AWS cost ownership
Understand the bill.
Engineer the improvement.
Make AWS costs attributable, explainable and manageable. We work with engineering, finance and service owners to understand expenditure, prioritise changes and establish a repeatable way to manage running costs.

Costs someone can explain
Cloud bills grow for good reasons and bad ones. A service gains users, a migration runs two estates at once, a log stream nobody reads keeps growing. When nobody can say which service, team or decision a cost belongs to, it becomes hard to forecast, hard to challenge and hard to reduce safely.
We aim for a bill that is attributable, explainable and managed as part of normal delivery, not a once-a-year savings exercise.
When to bring us in
- Spend is rising and nobody can say which services or decisions are driving it.
- Shared platforms, networking, support or data services are paid for centrally and not attributed to the services that use them.
- Recommendations exist, but nobody has the time, context or authority to change the running service safely.
What we take responsibility for
We take responsibility for the analysis, for the engineering changes we agree to make, and for the evidence that they had the intended effect. Budget authority, service priorities and the decision to accept a trade-off stay with your service and finance owners.
Engineering scope
Cost ownership
Allocate spend by service, team and environment, and find what remains unallocated. Agree how shared platforms, networking, support and data services are treated. Cost allocation tags and AWS Cost Categories help, but tags alone do not make a bill attributable; we document the denominator and the exclusions behind every allocation figure.
Forecasts and decisions
Make demand assumptions, seasonality, migration overlap and commitments such as Savings Plans explicit. Compare actual spend with forecast and agree who responds to an anomaly. AWS Budgets and Cost Anomaly Detection provide information; they do not cap spending, so someone must be able to act on them.
Engineering improvements
Assess utilisation, scheduling, idle resources, storage retention, database configuration, log volume, data transfer and architecture. Assess Kubernetes or AI costs where they actually exist. Record the operational constraints and compare the cost of making a change with its expected benefit.
An ongoing practice
Agree a review cadence, an owner and an improvement backlog. Check whether changes were implemented and whether the expected effect appeared. Develop cost-per-unit measures only where activity counts and shared-cost assumptions are meaningful.
An illustrative project
Illustrative example, not a client engagement
Starting state. A service’s AWS spend has grown for a year. Much of it sits in shared accounts, and finance cannot explain the growth to the service owner.
Agreed work. Establish a reproducible baseline for agreed accounts and months, allocate shared costs using rules both finance and engineering accept, separate demand growth from waste, and implement two low-risk changes with a verification window.
Deliverables. A cost baseline with stated exclusions, an allocation model, forecast assumptions, a prioritised change backlog with effort and risk, and a record of what was changed.
Acceptance. The baseline can be reproduced; material costs have an owner or an explicitly unresolved allocation; recommendations do not double count; and each implemented change has a verification method and an observed result.
What we need from you
- A service or finance owner who can make allocation decisions, and an engineer who knows the running service.
- Billing and usage data for an agreed period, through scoped read-only access or protected exports. Never send credentials in an enquiry.
- Context on discounts and commitments, planned demand changes and any service-level constraints.
Costs, risks and trade-offs
We keep three measures separate: the estimated opportunity, the change actually implemented, and the observed financial effect. Recurring changes are kept apart from credits and one-off savings, and overlapping recommendations are never summed.
Spend is shown alongside workload demand, resilience and service level. A fall in cost caused by fewer transactions is not an engineering efficiency gain, and a saving that weakens recovery may not be worth having. We do not promise a percentage saving without evidence from your own estate.
Where cost meets other work
Cost decisions run through the rest of the estate: account structure and shared networking in infrastructure engineering, environment expiry and deployment costs in platform engineering, dual running and retirement in migration, backup retention and exercise costs in reliability, and model, retrieval and review costs in AI engineering.
Read Who is responsible for the part of the AWS bill nobody recognises?, Modernising an AWS estate for cost control and A landing zone needs owners as well as accounts, or see anonymised experience.
A sensible first step
A short first conversation establishes whether a review would help and what scope makes sense. It does not commit either side to a review, and we do not ask for account access or billing data until a scope is agreed.
Further reading
FinOps Foundation: FinOps Framework
FinOps Foundation: Forecasting
FinOps Foundation: Unit economics
Working resource
The cloud cost and ownership worksheet is the blank version of the baseline and backlog a review produces.