Right sizing delivered faster savings than any discount negotiation, and committing first simply locks 15 to 30 percent of idle compute into the discount base
In a regulated estate the savings are real, and resilience, residency and audit controls set the boundary on every one of them.
Prepared by Redress Compliance · August 18, 2026 · AWS estates in banking and financial services. 15 to 25 estates reviewed, 2024 to 2025.
Executive summary
15 to 30 percent of compute spend sat on idle or oversized resources. Idle resources cost the same as busy ones, and removing them touches no compliance control.
Steady workloads ran on demand where commitments would have cut 20 to 40 percent. The workloads that must stay in a regulated estate are usually the easiest ones to commit.
In 3 of 4 banks, savings eroded within two quarters without a standing cost process. A single cleanup fades, and tagging with showback is what makes it hold.
Right size first, commit second, govern continuously. Cleaning before committing lowers the spend the commitment has to cover and protects the regulated controls throughout.
Why is right sizing the first saving rather than the last?
Because removing idle and oversized resources cuts spend immediately, with no commitment and no compliance risk. The usage data that makes oversized resources visible sits in AWS cost management documentation.
- Idle compute that no workload needs.
- Oversized instances and databases.
- Unattached volumes and orphaned snapshots.
Removing waste is compliance safe
Genuine waste does not touch resilience or residency, so it sits inside the regulated boundary without raising new risk. That is what makes it the safest saving as well as the fastest one.
How much do commitments actually cut?
Between 20 and 40 percent on predictable workloads, in exchange for a one or three year commitment. Steady banking workloads should rarely run at full on demand price.
| Lever | Saving type | Compliance note |
|---|---|---|
| Right sizing | Immediate, no lock in | Fully safe |
| Savings Plans | Rate cut on commit | Match to steady load |
| Reserved capacity | Rate cut on fixed use | Best for static workloads |
| Portfolio discount program | Discount across the portfolio | Size to real spend |
| Storage tiering | Lifecycle cost cut | Respect retention rules |
The commercial half of the same decision, meaning who you negotiate the rate with and when, sits in the AWS negotiation guide and the commitment flexibility paper.
Apply commitments only to load you are confident will persist. Cover steady compute with Savings Plans, reserve capacity for fixed regulated workloads, and leave variable load on demand until it stabilizes.
The AWS commitment optimisation brief
When reserved capacity beats a Savings Plan, when it does not, and how to size either against a regulated estate.
Get the brief →What 15 to 25 banking AWS estates showed
Across roughly 15 to 25 AWS estates in banking and financial services reviewed between 2024 and 2025, right sizing delivered faster savings than any discount negotiation. Three patterns recur.
- Idle waste: 15 to 30 percent of compute spend sat on idle or oversized resources.
- Under committed: steady workloads ran on demand where Savings Plans would have cut 20 to 40 percent.
- Governance decay: in 3 of 4 banks, savings eroded within two quarters without a standing cost process.
Committing before right sizing locks the waste into the discount base. You end up paying a negotiated rate on resources you never needed.
- 520 vendor benchmarks, from Microsoft EA to Oracle ULA to Salesforce
- Instant percentile standing: market low, median and high for deals like yours
- Renewal uplift exposure modeled over the full term, with the cap to ask for
Where does resilience set the boundary?
Wherever an optimization would reduce redundancy, cross region failover or the data residency a regulator requires. Those controls are not negotiable for cost, and the saving stops at the control line.
- Document the resilience and residency requirements first.
- Mark the workloads where optimization is constrained.
- Optimize freely outside those constraints.
Residency limits the menu, not the discipline
Residency rules can fix a region or a provider, which removes some options. Map them before you move or consolidate workloads, rather than discovering them during a migration.
Watch the briefing · 4:20Run AWS Between RenewalsWhy every optimisation lowers the spend that retires your commitment, and what to watch instead.
How do the savings survive past the first cleanup?
By making cost a continuous owned process rather than a one time exercise. In 3 of 4 banks the savings eroded within two quarters without one, because spend drifts back up without accountability.
Three controls keep the estate clean
- Tag every resource to an owner and a cost center.
- Use showback so business units see their own spend.
- Review commitments and waste every quarter.
Rates move often enough that a standing process pays for itself, since the AWS pricing documentation changes on its own schedule rather than yours.
Showback works because teams that see their own numbers change behavior, which is something a one time audit cannot do. The integration with the commercial side sits in the cost governance and negotiation guide.
What the reviews measured, 2024 to 2025
Two cuts of the engagement file frame the sequence rather than the size.
The waste that a commitment would have covered at a discounted rate if it had been signed first.
How long a cleanup held in 3 of 4 banks where no standing cost process owned it afterwards.
Neither figure argues against commitments. Both argue for doing them second, against an estate that has already been cleaned.
Your first five moves
- Document the resilience and residency requirements before touching anything, and mark the workloads where optimization is constrained by them.
- Remove idle compute, oversized instances and orphaned storage first, which is 15 to 30 percent of compute spend and carries no commitment and no compliance risk.
- Commit only to the load that remains and is confirmed to persist, worth 20 to 40 percent on steady workloads. The Savings Plans guide covers coverage shape.
- Size any portfolio discount program against real spend rather than aspiration, since an oversized commitment is the same mistake at a larger scale. The discount program guide prices it.
- Stand up tagging, showback and a quarterly review, because 3 of 4 banks lost the savings within two quarters without one. The AWS practice runs the sequence in that order deliberately.
Frequently asked questions
What is the first AWS saving in a regulated estate?
Right sizing. Removing idle and oversized resources cuts spend immediately with no commitment and no compliance risk, and it accounted for 15 to 30 percent of compute spend in the estates reviewed.
Why not commit first?
Because committing before right sizing locks the waste into the discount base. You end up paying a negotiated rate on resources you never needed, for the length of the term.
How much do commitments save?
Between 20 and 40 percent on predictable workloads, in exchange for a one or three year commitment. They should be applied only to load you are confident will persist.
What counts as waste?
Any resource running below its useful threshold: idle compute, oversized instances and databases, unattached volumes and orphaned snapshots. None of it touches a compliance control.
What must never be cut?
The redundancy, backup and failover that resilience requirements depend on. Cost savings stop at the regulated control line, and that line is documented before the work starts.
How does residency limit the options?
Residency rules can fix a region or a provider, which removes some consolidation and migration options. Map them before you move workloads rather than during.
Why do savings erode?
Because spend drifts back up without accountability. In 3 of 4 banks the savings eroded within two quarters where no standing cost process owned them afterwards.
Does showback actually change behavior?
Yes, because teams that see their own spend adjust it. That is the thing a one time audit cannot do, however thorough the audit was.
What is the right sequence?
Right size first, commit second, govern continuously. Each step de risks the next and protects the regulated boundary throughout.
Which workloads are easiest to commit?
In a regulated estate, the ones that must stay. Workloads pinned by resilience or residency requirements are the most predictable load on the bill, which is exactly what a commitment wants.