Copilot Credits in your Microsoft EA. How the meter is priced, bought and kept under control.

Copilot Credits are one pooled, one cent meter for agentic AI that sits on top of the Copilot seat. Budget the seat and the meter separately, set caps before provisioning, and size any prepay from measured burn.

What are Copilot Credits, and what do they meter?

Copilot Credits are Microsoft's single usage currency for agentic AI. One credit costs one cent at list, and every credit your tenant consumes comes out of one pooled balance, whichever product drew it.

The currency replaced the message based meter in September 2025. The per user Copilot subscription still covers Copilot Chat, the in app experiences and built in agents such as Researcher and Analyst. Credits meter the autonomous work on top of the seat, and they arrive on a different invoice from a different budget.

Which workloads draw on the same balance

• Copilot Cowork and Copilot agents. Autonomous tasks that read mail and files and write results back. Cowork has billed per task since its general availability on June 16, 2026, and it is typically the heaviest draw on the pool.
• Copilot Studio and Dynamics 365 agents. Custom agents your teams build and the first party agents in Dynamics 365.
• The Work IQ APIs. Generally available from June 16, 2026 and billed from the same pool. Tools API calls cost 0.1 credit each, while chat and context calls are billed by task complexity.
• Copilot Chat for unlicensed users. When staff without a Copilot seat use agents in Copilot Chat, their consumption is metered to the pool instead of a seat. This is the expansion path that grows without a purchase order.

One meter makes the invoice simpler. It also puts every AI workload in your tenant onto a single currency whose rate card Microsoft sets and can change at any renewal.

What one credit actually pays for

A credit bundles four things: the model, the runtime that orchestrates the agent, the context retrieved from mail, files and calendars, and the tool actions the agent takes. A task therefore bills by how hard it was, and a long, file heavy task can cost seven times or more what a short one does.

The bundle also means you cannot reverse a credit into a clean token rate, which makes the meter hard to benchmark against a direct model price.

Where Microsoft 365 E7 fits

E7 lists at $99 per user per month against $117 for its components (E5, Copilot, Agent 365 and Entra Suite) bought separately. It includes the Copilot seat, but agent execution still bills through Copilot Credits with no rollover, so the meter needs its own budget line even on E7. Our E7 guide covers the suite.

How much does a Copilot Credit task cost?

A single task costs anywhere from under $2 to more than $15, depending on how much work the agent does. Microsoft's own Copilot Credits Guide gives three Cowork examples that show the spread.

Microsoft calls these rough planning estimates. The same prompt can land in a different band depending on how many files the agent opens and how many tool calls it makes, so test on your own content. Our cost per task analysis goes further.

How do you buy Copilot Credits, and which routes count toward your MACC?

There are three routes: pay as you go, Capacity Packs and the Pre-Purchase Plan (P3). Microsoft's Copilot Credits Guide states that pay as you go and P3 both decrement your Microsoft Azure Consumption Commitment (MACC), so credits bought that way count like any other Azure spend.

Two details catch finance teams out. Capacity Packs are a license subscription, so do not assume they burn down the commitment. And Microsoft's P3 documentation says the plan is billed as a separate charge, not paid from an Enterprise Agreement's Azure Prepayment balance. Get the MACC treatment confirmed in writing.

If you hold more than one route, Microsoft draws Capacity Pack credits first, then P3 credits, then pay as you go. Oversized packs therefore push P3 credits toward expiry, so size the two together.

The P3 tier ladder in full

P3 has nine tiers, from 300,000 credits at 5 percent off to 300 million at 20 percent. The term is one year, purchases are final with no cancel or exchange, and plans renew automatically unless you change the setting.

Why the 0.8 cent floor matters

Across the whole ladder the effective price only falls from 0.95 to 0.8 cents. Even a $3 million commitment pays four fifths of list, and a Capacity Pack used to the last credit already reaches 0.8 cents without an annual pool.

At a discount of d percent, you must use at least 100 minus d percent of the pool to beat pay as you go. A Tier 4 buyer has to consume 92 percent of the credits and a Tier 9 buyer 80 percent. A pool that expires 30 percent unused costs more than pay as you go at every tier.

How should you size a Copilot Credits prepay?

Size it from measured burn, never from the adoption forecast in the proposal. Run pay as you go for a measurement period and let the demonstrated volume choose the tier, as our pay as you go versus prepurchase comparison explains.

A worked example with 400 Cowork users

Say you enable Cowork for 400 users with the mix below. Credits per task are points inside Microsoft's ranges, and the tasks per user are assumptions for illustration.

A year at that rate is 16,368,000 credits, or $163,680 on pay as you go. Tier 4 covers 15,000,000 credits for $138,000, and the remaining 1,368,000 credits bill at list for $13,680. The total is $151,680, a saving of $12,000, or about 7 percent.

The 20 heavy users are 5 percent of the population but draw more than a tenth of the credits, so they deserve their own threshold.

What a 30 percent forecast miss costs

Now assume real usage lands 30 percent below the model, at 11,457,600 credits for the year. On pay as you go that costs $114,576. On Tier 4 you still paid $138,000, and 3,542,400 credits expire at term end, worth $35,424 at list.

The prepay leaves you $23,424 worse off than paying as you go. About 24 percent of the pool expired, which is squarely inside the range we saw in real deals.

Why do Copilot Credits slip past spend governance?

Credit burn looks like progress on a commitment the company already signed, so it rarely gets a second look. In the renewals we worked on, credits drawn against an underused MACC felt free, so spend governance was the first control to slip.

