---
title: Multi-View AI Investigation
description: >-
  Run an agentic investigation across views to find root causes for a slow
  performance vital.
breadcrumbs: Docs > RUM & Session Replay > AI Investigations > Multi-View AI Investigation
---

# Multi-View AI Investigation

{% callout %}
##### Join the Preview!

Multi-View AI Investigation is in Preview.

[Request Access](https://www.datadoghq.com/product-preview/rum-investigation-optimization/)
{% /callout %}

## Overview{% #overview %}

Multi-View AI Investigation runs an agentic root-cause analysis across a sample of views that share a slow performance vital. While [Single-View AI Investigation](https://docs.datadoghq.com/real_user_monitoring/ai_investigations/single_view_ai_investigation.md) explains why one view was slow, Multi-View AI Investigation explains why an entire (view × vital) pair is consistently slow across users. Use it from the Optimization page when you have identified a specific page and vital and want to know what to fix first.

The investigation runs at the (view × vital) granularity and is available for the following vitals:

- Loading Time
- Largest Contentful Paint
- First Contentful Paint
- Interaction to Next Paint

## Prerequisites{% #prerequisites %}

Multi-View AI Investigation is available for Browser RUM applications only.

## Launch an investigation{% #launch-an-investigation %}

1. Open the [Optimization page](https://docs.datadoghq.com/real_user_monitoring/application_monitoring/browser/optimizing_performance.md) for your application.
1. Select a view and one of the supported vitals.
1. The Optimization page surfaces two elements for the selected (view × vital) pair:
   - A summary at the top with an at-a-glance overview of the issues detected.
   - Up to three recommendation cards below the summary. Each card represents a candidate root cause, ordered by impact.
1. Click Investigate on a card to launch the agent on that issue.

{% image
   source="https://docs.dd-static.net/images/real_user_monitoring/ai_investigations/multi-view-ai-investigation-recommendations.b9c0773905eb00d282c65caf93e7b8e1.png?auto=format&fit=max&w=850 1x, https://docs.dd-static.net/images/real_user_monitoring/ai_investigations/multi-view-ai-investigation-recommendations.b9c0773905eb00d282c65caf93e7b8e1.png?auto=format&fit=max&w=850&dpr=2 2x"
   alt="Optimization page for a Largest Contentful Paint metric, showing the p75 distribution, time series, and ranked recommendation cards with an Investigate button on each card." /%}

## What the agent investigates{% #what-the-agent-investigates %}

The agent groups its findings into four diagnostic types:

| Source               | What is examined                                                                                                               |
| -------------------- | ------------------------------------------------------------------------------------------------------------------------------ |
| Resource bottleneck  | Slow resources (HTML, scripts, images) impacting the vital across views.                                                       |
| Vital element        | DOM elements that contribute most to the Largest Contentful Paint (LCP) or Interaction to Next Paint (INP) score for the view. |
| Top JavaScript files | Large JavaScript bundles or packages that dominate execution time.                                                             |
| Long tasks           | Long-running tasks that block the main thread during the vital window.                                                         |

The richer the data available, the more precise the analysis. Correlating RUM with [APM Traces](https://docs.datadoghq.com/real_user_monitoring/correlate_with_other_telemetry/apm.md) and enabling [Browser profiling](https://docs.datadoghq.com/real_user_monitoring/correlate_with_other_telemetry/profiling.md) helps the agent attribute findings to specific code paths.

## Read the results{% #read-the-results %}

When you click Investigate on a recommendation card, a side panel opens and the agent streams its analysis in real time. Each investigation produces:

- **A step-by-step timeline** describing how the metric is built from your application code: which resources load, when scripts execute, and where time is spent during the vital window.
- **A root-cause explanation** with the supporting evidence the agent considered.
- **A code-level finding** when the agent can attribute the issue to a specific file or function.

{% image
   source="https://docs.dd-static.net/images/real_user_monitoring/ai_investigations/multi-view-ai-investigation-code-finding.f1e16faffba1588a67bc46d536013044.png?auto=format&fit=max&w=850 1x, https://docs.dd-static.net/images/real_user_monitoring/ai_investigations/multi-view-ai-investigation-code-finding.f1e16faffba1588a67bc46d536013044.png?auto=format&fit=max&w=850&dpr=2 2x"
   alt="A Code Investigation side panel showing the inspected DOM element, a timeline of layout and render events, the root-cause explanation, and the affected source code." /%}

## Take action{% #take-action %}

After an investigation completes, you can act on findings without leaving the panel:

- Fix with Bits: Opens the Bits AI dev assistant with the investigation context pre-filled to generate a code fix from your IDE.
- Save to a Notebook: Exports the full timeline and findings to a [Notebook](https://docs.datadoghq.com/notebooks.md) to share with your team.

## Further reading{% #further-reading %}

- [AI Investigations](https://docs.datadoghq.com/real_user_monitoring/ai_investigations.md)
- [Single-View AI Investigation](https://docs.datadoghq.com/real_user_monitoring/ai_investigations/single_view_ai_investigation.md)
- [Optimize page performance with RUM](https://docs.datadoghq.com/real_user_monitoring/application_monitoring/browser/optimizing_performance.md)