Avoid buying cloud commitments from a single busy week
On this page
Commit only to demand you expect to remain eligible and in scope for the full term. Treat each provider recommendation as a historical scenario, then reduce or defer it when the lookback includes a temporary peak, migration, recent purchase, or planned retirement.
Why this is worth a look
Provider recommendations do not remove forecasting risk. AWS bases Savings Plans recommendations on a selected 7-, 30-, or 60-day historical lookback and says they do not forecast usage. Azure analyzes eligible on-demand usage over recent periods, simulates candidate commitments, and warns that recent purchases can take time to affect recommendations. Google Cloud analyzes the previous 30 days using stable-usage and optimal-savings models, with the latter able to consider intermittent usage. Coverage and utilization describe benefit use, but neither proves that the next purchase is safe.
Run this check
CHECKLISTRun this review for each proposed purchase. It is a read-only worksheet for provider recommendation details and internal forecasts, with scope, time period, units, and access recorded before any separate purchase action.
COMMITMENT PURCHASE REVIEW
1. Provider recommendation record
[ ] Record provider, benefit type, term, payment option if shown, scope, sharing setting, currency if shown, and recommendation date.
[ ] Save the recommendation details exposed by the provider. Separately record the lookback period, existing-benefit treatment, coverage estimate, and any utilization estimate actually shown.
[ ] Confirm the provider-specific view or export permissions needed for each input. Use view-only cost and recommendation permissions where supported.
2. Demand evidence
[ ] Obtain the most granular eligible on-demand usage or cost data available for the relevant recommendation period.
[ ] If hourly eligible data is unavailable, use the provider recommendation details plus available billing exports and record the mismatch.
[ ] Record the data unit, scope, currency, and time window. Mark launches, incidents, load tests, month-end work, seasonal events, and temporary projects.
[ ] Separate continuously used baseline demand from burst or infrequent demand.
3. Forward changes
[ ] Ask workload owners about migrations, shutdowns, rightsizing, architecture changes, licensing changes, and region moves during the term.
[ ] Compare the proposed commitment with documented low, expected, and high demand cases.
[ ] Use the low case as the purchase ceiling unless the organization explicitly accepts underuse risk.
4. Benefit interaction
[ ] Review current utilization and coverage separately where the provider reports them.
[ ] Refresh or recalculate after recent purchases, returns, expirations, or sharing changes are reflected.
[ ] Do not add overlapping recommendation sets as if they were independent. In AWS, do not sum Compute and EC2 Instance Savings Plans recommendations.
[ ] In Azure, do not consider a reservation and savings plan together until recommendation systems have updated after a purchase.
5. Approval record
[ ] Record assumptions, excluded peaks, break-even method, and next review date.
[ ] Name the owner responsible for utilization review during the term.
[ ] Obtain finance and workload-owner approval before any separate purchase action.How to confirm it
- 01
Set the exact benefit scope
In each provider's cost management or recommendation interface, select the billing scope and sharing settings that would apply to the purchase. Save the recommendation details the provider exposes, then separately record its term, scope, lookback, existing-benefit treatment, coverage estimate, and any utilization estimate actually shown. Use view-only cost and recommendation permissions where supported, and confirm provider-specific permissions for each input.
- 02
Compare the available history
Compare the provider's available lookback periods rather than choosing the highest recommendation. AWS documents 7-, 30-, and 60-day periods; Azure analyzes 7-, 30-, and 60-day periods, while its portal recommendations use 30 days; Google Cloud analyzes the previous 30 days. Obtain the most granular eligible usage or cost data available, record its units and time window, and label exceptional periods. For Google Cloud, compare stable-usage and optimal-savings recommendations because optimal savings can consider intermittent usage above a financial break-even threshold.
- 03
Apply the workload forecast
Ask workload owners to identify retirements, migrations, rightsizing, licensing changes, and region moves during the term. Remove demand that will become ineligible or disappear. Keep uncertain launches and short peaks outside the committed baseline until their persistence is demonstrated, and document low, expected, and high demand cases.
- 04
Check existing benefit interaction
Review current utilization and coverage separately where the provider reports them. Refresh or recalculate after recent purchases, returns, expirations, or sharing changes are reflected. AWS says Compute and EC2 Instance Savings Plans recommendations use the same usage and are not meant to be taken together. Azure advises waiting at least seven days after buying either a reservation or savings plan before considering the other option.
- 05
Approve a bounded amount
Document the downside if usage falls below the proposed level and identify who will monitor utilization. Approve no more than the durable low-case demand in the eligible scope unless the organization explicitly accepts underuse risk. Keep purchasing as a separate controlled action outside this review.
Before making changes
Assumptions: the proposed benefit has a defined provider, scope, term, and eligible usage measure, and workload owners can provide a forward view for that term. Provider data can be historical, can omit planned changes, and can lag recent purchases or other scope updates. Benefit eligibility and flexibility differ by type. Before comparing savings, confirm which charges, discounts, taxes, licenses, Marketplace items, and capacity-related costs the selected recommendation includes. Review current provider terms before approval.