The invoice arrives as Azure spend that finance approved at signature, while the seat budget sits with another team, so the meter has no owner. Our credits and MACC analysis covers the accounting.

Controls to set before you provision

• Tenant spend caps. Set a monthly ceiling in the Microsoft 365 admin center before the first agent goes live.
• Per persona thresholds. Group and user limits sized to each role's task mix, so a few heavy users cannot drain the pool. Microsoft treats these spending policies as access controls, and a user who hits the limit loses access until the monthly reset.
• Alerts and an owner. Alerts well below the cap, routed to one named budget holder for credits, separate from the seat line and the Azure commitment.
• Apps built with Cowork. Bring them inside your spending policies from the start, since apps built with Cowork and calls to the Work IQ API consume credits under usage billing.

How to check what you are actually burning

The Microsoft 365 admin center reports consumption in credits, under Copilot and Cost Management, while Azure Cost Management shows the dollar charge and can lag by up to 24 hours. The admin center view can include non billable usage, so reconcile against the monthly Azure billing record, not the dashboard.

The Power Platform admin center shows how Copilot Studio capacity is allocated across environments. In Cowork, the /cost command shows what the open task has used and the user's remaining monthly limit, which suits spot checks with pilot users.

Filling an underused MACC with credits is a poor trade

Account teams often suggest pointing Copilot Credits at a MACC you are behind on, since the money is committed anyway. We disagree with that advice. A commitment shortfall is a contract problem, and closing it with agent tasks buys work you did not plan for, at list rates, with no governance attached.

Fix the commitment on its own terms through a mid term MACC renegotiation or a resized renewal, then approve credit spend on its own business case. Our MACC sizing guide shows how to set the next commitment from real consumption.

Are Copilot Credits cheaper than building agents on direct model contracts?

Not for heavy agentic work. Heavy workloads priced in Copilot Credits ran several times the cost of the same work built on direct model rates, and the same task on Claude direct costs a fraction per inference unit. The gap pays for the managed agent, meaning the runtime, retrieval and tool orchestration Microsoft runs for you.

Cost the same task on Copilot Credits and on a direct model contract that uses batch processing and prompt caching. The direct route's effective rate falls far below the 0.8 cent prepay floor. That runtime is worth real money without platform engineering; with it, price the runtime explicitly and show Microsoft the number.

Where to find the numbers for both sides

• Seat economics. The Copilot licensing guide.
• The EA the credits sit inside. The Enterprise Agreement guide.
• MACC sizing. The Microsoft overspending report.
• Direct model rates. The token economics report, the Anthropic pricing history and Copilot Credits versus Claude direct.

What have we seen in Microsoft EA renewals with Copilot Credits?

Across roughly 30 to 40 Microsoft EA renewals I benchmarked in 2024 and 2025, the Copilot conversation changed inside a single year. It began as a flat per user add on and ended as a variable meter sitting next to the seat.

• Expired prepay. Pools sized on vendor forecasts expired with 10 to 30 percent of their credits unused at term end, and the value was written off.
• A managed premium. Heavy agentic workloads cost several times what the same work cost on direct model rates.

The pattern matches every consumption meter we work on, with one Microsoft twist. The MACC makes credits the easiest AI spend in the company to approve, because the burn counts toward a commitment someone already signed, and the hardest to keep track of for the same reason.

A prepay pool sized on the adoption pitch is a bet that your users will work exactly as Microsoft forecast.

The buyers who kept costs predictable modeled the task mix in dollars first, installed caps before provisioning, kept the license and the meter as two budget lines, and treated the prepay ladder as a capped discount on an expiring pool, bought from measured burn or not at all.

What will the Microsoft account team say about Copilot Credits, and how should you answer?

Expect credits to arrive bundled with the seat renewal and the MACC. These are the lines we hear most often.

Contract terms to ask for

The discount tops out at 20 percent, so the terms that protect you from forecast error and rate changes are worth more than a deeper tier.

1. A rate hold. The one cent list rate and the credit weights per action fixed for the EA term, or a notice period before any change applies.
2. No automatic P3 renewal. Switch off the default so each year's tier is a fresh decision on measured data.
3. A tier step up. The right to move to a higher P3 tier mid term with credit for what you already paid.
4. Written MACC treatment. Confirmation of which purchase routes decrement the MACC and how P3 is invoiced against your Azure Prepayment.
5. A pack reduction right. The right to cut the Capacity Pack count at each anniversary, because packs bill annually and unused monthly credits are lost.
6. A funded pilot. Microsoft funded credits for a measurement period before any prepay decision.

The governance side of the negotiation is covered in our credits governance guide.

What to do next

1. Before provisioning. Model the task mix in dollars for each persona, using Microsoft's light, medium and heavy bands as the starting point.
2. Before the first agent goes live. Install tenant caps, per persona thresholds and alerts, and name one owner for the meter.
3. In the budget. Carry the license and the meter as two lines, the subscription and the agentic work it does not cover.
4. For the first three months. Run pay as you go, reconcile admin center credits against Azure Cost Management each month, and record burn per workload.
5. Before any prepay. Size P3 from measured burn, apply the break even test to the tier you are offered, and switch off automatic renewal.
6. Before the EA renewal. Run the buy or build comparison on outcomes, pricing the managed agent against direct model rates, and take the contract asks to Microsoft. Our Microsoft practice runs the model with you.

Frequently asked questions