The invoice is not the same as the usage.
Every downstream decision — what to commit to, what to optimise, what to negotiate — rests on knowing what you actually consume. On AWS that is not the number on the invoice, and the gap between them is where most commercial mistakes begin.
You cannot price a decision on a number you cannot decompose
AWS produces one of the most granular billing datasets in enterprise software — hourly, per-resource, per-usage-type, running to millions of line items a month for a large estate. That granularity is a strength for analysis and a problem for reporting: almost nobody consumes the detail, so most organisations work from summaries that have already discarded the distinctions the commercial decisions depend on.
Three different costs, one word
Unblended, blended and amortised cost answer different questions and routinely disagree. A commitment purchase shows as a spike in one view, a flat monthly charge in another, and is spread across consuming accounts in a third. Compare the wrong pair and a team looks efficient in a month it was not.
Tagging is never complete
Cost allocation depends on tags applied at resource creation, retroactively unfixable, and absent on much of what a mature estate accumulated before the tagging policy existed. Untagged spend does not vanish — it lands in a bucket nobody owns.
Shared cost has no natural home
Support charges, data transfer, and shared platform services are consumed by everyone and attributable to no one team without a chosen allocation rule. The rule changes the answer, and it is rarely written down.
The detail exists. Reading it commercially is the hard part.
This is not a data availability problem. The Cost and Usage Report contains what is needed. The difficulty is that answering a commercial question requires joining billing detail to organisational structure and contract terms that live nowhere near it.
- 01
Discounts are applied at the payer level but consumed at the linked-account level, so per-account cost depends on an allocation choice rather than a fact — and the choice is invisible to the account owner.
- 02
Commitment benefit floats to whichever eligible usage maximises the discount, so the same workload can appear to cost different amounts month to month without changing at all.
- 03
Usage types and operation codes describe what AWS billed for, not what the business was doing. Mapping one to the other requires knowledge of the architecture, not just the invoice.
- 04
Marketplace subscriptions, private-offer purchases and support tiers sit on the same bill as consumption but behave entirely differently under a commitment or discount agreement.
- 05
Reserved capacity purchased historically may still be applying benefit to workloads nobody associates with the original purchase decision.
What we read, and what we join it to
The analysis starts from your own billing and usage data — read-only — and joins it to the organisational and contractual facts that determine what any line actually means.
- 01Cost and Usage Report at hourly, resource-level granularity across the full retained history
- 02Unblended, amortised and net-amortised views reconciled against each other
- 03Organisation structure: payer, linked accounts, OU hierarchy and how they map to business units
- 04Cost allocation tag coverage, by spend rather than by resource count
- 05Untagged and unallocable spend, quantified rather than distributed by assumption
- 06Usage-type and operation breakdown by service, region and instance family
- 07Data transfer paths — inter-AZ, inter-region, egress — as a distinct cost class
- 08Storage class distribution and lifecycle-policy coverage
- 09Marketplace and private-offer charges separated from native consumption
- 10Support tier and its billing basis relative to eligible spend
- 11Credits and their expiry profile, kept separate from underlying run-rate
- 12Existing commitment inventory and where its benefit is landing
What the analysis establishes
A defensible run-rate
Consumption stated net of commitment benefit, credits and one-off events — the number a forward commitment should actually be sized against, rather than a headline that flatters or overstates depending on which view produced it.
The commitment-eligible base
Not all spend can be covered by a Savings Plan or Reserved Instance. Separating eligible from ineligible consumption is the precondition for every coverage figure downstream, and getting it wrong makes coverage look better than it is.
Where cost is genuinely attributable
Which spend maps cleanly to a business unit, which requires an allocation rule, and how much is honestly unattributable. Stated plainly, so the shared-cost conversation happens once rather than every quarter.
The structural drivers
The architectural facts producing the bill — data transfer topology, storage class distribution, instance family mix — which is where cost changes come from, as opposed to the line items, which are only where they appear.
The point of decomposing the bill is not the report. It is that every commercial decision after it — commit or wait, optimise or leave, negotiate now or at renewal — becomes answerable with a number you can defend to a board.
- Whether current commitment coverage is sized against real eligible consumption or an inflated base
- Which cost movements are architectural and which are pricing, so remediation goes to the right team
- What a business unit actually costs to run, including the shared cost it genuinely causes
- Which workloads deserve optimisation effort first, measured by recoverable spend rather than by visibility
- What your run-rate will be after in-flight efficiency work lands — the input a commitment decision needs
- Where the estate is exposed to cost changes it has not modelled, such as credit expiry or tier thresholds
What we hand over
The reconciled baseline
A single stated run-rate with its derivation shown — which views were used, what was excluded, and why. The figure a commitment decision or a negotiation can be built on without being re-litigated.
Attribution map
Spend mapped to business structure, with allocation rules made explicit and unallocable spend quantified rather than quietly spread.
Structural findings
The architectural drivers behind the bill, each with the evidence attached, ordered by the spend they control rather than by how easy they are to describe.
Know which number that was
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.