The reserved portfolio was usually sized to last year's footprint rather than next year's, and standard reservations locked the wrong instance family in about a third of estates
Both commitment vehicles deliver the same discount band. The choice between them is not about price, it is about which kind of change you expect to survive.
Prepared by Redress Compliance · August 19, 2026 · AWS commitment reviews. 25 to 35 reviews run, 2024 to 2025.
Executive summary
Standard reservations locked the wrong instance family in about a third of estates after a workload refactor the commitment could not follow.
Coverage targets set above 80 percent left idle reservations worth 10 to 20 percent of the commitment. Full coverage is a target that costs money at the margin.
Convertible reservations and spend based plans together covered volatility better than reservations alone in most accounts, which is a portfolio decision rather than a product one.
Both vehicles deliver thirty to seventy two percent off on demand. The choice sits in flexibility, and the programme discount then stacks on top of whichever you signed.
What is the actual difference between the vehicles?
Flexibility, not discount. Reservations pre commit to capacity in a region for one or three years. Spend based plans pre commit to a dollar figure across services for the same terms and the same discount band, priced on the compute pricing pages.
| Mechanic | Discount band | Family lock | Service coverage |
|---|---|---|---|
| Standard reservation | 40 to 72 percent | Yes | Compute instances only |
| Convertible reservation | 35 to 66 percent | No | Compute instances only |
| Instance spend plan | 40 to 72 percent | Yes | Compute instances only |
| Compute spend plan | 30 to 66 percent | No | Instances, containers and functions |
Each one wins in a different estate
Standard suits predictable workloads on stable families running continuously. Convertible suits workloads that may shift family inside the term. The broad spend plan suits a mixed compute estate, and pays roughly five points for that reach.
Where does the break even actually sit?
Around six months of utilization on a one year term, and around eighteen months on three years. Below that the commitment costs more than paying as you go, which is the arithmetic behind every stranded reservation.
At a 40 percent discount the break even utilization is roughly half the term. At 72 percent it falls to about a third. The deeper the discount, the more forgiving the commitment is of a workload that moves.
That is the number to test a coverage target against. Pushing coverage above 80 percent left 10 to 20 percent of the commitment idle in the reviews, which is the break even failing quietly across the tail of the portfolio.
The reserved capacity optimisation brief
When reserved capacity beats a spend based plan, when it does not, and how to size either.
Get the brief →What 25 to 35 AWS commitment reviews showed
Across roughly 25 to 35 AWS commitment reviews run between 2024 and 2025, the reserved portfolio was usually sized to last year's footprint rather than next year's. Three patterns recur.
- Standard reservations locked the wrong instance family in about a third of estates after a workload refactor.
- Coverage targets set above 80 percent left idle reservations worth 10 to 20 percent of the commitment.
- Convertible reservations and spend based plans together covered volatility better than reservations alone in most accounts.
A commitment is a forecast with a penalty attached. Sizing it from the previous year is forecasting the past, which is why the family lock kept catching people.
- Commitments sized from your actual consumption curve rather than a vendor forecast
- Right sizing for databases, storage and compute schedules with dollar figures attached
- A ranked savings queue your team can work through in order
What does the convertible option really buy?
Roughly five points of discount traded for the right to change instance family inside the term. In an estate that refactors, that is the cheapest insurance available.
The mistake is treating it as a worse version of the standard reservation. It is a different instrument for a different level of certainty, and the estates that mixed both covered volatility better than the ones that picked a side.
The portfolio is the decision, not the product
Stable base load on the cheapest locked instrument, volatile capacity on the flexible one, and the genuinely uncertain remainder left on demand until it settles. The wider coverage question sits in the spend plan pricing guide.
Watch the briefing · 4:10The Discount StackReservations, then spend plans, then the negotiated rate, multiplied rather than added.
How does the programme discount stack on top?
Multiplicatively rather than additively. The negotiated programme rate applies on top of the term discount, worth roughly three to fifteen percent above whatever the commitment already delivered.
The rate is the second conversation
The commitment mechanics themselves are documented in the savings plan documentation, which is worth reading against your own consumption curve rather than a proposal.
That is why the two negotiations belong together. Sizing the commitment first and then negotiating the programme against it produces a better effective rate than arguing the headline percentage alone.
The programme mechanics, and how the commitment feeds the band, sit in the programme negotiation guide, the cost governance integration guide and the AWS negotiation guide.
The head to head between the two vehicles is worked through in the vehicle comparison, and the cross provider position in the multi cloud leverage guide.
What the reviews measured, 2024 to 2025
Two cuts of the engagement file, both about sizing rather than rate.
After a workload refactor that the standard reservation could not follow, leaving the commitment pointed at capacity nobody ran.
Where coverage targets were pushed above 80 percent, past the point at which the break even still holds.
Neither is a discount failure. Both are forecast failures, and a discount applied to a wrong forecast makes the error cheaper per hour and no smaller in total.
Your first five moves
- Size the portfolio against next year's footprint, not last year's, because forecasting the past is what locked the wrong family in about a third of estates.
- Set the coverage target below 80 percent deliberately, since pushing past it left 10 to 20 percent of the commitment idle at the margin.
- Test every commitment against the break even, roughly six months on a one year term and eighteen on three years, before signing rather than after.
- Mix convertible capacity and spend based plans for the volatile layer, which covered change better than reservations alone in most accounts.
- Negotiate the programme discount and the commitment together, because the programme rate stacks on top for a further three to fifteen percent. The AWS practice sizes the curve before the commitment is signed.
Frequently asked questions
What discount do the commitment vehicles deliver?
Both deliver thirty to seventy two percent off on demand. The choice between them sits in flexibility rather than in the discount band itself.
What does convertible capacity cost?
Roughly five points of discount, traded for the right to change instance family inside the term. In an estate that refactors, that is the cheapest insurance available.
Where is the break even?
Around six months of utilization on a one year term and around eighteen months on three years. At a deeper discount the break even falls, which makes the commitment more forgiving.
Should coverage be pushed as high as possible?
No. Targets above 80 percent left idle reservations worth 10 to 20 percent of the commitment, because the tail of the portfolio stops clearing the break even.
What goes wrong most often?
The portfolio is sized to last year's footprint. Standard reservations then locked the wrong instance family in about a third of estates after a workload refactor.
Should you pick one vehicle or mix them?
Mix them. Convertible capacity and spend based plans together covered volatility better than reservations alone in most accounts, because the estate is not uniformly predictable.
Which layer belongs on which instrument?
Stable base load on the cheapest locked instrument, volatile capacity on the flexible one, and genuinely uncertain demand left on demand until it settles into a shape.
How does the programme discount interact?
It stacks on top of the term discount rather than replacing it, worth roughly three to fifteen percent above whatever the commitment already delivered.
Should the two be negotiated together?
Yes. Sizing the commitment first and negotiating the programme against it produces a better effective rate than arguing the headline percentage on its own.
Does a discount fix a bad forecast?
No. It makes the error cheaper per hour and no smaller in total, which is why sizing is the first conversation and the rate is the second.
