Microsoft puts a dollar value on AI agent work across 6000 task types, but it's only an estimate
A CFO asks: are our AI agents paying off? Microsoft has launched Agent 365 Business Analytics to help answer that. The service takes a task an agent performed, such as processing customer requests, estimates how many hours a human would have needed, and converts those hours into dollars using rates from US Department of Labor data. The system describes more than 6000 task types. But Microsoft itself calls the result an estimate, not actual savings, so you can't build a budget on a single dashboard.
AI-processed from ZDNet AI; edited by Hamidun News
Companies are spending more and more on AI agents, and the CFO asks a simple question: we pay for tokens, subscriptions and cloud, but is it paying off? On November 17, Microsoft answered with a product called Agent 365 Business Analytics. Say an agent processed a hundred customer requests in a day.
The service estimates how many hours a human employee would have spent on that, prices those hours at market rates, and compares the figure with the actual cost of the AI. The system describes more than 6000 types of work tasks, so the calculation can be applied to a wide range of scenarios. But there is an important caveat: the result is an estimate of "what this work would have cost if a human did it," not proven savings.
If you are responsible for an AI budget, that difference is the main reason to read on.
How the cost of agent work is calculated
First, some context. Microsoft showed Agent 365 earlier this year. It is a registry and control panel for agents that treats them roughly the way an HR department treats employees: each one has an identity, access permissions and a history of actions. Business Analytics adds a financial layer on top.
The logic works like this. The service takes a task completed by an agent and determines how long a human would typically need for it. It then prices that time at typical rates. For this, Microsoft uses data from the US Bureau of Labor Statistics, the US Department of Labor's O*NET occupation database, and LinkedIn's own occupation graph (LinkedIn Occupational Graph). Next, the system checks which part of the task was done by a human and which by the agent, and adds the actual AI costs: subscriptions, tokens, in-house infrastructure and cloud.
The output is a hybrid estimate: the human contribution, the notional savings from replacing human hours, AI costs, and the net difference. The service is supposed to show what individual agents, models and users cost, what exactly the AI is doing, and whether it makes sense to increase or cut the budget.
Why you should treat the numbers with caution
ZDNET columnist David Gewirtz points to a line from Microsoft's own blog: "assisted hours are a directional estimate of AI-assisted work, not measured time savings or financial return." The dashboard shows what the work would have cost if a human had done it. But it does not prove that the company actually saved a single dollar.
The reasons are straightforward. If no one reduced employees' hours, salaries are still being paid, and the "savings" exist only on paper. Also, a task may have gone to an agent simply because it was handy, not because a human would have taken longer and cost more. Gewirtz jokes that this tool comes from the company that created PowerPoint: impressive charts may well be detached from reality. Still, he does not consider the product useless. In his words, it is one possible source of truth, but not the only one.
Why Microsoft wants this
The product fits into a broader strategy. Microsoft wants to be the control center for AI in the enterprise. To that end, it offers to monitor agents from different vendors, manage their permissions and security, cap spending, assess returns, and decide which AI systems deserve further investment. Integrations are announced with OpenAI, Anthropic, Salesforce, Google, Databricks, Snowflake and others. That means Microsoft does not have to sell you a model or an app. It is enough to run the "plumbing" and play the role of HR for AI employees to take a central place in a business's day-to-day operations.
What to do in practice
For executives and product managers, the takeaway is simple. Use these reports as a yardstick for comparing agents and scenarios with each other: where AI takes on the most work, and where token costs are growing faster than the benefit. But don't substitute them for real financial reporting. To claim savings, check the estimates against measurable indicators: whether process turnaround time has dropped, whether overtime has disappeared, whether team throughput has grown, whether actual AI bills have changed.
Those building their own internal agent-tracking tools should borrow both the idea and the caution. Keep "estimated labor value" and "confirmed savings" in separate columns, and don't merge them into one tidy ROI figure, that is, return on investment. Otherwise it is easy to make budget decisions that rest on nothing but a dashboard.
- Nov 20, 2026Microsoft and Salesforce Build CRM Into Copilot: 38+ Ready-Made Skills for Sales Reps
- Nov 20, 2026US Outage Wave: Officials See No Attack, Experts Blame Rush to Patch After AI Discoveries
- Nov 20, 2026Unsiloed Raises $7.5M to Help AI Agents Read PDFs: 99.78% of Values Correct in Its Own Test
- Nov 20, 2026Anthropic raises $81.3 billion in IPO, beating SpaceX's $75 billion record
Key Questions Answered
How the cost of agent work is calculated?
First, some context. Microsoft showed Agent 365 earlier this year. It is a registry and control panel for agents that treats them roughly the way an HR department treats employees: each one has an identity, access permissions and a history of actions. Business Analytics adds a financial layer on top. The logic works like this. The service takes a task completed by an agent and determines how long a human would typically need for it. It then prices that time at typical rates.
Why you should treat the numbers with caution?
ZDNET columnist David Gewirtz points to a line from Microsoft's own blog: "assisted hours are a directional estimate of AI-assisted work, not measured time savings or financial return." The dashboard shows what the work would have cost if a human had done it. But it does not prove that the company actually saved a single dollar. The reasons are straightforward. If no one reduced employees' hours, salaries are still being paid, and the "savings" exist only on paper.
Why Microsoft wants this?
The product fits into a broader strategy. Microsoft wants to be the control center for AI in the enterprise. To that end, it offers to monitor agents from different vendors, manage their permissions and security, cap spending, assess returns, and decide which AI systems deserve further investment. Integrations are announced with OpenAI, Anthropic, Salesforce, Google, Databricks, Snowflake and others. That means Microsoft does not have to sell you a model or an app.
Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews
Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.