Savings Plans Rightsizing 2026

Savings Plans commit to spend. The rightsizing approach that maximises commitment without overcommitting.

Find your baseline

Savings Plans rightsizing is the discipline of buying enough commitment to capture significant savings without buying so much that the commitment becomes a liability. Over-commit means paying for capacity the team does not use; under-commit means leaving savings on the table. The right level reflects the team's actual usage pattern, not their aspirational growth.

What finding the baseline looks like:

• Trailing 90-day spend on covered usage.: The starting point is the actual usage over the past 90 days. The Savings Plans cover specific usage types (compute, EC2 instance family, etc.); the baseline is the spend on covered usage, not all spend.
• The floor.: The trailing 90-day baseline represents the floor of recent usage. Committing below this floor is generally safe; usage will not drop below the historical baseline without significant operational changes.
• Below this, cost is too low to commit.: If usage routinely drops below a level, the commit at that level becomes wasted. The team pays for capacity that is not used. Avoiding this is the primary discipline.
• Historical seasonality.: Many workloads have seasonal patterns: lower usage on weekends, lower usage during holidays, higher usage during peak business periods. The baseline reflects the lowest-usage portions of the seasonal cycle.
• Commit below the floor of the seasonal pattern.: If the lowest weekly usage is 80 units of compute, the commit should be at or below 80. Committing above the seasonal floor produces idle capacity during the low-usage portions.

The baseline is the foundation. Committing wisely starts with understanding what is actually used.

Commit level

The commit level is below the baseline by some safety margin. The margin protects against unexpected usage drops; the level captures most of the available savings. The right margin depends on workload variability and the team's confidence in usage stability.

• 70 to 80% of baseline.: A common rule of thumb: commit at 70% to 80% of the trailing baseline. The remaining 20% to 30% provides safety margin; usage drops within this margin do not strand commitment.
• Leaves room for short-term reductions.: Workloads sometimes shrink temporarily: a feature is decommissioned, a workload moves to spot, an experiment ends. The safety margin absorbs these reductions without producing wasted commit.
• Avoids over-commit.: Over-commit costs more than no commit at all. The unused commit is paid for at the committed rate; the net cost is higher than on-demand for the same usage. The safety margin is what prevents this.
• Costs more than no commit.: The math: if the commit is for 100 units and only 80 are used, the team pays for 100 at the committed rate. On-demand for 80 would cost less than committed for 100 (depending on the discount level). Over-commit is a real cost, not just lost savings.
• Adjust based on confidence.: Stable workloads can commit closer to baseline; volatile workloads should commit further below. The team's confidence in usage stability determines how aggressive the commit can be.

The commit level is the lever. Conservative commits leave savings on the table; aggressive commits risk over-commit; the right level captures most of the savings safely.

Layer commitments

Single large commitments are riskier than layered shorter ones. Layering produces flexibility: as workloads change, individual layers can mature off and be replaced or not. The total commitment is the sum of layers; the flexibility comes from the layered structure.

• Multiple shorter terms (1-year) instead of one long term (3-year).: Three-year commitments offer the deepest discounts but the longest lock-in. One-year commitments offer shallower discounts but more flexibility. The team chooses the trade-off based on confidence.
• More flexibility.: One-year layers can roll off as workloads change. A workload that exists in year one may not exist in year three; a one-year commit on that workload matures off naturally. A three-year commit on the same workload becomes stranded.
• Small price difference.: The discount difference between 1-year and 3-year commitments is real but typically modest (a few percentage points). The flexibility benefit often outweighs the discount difference for workloads with any uncertainty.
• Stagger purchases.: Buy commitments at different times to stagger their expiration. The portfolio expires gradually rather than all at once. The team's renewal decisions are spread across the year.
• Layer by workload stability.: Highly stable workloads can take 3-year commits; less stable workloads take 1-year. The commitment portfolio matches the workload portfolio's stability profile.

Savings Plan rightsizing is one of the highest-leverage FinOps disciplines available. The savings are substantial; the discipline is in calibrating commitments to actual usage rather than aspirational projections. Nova AI Ops integrates with cloud usage and commitment data, surfaces over-commit and under-commit, and produces the recommendation queue that the FinOps team uses to optimize the commitment portfolio.