back to blog

Microsoft's Copilot Credits Could Change MSP Billing Economics

Read Time 4 mins | Written by: Gradient MSP

Copilot Credits bill AI by usage, not by seat. Here is how consumption pricing changes MSP billing, agreement design, and financial exposure.

For most of the last decade, Microsoft billing has been a seat game. A client has 50 users, each user has a license, and the MSP bills 50 times the per-seat price. The number changes when people join or leave, and the arithmetic is simple enough to check in a spreadsheet.

 

Copilot Credits change the shape of that arithmetic. They are the billing unit for Microsoft Copilot Studio, the platform used to build AI agents, and they replaced the older "messages" unit in September 2025. You do not pay per person. You pay for what the agents consume.

 

That is a different kind of bill, and it lands differently on an MSP's margin, contracts, and cash flow.

 

What Are Copilot Credits?

 

Copilot Credits are the unit Microsoft uses to meter the work AI agents do in Copilot Studio. Every interaction an agent handles draws down credits, and different features draw different amounts.

 

Microsoft offers three ways to buy them. Pay-as-you-go is billed monthly in arrears through an Azure subscription at a published rate of $0.01 per credit, with no upfront commitment. A capacity pack is $200 a month, billed annually, for 25,000 credits pooled across the tenant, which works out to $0.008 per credit. A pre-purchase plan lets customers commit upfront for a year. When a pack runs out, usage can overflow to pay-as-you-go.

 

Two details matter for MSPs. Rates can differ under enterprise or CSP agreements, so the published numbers are a reference point and not a guarantee. And there is no per-user fee for people who simply use an agent, which means the cost no longer tracks headcount.

 

Why Does Usage-Based Pricing Break the Per-Seat Model?

 

Per-seat pricing is predictable because the client's headcount drives it, and the MSP controls the count. Consumption pricing is driven by behavior, and the client controls the behavior.

 

Consider an illustration. A client's agent consumes 40,000 credits in a month. On pay-as-you-go that is $400. With a 25,000-credit pack plus overflow, it is $200 for the pack and $150 for the extra 15,000 credits, or $350. Next month the agent is more popular and consumes 90,000 credits, and the bill moves with it. Nothing about the client's headcount changed.

 

This is the economic shift. Seat pricing punishes you for adding people who use a tool lightly. Credit pricing punishes you for success, because the better an agent works, the more it runs. An MSP that resells credits at a flat monthly price has quietly taken on the risk of the client's enthusiasm.

 

Where Does the Financial Exposure Land?

 

Exposure lands in three places, and most agreements are silent on all of them.

 

The first is timing. Pay-as-you-go usage is billed after it happens. The client can consume heavily for weeks before anyone sees the number. If the MSP is the billing party, it may owe Microsoft before the client has been invoiced.

 

The second is control. Someone has to decide who may build agents, which environments are linked to a billing policy, and what happens when usage spikes. An agent that loops or is widely shared can consume credits quickly, and the bill does not distinguish between valuable use and accidental use.

 

The third is ownership. Pay-as-you-go requires an Azure subscription linked to each environment. Whose subscription it is, and who pays for it, shapes who carries the risk.

 

How Should MSPs Design Agreements for Credit-Based Services?

 

The goal is to decide, in writing and in advance, who carries the variable cost. A few structures work, and each trades predictability against margin.

 

  • Resell a fixed pack. The client pays a flat monthly amount for a defined credit allowance. The MSP buys capacity to match. This is predictable, but unused credits are waste and overages need a rule.
  • Pass through at cost plus a handling fee. The client pays actual consumption plus a defined markup. This protects the MSP from usage risk but requires clear reporting, because the client will want to see what they are paying for.
  • Tier the allowance. Offer a small, medium, and large credit allowance with a stated overage rate. This gives clients a menu and gives the MSP a reason to revisit the tier when usage grows.
  • Cap and alert. Whatever the model, define a usage threshold, who is notified when it is crossed, and who must approve spending beyond it.

 

Any of these can work. What does not work is leaving consumption to be settled after the fact, because that is where disputes begin.

 

What Does Reconciliation Look Like for Consumption Billing?

 

Seat-based billing can be checked against a count. Consumption billing has to be checked against a meter, and that raises the standard for reconciliation.

 

Each month there are at least three records that should agree: what Microsoft billed for credits, what the agreement says the client owes, and what was actually invoiced to the client. When the first moves and the third does not, margin disappears. When the third moves and the second does not, the client notices.

 

Reconcile exists for this kind of comparison. It checks vendor invoice data against billing agreements and surfaces the differences before they reach a client invoice. As Microsoft pricing moves from fixed seats toward metered usage, that monthly discipline matters more, not less. The Microsoft Add-on applies the same approach to Microsoft licensing, so variances in credits or seats are visible as they occur and not at quarter end.

 

FAQ

 

What are Copilot Credits? Copilot Credits are the billing unit for Microsoft Copilot Studio. They meter the work AI agents do, replaced the older "messages" unit in September 2025, and can be bought as pay-as-you-go, as a capacity pack, or through a pre-purchase plan.

 

How is Copilot Credit billing different from per-seat billing? Per-seat billing follows headcount, which the MSP can count and control. Copilot Credit billing follows usage, which the client's behavior drives. The bill can rise or fall without any change in the number of users.

 

Who carries the financial risk when a client's agent uses more credits than expected? Whoever the agreement says it is. If the agreement is silent, the risk often falls on the billing party, which may be the MSP. That is why agreements for credit-based services should define allowances, overage rates, and approval thresholds.

 

How should MSPs reconcile consumption-based Microsoft billing? By comparing three records each month: the vendor bill for credits, the amount the agreement says the client owes, and the amount actually invoiced. Differences between any two are the places to investigate.