A risk adjusted Azure commitment strategy covers only the stable baseline of your usage and leaves the variable top uncommitted. Reservations give the deepest discount, often in the region of 40 to 62 percent against pay as you go for a three year term (indicative, confirm on the current Azure pricing page), but they bind you to a VM family and region. The Azure Savings Plan for compute gives a smaller discount in exchange for flexibility across families, regions, and operating systems. Reservations can be exchanged or refunded within limits; the Savings Plan cannot be cancelled at all. Coverage should follow a forecast you can defend, not the largest number a discount calculator shows.
Commitments are where most Azure savings live and where most stranded spend is created. The mechanics below decide which of those two outcomes you get.
What is the difference between a Reservation and the Azure Savings Plan?
A Reservation is a one or three year commitment to a specific VM family in a specific region. In return you receive the deepest available discount on that capacity. Reservations apply to virtual machines and also to many other services such as SQL Database, Cosmos DB, and storage. They carry instance size flexibility within a family, so a reservation for one size automatically covers smaller and larger sizes in the same group at a proportional ratio.
The Azure Savings Plan for compute is a one or three year commitment to an hourly dollar amount of compute spend. Any eligible compute usage, across VM families, regions, and operating systems, draws down that hourly commitment at discounted rates until the hourly amount is consumed. The discount is lower than an equivalent reservation, and that gap is the price you pay for flexibility.
The buyer rule is simple. Reservations win on price for steady, predictable workloads that will not move families. The Savings Plan wins on resilience for a fleet that changes shape, migrates regions, or rebalances families across the term.
Why does the cancellation rule decide your real risk?
Discount depth is only half the decision. The other half is what happens when your forecast is wrong.
Reservations can be exchanged or refunded, within limits. You can exchange a reservation for another of equal or greater value, and you can cancel for a refund subject to an early termination fee and an annual cancellation cap (indicative figures, confirm current Microsoft terms). That escape hatch makes a reservation less risky than its term length suggests.
The Azure Savings Plan cannot be cancelled, exchanged, or refunded. Once purchased it bills the committed hourly amount for the full term whether or not you use it. Its flexibility is in where the commitment applies, not in whether you can exit. That makes oversizing a Savings Plan a more permanent mistake than oversizing a reservation.
So the apparent safety order inverts. The Savings Plan feels safer because it floats across your estate, but it is the one you cannot walk back. Size it to the floor of your usage, the spend you are certain will exist for the whole term.
How much coverage should you actually buy?
Coverage is a forecast problem, not a discount problem. Start from twelve months of billing history, strip out anything already retired or migrating, and find the stable baseline that has held for the trailing quarter. That baseline, not the peak and not the average, is your commitment target.
A defensible split for most enterprises looks like this. Cover the hardened baseline with three year Reservations where the family and region are settled. Cover the next band of steady but less certain usage with a one year Savings Plan for flexibility. Leave the volatile top, the part that scales with traffic or seasonality, on pay as you go. Layer in Azure credits and MACC drawdown so committed purchases count toward any enterprise commitment you already owe, and stack Azure Hybrid Benefit where you own eligible licences.
A worked example: covering a baseline, not a peak
A European SaaS company ran a steady production fleet plus a spiky batch tier on Azure. A vendor calculator suggested a three year Savings Plan sized to the total run rate, promising the biggest headline discount. Read against billing history, only about 64 percent of that run rate had held for a full year; the rest was batch that swung with customer load. We covered the hardened 64 percent with three year Reservations on the settled families, added a modest one year Savings Plan for the migrating tier, and left the batch on pay as you go. Effective discount on covered spend landed near the reservation rate, commitment utilization stayed above 95 percent, and no spend was stranded when the batch tier later moved families. Figures are verified against billing data and anonymised.
The lesson: the largest commitment is rarely the cheapest one. The cheapest one is the largest commitment you will still fully use in month thirty six.
How this fits a wider Azure cost program
Commitments only pay off on top of a clean estate. Rightsize and retire waste first so you are not committing to oversized machines, decide deallocating versus stopping VMs for non production, then layer commitments over the baseline that remains. Track utilization and coverage monthly and exchange reservations as families shift. The full sequence sits in the Azure cost optimization guide, and the same forecast discipline carries across providers in the cross cloud cost optimization guide.
Frequently asked questions
Get the buyer side commitment playbook
We size Azure Reservations and Savings Plan coverage to a forecast your finance team can defend, with zero provider commissions and no incentive to oversell a commitment. It is the same discipline behind the 31 percent median reduction we see in the first 90 days. Our guarantee: we reduce your cloud spend or we reimburse our service fee.
Put a defensible number on your cloud spend.
No provider in the room, no published price list. Tell us your footprint and we will scope the savings against your billing data — we reduce your cloud spend or we reimburse our service fee.
The Cloud Spend Navigator: what changed in cloud pricing, commitments, and FinOps — no vendor spin.