

{{< arcade id="vw2t8jZLKjw8BX6W1vUg" title="Introducing AI run comparison" >}}

## How to use it

1. From a run detail page, click **Compare runs** to open the **Run Comparison** view.
2. Select **2 to 5 runs** to compare. The **Compare with AI** button becomes active once at least 2 runs are selected.
3. Click **Compare with AI**.
4. The report appears above the comparison chart. The chart stays available for further exploration.

{{<  alert info >}}
AI features can be disabled organization-wide by an administrator. If you don't see the **Compare with AI** button, contact your organization admin.
{{< /alert >}}

{{< img src="ai-run-comparison.png" alt="Run Comparison page with the AI comparison button" >}}


## What you get

The report is structured around three blocks:
 
**Findings** opens with a one-sentence summary of the most significant pattern across the selected runs, followed by a breakdown of the key differences: throughput, error counts, latency percentiles, CPU usage, and request-level details where relevant. No interpretation, just what the data shows.
 
**Recommendations** opens with a one-sentence focus for your investigation, followed by 2 to 4 concrete actions to take before the next run.
 
**Explore in the chart** opens with a one-sentence summary of what to look at, followed by a precise starting point, which metrics to select and which run pairs to compare in the chart below.
 
The report also includes a **verdict** and a **confidence level** tags. The verdict summarizes the overall relationship between the runs in a single word (`Similar`, `SomeDiscrepancies`, or `Divergent`). The confidence level (`Low`, `Medium`, or `High`) reflects how much signal was available. A low confidence may indicate that a run was stopped early or that data was insufficient to draw firm conclusions.

{{< img src="ai-run-comparison-details.png" alt="Example of generated AI run comparison" >}}


## Notes

- Select the same set of runs again to retrieve a previously generated report.
- AI Run Comparison is only available when the AI feature is enabled for your organization.
- If report generation fails, it cannot be retried from the UI in the current version. Please contact the support.

{{<  alert info >}}
AI-generated insights are provided for informational purposes only. Always verify results before making decisions.
{{< /alert >}}