A discount you cannot exit is a liability.
Savings Plans and Reserved Instances are usually discussed as savings instruments. They are more accurately fixed-term purchase obligations that happen to carry a discount — and the difference matters most in exactly the situations nobody models in advance.
You are pricing a bet on consumption you have not had yet
Compute Savings Plans and new EC2 Reserved Instances typically use one- or three-year terms, exchanging a payment commitment for lower eligible usage rates. That trade is sound when the forecast holds. When it does not, the commitment does not adjust: an over-sized commitment bills in full against consumption that never arrived, and an under-sized one leaves the remainder at on-demand rates. Both failures are decided at purchase and discovered much later.
The instruments are not interchangeable
Compute Savings Plans span eligible EC2, Fargate and Lambda usage. EC2 Instance Savings Plans are tied to a family and Region. Standard and Convertible RIs have different flexibility rules. Compare actual rates and workload requirements instead of assuming a universal discount ranking.
Term and payment are separate levers
One-year versus three-year changes the discount materially, and so does no-upfront versus partial versus all-upfront. All-upfront improves the rate but converts the commitment into cash out today — a treasury decision that is usually made by an engineer.
Exit rights differ sharply by instrument
Standard RIs can be sold on the Reserved Instance Marketplace under conditions. Convertible RIs can be exchanged rather than sold. Savings Plans have no marketplace and no exchange. The instrument that discounts hardest is the one you are most stuck with.
The decision needs a forward view the purchase process rarely has
Commitment tooling is built around the past. It recommends against trailing consumption, which is exactly the input that is wrong whenever the business is about to change — and the business is usually about to change.
- 01
Native and third-party recommendations extrapolate a trailing window. Anything the business already knows about — a migration, a launch, a decommission, a renegotiated headcount plan — is invisible to that method and material to the answer.
- 02
Efficiency work in flight will reduce the eligible base the recommendation is sized against, so acting on the recommendation locks in the inefficiency you were about to remove.
- 03
Architectural direction changes eligibility itself: moving to Graviton, to serverless, or to managed services shifts which spend a given commitment type can even cover.
- 04
Commitments stack and interact. New purchases layer on expiring ones, and the benefit application order determines what the marginal purchase is actually worth.
- 05
Commitment strategy and discount-programme negotiation are usually run by different people on different calendars, so each is optimised while the combination is not.
- 06
A long commitment needs ownership of its assumptions and periodic review as workloads and plans change.
What we examine before any commitment recommendation
Sizing a commitment responsibly requires the current portfolio, the eligible base after planned change, and the business events that will move both.
- 01Full inventory of existing Savings Plans and Reserved Instances with term, payment option and expiry date
- 02Expiry laddering — how much commitment rolls off, and when, across the next 36 months
- 03Realised utilisation of each existing commitment, not the portfolio average
- 04Coverage of eligible spend by instrument type, measured in dollars rather than hours
- 05Eligible versus ineligible consumption, established from the underlying usage data
- 06Instance family, size and region distribution, and its stability over time
- 07Graviton and modernisation migration status and roadmap, with its effect on eligibility
- 08Planned architectural change: containerisation, serverless adoption, managed-service moves
- 09Business events with cloud consequences — migrations, product launches, decommissions, M&A
- 10Seasonality and its amplitude, distinguished from underlying growth
- 11Interaction between commitment coverage and any private pricing or discount agreement
- 12Treasury position and cost of capital, which price the upfront payment options
What we determine
The defensible commitment floor
The level of consumption you would still have in a downside scenario — the portion that can be committed with confidence at the hardest available discount, because it survives the plans not landing.
The layer that must stay flexible
Consumption that is real but uncertain, which belongs in flexible instruments or stays on demand. Paying a lower discount for the right to be wrong is frequently the better commercial trade, and it is rarely the recommended one.
The instrument and term mix
Which portion belongs in Compute Savings Plans, which in instance-locked instruments, at what term, on what payment structure — with the exit position of each stated up front rather than discovered later.
The expiry ladder
How commitments should be staggered so a single renewal date never forces a large decision under time pressure — which is when the worst commitment decisions are made.
The output is not a recommended purchase. It is a defensible position on how much of your consumption should be locked, for how long, in which instrument — with the downside of each choice quantified before you sign rather than discovered when the forecast misses.
- How much to commit, at which term, and what the downside costs if the forecast is wrong
- Which instrument each portion belongs in, weighed on flexibility rather than headline discount
- Whether to buy now or wait until planned efficiency work has changed the eligible base
- Whether upfront payment is worth the rate improvement at your actual cost of capital
- How to stagger expiries so renewals never all land in one quarter
- Which existing commitments are recoverable through exchange or marketplace sale, and which are not
What we hand over
Commitment position paper
The recommended floor, flexible layer and instrument mix, each with the reasoning and the downside case attached — written to be read by a CFO, not a console.
Scenario model
What the portfolio costs if consumption grows, holds, or falls, so the decision is made against a range rather than a single forecast that will not hold.
Expiry ladder and calendar
Every existing and proposed commitment mapped to its expiry, with the decision date that precedes each one flagged in advance.
Model the downside first
Stage 00 is a 30-minute qualifying call at no cost. If the timing or the estate does not justify an engagement, we say so on that call.