← All insights

News analysisLens: United States4 min read

Microsoft calls Agent 365 Business Analytics an ERP for AI usage. Its value metrics need your definitions

At Ignite 2026 Microsoft added business analytics and cost management to Agent 365. Spend data is solid. Success rate, satisfaction and assisted hours become value only when the outcome is defined.

An old hanging scale with a brass pan against a white wall, in La Madre duotone, beside the words Spend meets value
Photo: rawpixel (CC0)

“You can think of it as your ERP system for AI usage.” That is how Satya Nadella introduced Agent 365 Business Analytics in the Ignite 2026 keynote on November 17. It is a useful comparison, and it also contains the catch. An ERP records transactions accurately. It does not decide whether the purchase was worth making.

What Microsoft announced, and its status

Agent 365 is Microsoft’s control plane for agents, which Nadella said already holds more than 50 million agents. At Ignite it gained two related layers, aimed at two audiences:

  • Agent 365 Cost Management, for IT, in the Microsoft admin center: one place to see and control usage-based AI services, including Copilot Cowork, Copilot Code, GitHub Copilot CLI, Copilot Studio, Copilot Managed Runtime, Autopilot and Work IQ APIs. Admins can see what drives spend, set spending policies, limits and alerts, and extend the analysis in Microsoft Fabric for chargeback. The Book of News lists this in preview; forecasting, plain-language questions and optimization recommendations come through the Frontier program.
  • Agent 365 Business Analytics, for business leaders, available in the Frontier program: what drives cost across agents, services, models and users; measures of value for specific jobs and tasks, namely success rate, user satisfaction and assisted hours; and cost scenarios with recommendations.

Insights for Microsoft services require at least one Microsoft 365 Copilot license in the tenant; insights for non-Microsoft services require an Agent 365 license. In December, Microsoft says, value insights will extend across Microsoft and non-Microsoft agents, and analysis can be delegated to business leaders.

Frontier is Microsoft’s early-access program. None of this is generally available yet, and the cross-vendor part is a plan for December.

Spend visibility is the part to adopt now

The cost side solves a real problem. AI spend at most companies is spread across seat licenses, credit pools, Azure meters and separate products, each with its own report. A single view with policies and alerts, exportable to Fabric for chargeback, is what finance has been asking for. We listed five layers of agent spending controls and noted that only a few actually stop a bill; limits enforced per service, in the admin center, are among them.

If you run Copilot Studio agents or Copilot Cowork at any scale, put an owner on the Cost Management view during the preview, set alerts per department, and check whether its numbers match what each product’s own report says. Two reports that disagree are a finding.

Value metrics are where the definitions come in

The value side is more delicate. Success rate, satisfaction and assisted hours are activity and quality signals. Each one becomes business value only after an operating decision:

Success rate depends on what counts as success. For a support agent, is it a conversation that ended without escalation, or a customer whose problem stayed solved a week later? The platform can count the first. Only the process owner can define the second.

User satisfaction measures how people felt about the assistance. It correlates with value when the task is discretionary and poorly when the task is mandatory: people rate a compliance check badly even when it saved the company from a fine.

Assisted hours is the most tempting and the most misleading. As we argued in our analysis of agent ROI, hours saved are capacity. They become value only when someone decides what that capacity is used for: more cases handled, a hiring plan changed, a backlog cleared. A dashboard that adds up assisted hours across the company will produce an impressive number that no budget line reflects.

None of this is a flaw in Microsoft’s product. It is the reason the product needs a counterpart inside your company: for each agent that matters, a written definition of the outcome it serves, the baseline before it existed and the operating decision its savings fund.

A practical setup for the preview

Name an outcome per agent. One sentence, owned by the business, for example “invoices matched without manual review” rather than “time saved in finance.”

Record the baseline before rollout. Volume, cycle time and error rate of the process without the agent. Agent 365 will show the after; you need the before.

Pair cost with the outcome, not with usage. Cost per resolved case or per approved invoice says more than cost per user or per token.

Treat Microsoft’s value metrics as inputs. Use success rate and satisfaction to find agents that are failing. Use your own outcome measure to decide which ones deserve more budget.

The ERP comparison holds: it will tell you, accurately, where the money went. Whether it was well spent is still a decision your company has to make.

Have an AI use case stuck between prototype and production?

Tell us what you’re trying to ship. We’ll reply with honest next steps.

Discuss a use case