The commitment purchase playbook: buying discounts without regret

Quick answer

Follow five rules. 1) Never buy before cleanup and right-sizing, or you lock in waste for the full term. 2) Size against the minimum hourly spend over the last 90 days, not the average, so you commit only to the floor. 3) Target 70 to 80 percent coverage of steady compute and leave the variable top on demand or Spot. 4) Ladder purchases in monthly or quarterly tranches so renewals never all land at once. 5) Prefer flexible instruments over family specific ones unless the extra few points of discount outweigh the risk of a migration stranding coverage. Expected saving is 25 to 35 percent on covered spend for 1 year terms and 45 to 60 percent for 3 year terms.

Every other cost lever is reversible. Delete the wrong thing and you restore it. Right-size too far and you resize back. A commitment is different: you have signed for a dollar amount per hour for one or three years, and if your usage drops, you pay anyway. That asymmetry is why commitments come last in every sensible reduction sequence and why sizing deserves more care than any other decision in a cost program.

Rule 1: sequence, always

Buy after deletion and right-sizing, never before. Consider an account spending $100,000 a month on compute where 20 percent is idle or over-provisioned. Buying a three year commitment at 60 percent discount on the full $100,000 looks like $60,000 of annual saving but locks in $20,000 a month of waste for 36 months. Clean up first, then commit against $80,000, and you save the discount plus the waste. The sequencing is worth more than the discount rate.

Rule 2: size against the floor, not the average

Metric over 90 days	What to do with it
Minimum hourly spend	Safe commitment baseline
p25 hourly spend	Upper bound for a conservative buy
Average hourly spend	Do not commit here
Peak hourly spend	Leave entirely on demand or Spot

Plot hourly on demand compute spend for 90 days. The minimum across that window is the amount you are certain to consume in any hour, including quiet weekends and holiday periods. Committing to the average guarantees unused commitment during every trough, and unused commitment is money burned with no offsetting benefit. A common conservative approach is to commit between the minimum and the p25 value.

Rule 3: coverage target of 70 to 80 percent

One hundred percent coverage is the wrong goal. It means any workload you decommission, migrate, or move to Spot leaves you paying for capacity you no longer use. Seventy to eighty percent of steady state compute is the standard target, leaving the volatile top slice on demand where it costs more per hour but can vanish for free.

Track two metrics continuously, because they fail in opposite directions. Coverage is the share of eligible spend that a commitment discounts, and low coverage means you are leaving discount unclaimed. Utilisation is the share of your commitment that is actually consumed, and anything under 100 percent means you are paying for hours you did not use. The distinction and the monitoring cadence are in coverage versus utilization.

Rule 4: ladder the purchases

Buying your entire commitment in one transaction creates a cliff: in 36 months every commitment expires on the same day, at whatever prices and whatever usage level exists then. Instead buy in tranches, for example a quarter of the target each quarter. Laddering smooths renewal risk, lets you correct sizing as you learn, and means a business change never leaves you fully exposed or fully committed. The mechanics are in commitment laddering.

Rule 5: pay for flexibility unless the discount gap is large

Every cloud offers a flexible instrument and a specific one. Flexible compute commitments apply across instance families, sizes, regions, and often architectures, at a slightly lower discount. Family or resource specific commitments give a few more points but pin you to a shape.

The rule: if there is any chance of migrating architecture, changing instance family, moving region, or shifting from virtual machines to containers or serverless within the term, buy flexible. The extra three to seven points of discount on a specific instrument is not worth a stranded commitment. Only buy specific for genuinely static workloads, such as a database that will not move. Comparison in reserved instances versus savings plans.

Term and payment option

One year terms typically discount 25 to 35 percent and three year terms 45 to 60 percent, with all upfront payment adding a few points over no upfront. The decision is about confidence, not arithmetic: a three year term on a workload whose architecture you expect to change is a bad trade even at a better rate. A practical split is three year terms on the genuinely permanent base layer, one year on the merely stable layer, and nothing on the rest.

All upfront payment improves the rate but consumes cash. For a company where cash is scarce, no upfront at a slightly lower discount is usually the right call, and the difference is typically a few percentage points rather than a transformation.

The monthly review

Commitments need ongoing management, not a one time purchase. Each month check utilisation (anything under 98 percent needs explaining), check coverage against the 70 to 80 percent target, and check the expiry calendar for the next 90 days. Before any large migration, check whether it strands coverage.

The forward looking half of this is knowing what new infrastructure will cost before it lands, so your baseline projection is accurate rather than retrospective. Price planned Terraform changes against the resource catalog so commitment sizing reflects where the estate is going, not only where it has been.

FAQ

When should I buy cloud commitments?

After deletion and right-sizing, never before. On an account spending $100,000 a month on compute where 20 percent is idle or over-provisioned, buying a three year commitment on the full amount locks in $20,000 a month of waste for 36 months. Clean up first, commit against $80,000, and you capture both the discount and the waste reduction. Sequencing is worth more than the discount rate.

How much commitment should I buy?

Size against the minimum hourly spend over the last 90 days, not the average. The minimum is what you are certain to consume in any hour including quiet weekends and holidays. Committing at the average guarantees unused commitment during every trough. A conservative approach commits somewhere between the 90 day minimum and the p25 hourly value.

What is the right commitment coverage target?

70 to 80 percent of steady state compute. One hundred percent coverage means any workload you decommission, migrate, or move to Spot leaves you paying for capacity you no longer use. Leaving the volatile top slice on demand costs more per hour but can disappear for free, which is exactly the flexibility you want on the unpredictable portion.

What is the difference between coverage and utilization?

Coverage is the share of eligible spend that a commitment discounts, so low coverage means unclaimed discount. Utilization is the share of your commitment actually consumed, so anything under 100 percent means paying for hours you did not use. They fail in opposite directions, which is why both need tracking: high coverage with low utilization means you over-bought.

Should I buy flexible or instance specific commitments?

Flexible, unless the workload is genuinely static. Family or resource specific instruments give three to seven more points of discount but pin you to a shape. If there is any chance of changing instance family, migrating architecture, moving region, or shifting from virtual machines to containers within the term, those extra points are not worth a stranded commitment.

Should I ladder commitment purchases?

Yes. Buying the entire target in one transaction creates a cliff where every commitment expires on the same day at whatever prices and usage exist then. Buying in tranches, for example a quarter of the target each quarter, smooths renewal risk, lets you correct sizing as you learn, and means a business change never leaves you either fully exposed or fully committed.

What to do next

Size commitments against where the estate is going. C3X prices planned Terraform against a live resource catalog. Start with the quickstart.