LEARN. PRACTICE. GROW.Choose scope after designing demand
A new reconciliation platform has steady demand, closing peaks, and a planned move from EC2 to Fargate. Before buying a commitment, separate recurring usage from temporary coexistence hours and peaks. Compute Savings Plans allow flexibility across EC2 families and Regions and eligible Fargate and Lambda usage. EC2 Instance Savings Plans restrict scope to the chosen family and Region while allowing size flexibility within that scope. Current documentation also includes Database Savings Plans; do not generalize Compute rules to every product sharing the name. Estimate eligible demand after rightsizing and retirement. The financial owner should understand the term, underutilization exposure, and migration assumptions. An advertised maximum discount is not the guaranteed saving of the specific proposal.
Calculate per hour and distinguish denominators
Commitment unused in one hour does not carry into the next. In an original model, the purchase is 8 currency units per hour and eligible usage, already valued at plan rates, is 5 in one hour and 11 in the next. Used commitment is 5 + 8 = 13 out of 16 purchased units: 81.25% utilization. The three unused units from the first hour do not cover the second-hour excess. That excess must be revalued at On-Demand rates to calculate the bill, which this model does not estimate. Coverage asks a different question: how much eligible usage, valued at its On-Demand equivalent, was covered? If 120 out of a comparable total of 200 was covered, coverage is 60%. Record period, filters, and valuation basis; mixing these denominators produces meaningless indicators.
Separate discount, capacity, and interruption tolerance
A Savings Plan lowers the price of eligible consumption but does not reserve launch capacity. For a critical window, assess Capacity Reservations with matching instance type, platform, zone, and tenancy, together with quotas and actual launch behavior. An immediate-use reservation differs from a future-dated reservation with its own commitment; do not assume identical cancellation costs. Reserving capacity does not itself provide a discount: the two decisions can be combined subject to eligibility. For interruptible calculation workers, Spot can be an alternative if durable checkpoints, safe replay, and the completion deadline are demonstrated. Interruption notices are best effort; hibernation starts without two minutes of advance warning. The design should recover even if the notice never arrives, rather than depending on it to preserve the only copy of work state.
Present a financial decision that can be checked
The proposal should show the commitment, expected hourly eligible usage, and a scenario where an application leaves scope. Applicable Reserved Instances are considered before Savings Plans; do not count two discounts on the same consumption. Among compute plans, the more specific EC2 Instance scope is applied before Compute. Sharing across accounts also depends on configuration and starts with the owner account’s usage. For the fictional project, compare conservative demand, growth, and early retirement, identifying who can change each assumption. Keep a review after the first closing cycle and record deviations. High utilization can coexist with low coverage; this does not establish a billing error or require another purchase. The decision depends on future stability, risk, and total cost, including uncovered consumption.
commitment_per_hour = 8
usage_at_plan_rates = [5, 11]
used_commitment = sum(min(commitment_per_hour, u) for u in usage_at_plan_rates)
purchased_commitment = commitment_per_hour * len(usage_at_plan_rates)
utilization_pct = 100 * used_commitment / purchased_commitment # 81.25
unused_commitment = purchased_commitment - used_commitment # 3
coverage_pct = 100 * 120 / (120 + 80) # 60; On-Demand equivalent values
# Fictional currency units; this is not an AWS invoice calculation.On a fictional funds platform, two environments coexist for six weeks. The team excludes this temporary peak from a new commitment baseline and validates cutover capacity in a separate decision.
Common pitfalls
Using a monthly average as an hourly commitment; stacking discounts on the same usage; confusing coverage with utilization; assuming a discount guarantees capacity.
Related topics: FinOps and evidence for optimization
Choose the commitment against eligible future demand and test capacity, continuity, and financial exposure separately.
Reference: Savings Plans types · SAP-C02