Efficiency after commitment needs a plan.
Commitments lower eligible usage rates; efficiency reduces required consumption. Model them together so a reduction does not leave unused commitment, while recognising that other eligible demand may absorb the released benefit.
The order you do these in decides whether either one pays
Existing commitment payments continue when consumption falls. Savings may still arise from reduced On-Demand usage or from applying released benefits to other eligible demand. Model the portfolio effect before treating a resource reduction as an equal bill reduction.
Optimise first, then commit
Removing waste before sizing means the commitment is placed against consumption that will persist. The discount then applies to real demand rather than to inefficiency you were about to remove.
Commit first, then optimise
If no other eligible demand absorbs the released benefit, part of the commitment may become unused. Estimate that exposure alongside reductions in uncovered usage.
Why both teams are right
Engineering reports resource reduction accurately. Finance reports flat spend accurately. Neither is wrong, because the commitment sits between them and nobody owns the sequencing decision that connects the two.
Headroom is not a list of idle resources
Optimisation is usually delivered as a findings list — unattached volumes, oversized instances, old snapshots. That is the easy part. The commercially significant question is which of those changes is safe, durable, and worth making before a commitment decision rather than after it.
- 01
Right-sizing recommendations derived from short observation windows miss periodic load, and a resize that breaks a quarterly batch job costs more than it saved.
- 02
Instance family migration — to Graviton, for example — changes both cost and commitment eligibility, so it is a commercial decision as well as an engineering one.
- 03
Storage optimisation has retrieval-cost and access-pattern consequences that a tiering recommendation alone does not price.
- 04
Idle and orphaned resources accumulate faster than they are removed, so a one-off cleanup without an ownership model regenerates the same list within a year.
- 05
Some inefficiency is deliberate: capacity held for resilience, burst headroom, or disaster recovery. Removing it looks like optimisation and is actually risk transfer.
- 06
Optimisation work competes with product work for the same engineering capacity, so recommendations without effort estimates and sequencing do not get executed.
What we examine
We look for durable, safe reductions in the consumption base — and specifically for the ones worth completing before a commitment decision is taken.
- 01Instance utilisation over windows long enough to capture periodic and batch load
- 02Right-sizing candidates with the observation window and confidence stated
- 03Instance family modernisation opportunities, including commitment-eligibility effects
- 04Idle and orphaned resources: unattached volumes, unassociated addresses, stale snapshots, dormant load balancers
- 05Non-production environment scheduling opportunity, measured against actual access patterns
- 06Storage class distribution against real access patterns, priced with retrieval cost included
- 07Lifecycle policy coverage and the spend sitting outside any policy
- 08Data transfer topology — inter-AZ, inter-region, egress — as an architectural cost driver
- 09Managed-service tier and configuration relative to actual load
- 10Capacity deliberately held for resilience, separated from genuine waste
- 11Effort and risk estimate per opportunity, so the list can be sequenced rather than admired
- 12Interaction between each opportunity and the existing commitment position
What we determine
The optimised baseline
What consumption will be once the safe, durable reductions land. This — not today's consumption — is the number a commitment should be sized against, and producing it is the entire point of doing this work first.
The correct sequence
Which optimisations must complete before a commitment decision, which can run alongside it, and which should wait until after — decided by whether each one changes the commitment-eligible base.
Durable versus one-off reduction
Which savings persist and which regenerate without an ownership model. Only the durable portion should inform a multi-year commitment; the rest is a cleanup, not a baseline change.
What is waste and what is insurance
Capacity held deliberately for resilience or burst, separated from genuine waste and named as such — so a cost programme does not quietly remove protection the business chose to buy.
The deliverable is not a list of savings opportunities. It is a sequence: what to fix before you commit, what the base will be when you do, and which reductions are real enough to build a multi-year position on.
- Which optimisation work must complete before the next commitment decision
- What the commitment-eligible base will be after that work lands
- Which opportunities are worth the engineering capacity they will consume
- Which reductions are durable enough to inform a three-year commitment
- Where an ownership model is needed to stop the same waste regenerating
- Which inefficiency is deliberate and should be defended rather than removed
What we hand over
Sequenced opportunity register
Every finding with recoverable spend, effort, risk, durability and its effect on commitment eligibility — ordered as a plan rather than presented as a list.
The optimised baseline
Projected consumption after the recommended work lands, with the derivation shown, ready to feed the commitment decision instead of today's inflated figure.
Sequencing plan
The order of operations across optimisation, commitment and contract events, with the dependencies between them made explicit.
Fix the order
Stage 00 is a 30-minute qualifying call at no cost. If the timing or the estate does not justify an engagement, we say so on that call.