Why the bill stays high after migration
A rehosted estate often carries its old shape with it. Servers sized for peak load on physical hardware run all day and night, development environments never switch off, older storage types remain in place, and network traffic passes through gateways that charge for every gigabyte. Finance receives one large invoice that no individual service owner recognises as theirs.
Make costs visible and owned first
The AWS Well-Architected cost optimisation pillar begins with cloud financial management and awareness of expenditure, before it reaches resource choices. In practice that means separating workloads into accounts, applying consistent cost allocation tags, setting budgets and alerts for each owner, and reviewing costs with the people who can change them. In the public sector, map those owners to budget holders so that forecasts can support spending decisions.
Start with the low-risk changes
AWS Compute Optimizer recommends sizes for resources including EC2 instances, EBS volumes, Lambda functions, RDS and Aurora databases and NAT gateways, based on their configuration and, by default, the last fourteen days of CloudWatch metrics. Check memory use before acting on a recommendation, since it needs additional metrics to be considered.
Moving EBS volumes from gp2 to gp3 is often straightforward, and AWS states that gp3 costs up to 20% less per gigabyte, with a baseline of 3,000 IOPS regardless of volume size. Schedule non-production environments to stop outside working hours, and delete unattached volumes and old snapshots under an agreed lifecycle policy.
Commit after rightsizing
Savings Plans and reserved capacity reduce the price of steady usage in exchange for a commitment. Make that commitment after rightsizing, once the steady baseline is known, and involve finance early, because annual budgets and multi-year commitments need to be reconciled before anyone signs.
Modernise where it changes the cost curve
Some savings need architectural change. Managed databases remove licensing and administration effort, while serverless functions and containers that scale down when idle suit uneven demand. Data transfer deserves particular attention: NAT gateways charge for each gigabyte they process, so traffic to AWS services that could use a VPC endpoint is worth examining. Architectural changes cost money to make, so give each one a business case with an expected payback and a measured result.
Keep it going
Cost control works as a routine. Review each account's costs monthly with its owner, track a unit cost that the service recognises, such as the cost per application processed, and record each optimisation with its expected and actual saving.
Working document: a cost review record
A monthly record for each service keeps cost decisions with the people who can make them.
| Field | What to record |
|---|---|
| Service and budget holder | The service reviewed and the person accountable for its spend. |
| Scope | Accounts and tags included in the review. |
| Monthly cost and trend | Current cost, change from last month and forecast. |
| Unit cost | A measure the service recognises, such as cost per transaction or per user. |
| Main cost drivers | The resources or behaviours responsible for most of the spend. |
| Actions | Each change with its expected saving, risk, owner and date. |
| Commitments | Savings Plans or reservations in place, their coverage and expiry. |
| Next review | Date of the next review and any decisions needed beforehand. |
Further reading
AWS Well-Architected: cost optimisation pillar
AWS Compute Optimizer: what is Compute Optimizer?
Amazon EBS: general purpose volumes
AWS Architecture Blog: data transfer costs for common architectures
