Foundry Agent ROI Dashboard: What Admins Should Prepare

Microsoft opened a private preview of the Foundry agent ROI dashboard on July 8, 2026. It translates raw agent runs into business value: task completion, time saved and cost per completed task. On the same day, Copilot in Microsoft Forms reached global availability. Together, the two announcements mark a shift from “ship the agent” to “prove the agent pays off,” and give admins, finance partners and platform owners a new reporting surface to plan around.

What the Foundry agent ROI dashboard measures

The dashboard sets the cost of running an agent against three business dimensions: task completion rate, time saved and cost efficiency. The metrics appear side by side in the Foundry portal and through the platform API. Teams can compare versions of the same agent, track daily trends and drill into low-ROI traces to debug specific runs.

It isn’t a separate data store. The dashboard reads from Foundry’s existing tracing pipeline, which already captures inputs, outputs, tool use, retries, latency and per-run cost. ROI is a new view on traces you’re probably already collecting.

• Task completion rate across agent versions and time windows.
• Time saved per successful run, expressed in human-hour equivalents.
• Cost efficiency in dollars per completed business task, not just per token.
• Version deltas so a v2 rollout can be measured against the v1 baseline.
• Low-ROI trace drill-down for expensive or failing runs.

Why Microsoft launched it now

The timing wasn’t accidental. Microsoft 365 list prices rose on July 1, 2026, and the Copilot Business bundles became permanent SKUs (see Microsoft 365 pricing changes in 2026). A week later, CFOs were asking whether agents pay for themselves. The ROI dashboard gives IT a native answer instead of a Power BI report stitched together from billing exports.

It also follows Build 2026, where Foundry positioned itself as the runtime, tooling and governance layer for production agents. Microsoft has said granular cost-per-agent-run reports linked to Azure billing will follow, which would make chargebacks between IT and business units far easier.

Framework coverage: LangChain, LangGraph and OpenTelemetry

The dashboard isn’t limited to agents built with Microsoft Agent Framework. Foundry’s tracing and evaluations extend to agents built on LangChain, LangGraph and the OpenAI SDK, and any custom framework can plug in through OpenTelemetry. A SharePoint knowledge agent, a LangGraph finance analyst and an OpenAI SDK support bot can all be compared in one view.

1. OpenTelemetry emitters on custom agents become a governance requirement.
2. Business-outcome tags on traces (for example, ticket resolved) are what unlock ROI comparisons.
3. Consistent version labels on deployed agents make version-over-version deltas work.
4. Azure cost-allocation tags should already match the agent boundary you want to report on.

Getting access to the preview

At launch, access was gated. Organizations applied through the Microsoft Foundry preview portal, and eligibility required either Microsoft AI Cloud Partner Program membership or an enterprise agreement. Microsoft said it was prioritizing manufacturing, financial services and healthcare scenarios. A strong nomination names one production agent, one measurable business outcome and one executive sponsor who will use the dashboard. Check the Foundry blog for the current preview status.

Copilot in Forms went global the same day

Also on July 8, Copilot in Microsoft Forms reached global availability for customers with qualifying Microsoft 365 Copilot licenses, rolling out to every Copilot market within two weeks. It works in the Forms web app, Teams and mobile. It generates surveys and quizzes from natural language, suggests improvements to existing forms, summarizes responses, and translates forms while keeping branching logic.

Admins manage it in the Microsoft 365 admin center under Copilot settings, with controls at the tenant and user-group level, so a phased pilot (for example, HR first) works without a tenant-wide switch.

Governance and chargebacks

• Cost center tagging on the Azure resources behind each agent must be clean before billing integration lands, or chargebacks fall apart on day one.
• Business-outcome labels should be defined by the business owner, not the platform team, so leadership trusts the numbers.
• Retention of ROI data needs an explicit decision: some organizations want multi-quarter trends, while regulated tenants may prefer shorter windows.
• Forms data will grow faster once authors use Copilot, so review retention, sensitivity and DLP policies for Forms too.

Foundry agent ROI dashboard readiness checklist

1. Confirm every production agent sends OpenTelemetry traces to Foundry.
2. Tag traces with a stable business-outcome field (for example, ticket_resolved=true).
3. Standardize agent version labels.
4. Check Azure cost-allocation tags match the agent boundary you’ll report on.
5. Prepare a preview nomination with a named sponsor and one measurable outcome.
6. Run a Forms Copilot pilot with a friendly business unit and clear success criteria.
7. Add both features to your existing Copilot governance and DLP baseline. Our Copilot Agent Builder governance guide covers the owner and review model.

Bottom line

The Foundry agent ROI dashboard puts business value next to latency and cost in the same portal where agents are built. Tenants that clean up tracing, tagging and outcome labels now will be reporting chargebacks when Azure billing integration arrives, while others are still tagging resources.

Last reviewed: October 2026

Sources

• Build 2026: From observability to ROI for AI agents on any framework (Microsoft Foundry Blog)
• Microsoft pilots agent ROI dashboard in Foundry, Forms Copilot hits global availability (Windows News)
• Agent tracing in Microsoft Foundry (Microsoft Learn)
• Monitor agents with the Agent Monitoring Dashboard (Microsoft Learn)
• Tracking every token: cost and usage metrics for Foundry agents (Microsoft Community Hub)
• Release notes for Microsoft 365 Copilot (Microsoft Learn)