Copilot Credits: Pay-As-You-Go vs Prepaid vs Pre-Purchase

Microsoft sells the same Copilot Credit three different ways once you set aside the zero-rated inclusion path: pay-as-you-go, prepaid capacity packs, and an annual pre-purchase plan. They buy an identical credit, but at different unit prices and with very different failure modes. Choosing badly is the most common avoidable cost in an early Copilot rollout — either paying the flexibility premium on volume you could have committed, or committing hard against a forecast you cannot yet defend. This guide sets out the economics and the decision logic. It is a companion to our Copilot Credits economics pillar.

The three paths at a glance

Pay-as-you-go: the flexible, expensive default

Pay-as-you-go is the most expensive credit and the most forgiving structure. There is no commitment and no bucket; you pay a cent a credit, billed in arrears through your Azure subscription, for exactly what your agents consume. That makes it the right starting point for any new deployment, because it lets you discover what an agent actually burns before you commit a dollar. It is also the correct structure for genuinely unpredictable or low-volume workloads, where the premium over a pack is small in absolute terms and the flexibility is worth it. The risk of pay-as-you-go is not a cliff — it has none — but an open-ended bill, which is why it must be paired with the governance caps covered in our governance guide.

Prepaid capacity packs: cheaper, with a cliff

A capacity pack is two hundred dollars for twenty-five thousand credits — eight tenths of a cent each, roughly a quarter cheaper than pay-as-you-go. The credits are pooled across every agent in the tenant and, critically, do not roll over: unused credits in a month are forfeited. The sharper catch is enforcement. Cross 125% of your pack capacity and Microsoft disables agents until you add credits or switch on pay-as-you-go. Because the pool is shared, one team’s reasoning-heavy experiment can exhaust the credits a different team’s production chatbot depends on, taking it offline with no warning.

Packs are the right structure for a predictable base load you can confidently consume each month. The art is sizing them to cover roughly 80% of expected usage rather than the peak, because over-buying packs wastes credits that do not carry forward, while under-buying without a PAYG fallback risks the cliff. They reward an estate that has enough history to know its floor.

The pre-purchase plan: lowest rate, annual commitment

The pre-purchase plan — bought as commit units through Azure — takes another bite off the rate, advertised at up to 20% below standard, in exchange for an annual upfront commitment. It has the feature packs lack: automatic pay-as-you-go overflow, so exhausting the commitment does not disable agents, it simply bills the excess at the pay-as-you-go rate. That removes the cliff while keeping the lower price. The trade is commitment risk: an annual pre-purchase sized on an optimistic forecast can leave you having paid upfront for credits you never consume. It belongs to mature, high-volume estates with months of consumption data to commit against, not to a first-year rollout.

All three touch your Azure commitment

One factor cuts across the choice: pay-as-you-go and the pre-purchase plan are both billed through Azure and count toward a Microsoft Azure Consumption Commitment. If you hold an underused MACC, that turns credit spend into a way to retire commitment you have already made — which can tip the economics toward routing AI spend through credits rather than a separate model contract. The interaction is worked through in our Credits and MACC guide, and it is the kind of detail your account team will model and most buyers will not.

The answer is usually to layer them

The structures are not mutually exclusive, and treating the decision as a single pick is the mistake. The pattern that works for most enterprises is layered and sequenced. Begin a deployment on pay-as-you-go and the inclusion path while you gather consumption data. Once a stable base load is clear, cover it with capacity packs or a pre-purchase commitment at the lower rate, and keep pay-as-you-go enabled underneath as the overflow valve that prevents the 125% cliff. Move from packs to a pre-purchase commitment only when you have the history to size the annual number without over-buying. The organisations that overspend are the ones that invert this order and commit hard on day one.

Four mistakes that cost real money

Buying packs to the peak. Because pack credits do not roll over, sizing them to your busiest month wastes credits in every other month. Size to the floor and let PAYG cover spikes. Running packs without PAYG. The 125% enforcement cliff disables agents tenant-wide when the shared pool empties; the fix costs nothing but a configuration toggle. Committing pre-purchase on a forecast. An annual commitment sized on a roadmap rather than consumption history locks you into credits you may never burn. Ignoring the MACC interaction. If you hold an underused Azure commitment, leaving credits on a separate billing path forfeits the chance to retire commitment you have already paid for.

A decision checklist

Before you pick a path, answer five questions in order. First, is the agent internal and serving licensed users? If so, it is zero-rated and none of this applies — build it on inclusion. Second, do you have at least two months of consumption data for this workload? If not, stay on pay-as-you-go until you do. Third, what is the demonstrable monthly floor — the consumption you are confident will recur? Cover that with packs or pre-purchase. Fourth, is pay-as-you-go enabled underneath as overflow? It must be, for any production agent. Fifth, do you hold an Azure commitment the credits could decrement? If so, weight the Azure-billed paths accordingly. Work those five in sequence and the right structure falls out of the answers rather than out of a sales conversation.