An AWS cost optimisation checklist is a short list of checks that find the spend nobody chose: resources left running, storage on the wrong tier, prices paid at the full On-Demand rate, and traffic routed the expensive way. The twelve below can be run from your own console in an afternoon, need no code changes to investigate, and each says where to look and what it can save.
The checks are grouped by how quickly they pay back. They build on the patterns in Why Your AWS Bill Is Higher Than It Should Be; this page turns them into something you can work through.
Before you start: two settings that make the rest possible
1. Turn on cost allocation tags. Tags such as owner, environment and application only appear in billing reports once activated in Billing and Cost Management. Without them, every later check ends in “whose is this?”. Our tagging strategy guide has a starter set.
2. Check that Cost Anomaly Detection is on. AWS sets up a service monitor and a daily email summary automatically for new Cost Explorer users, at no additional cost, alerting when spend is over $100 and more than 40% above what was expected. Accounts set up earlier may not have it. Look under Billing and Cost Management → Cost Anomaly Detection, and send alerts to a person, not a shared inbox nobody reads.
Quick wins: waste you can remove this week
| # | Check | Where to look | What it can save |
|---|---|---|---|
| 3 | Unattached EBS volumes | EC2 → Volumes, filter State = available | Their full storage cost: nothing is using them |
| 4 | Old snapshots and unused AMIs | EC2 → Snapshots and AMIs, sort by date | Storage for copies nobody will restore |
| 5 | Idle public IPv4 addresses | VPC → Public IP insights; EC2 → Elastic IPs | $0.005 an hour each, about $3.65 a month |
| 6 | gp2 volumes that could be gp3 | EC2 → Volumes, filter Volume type = gp2 | AWS prices gp3 20% lower than gp2 |
| 7 | Development and test running around the clock | Cost Explorer, grouped by your environment tag | About 70% of those instance hours, if they run only in working hours |
Some notes on these:
- Idle public IPv4. Since 1 February 2024, AWS has charged $0.005 per IP per hour for all public IPv4 addresses, whether attached to a service or not. An Elastic IP kept “just in case” now costs money every hour.
- gp2 to gp3. AWS launched gp3 at a 20% lower price than gp2, with a 3,000 IOPS and 125 MiB/s baseline regardless of size. Elastic Volumes changes the type without interrupting the instance. Check large gp2 volumes first: above 1 TB, gp2’s IOPS grow with size, so match the gp3 IOPS to what the volume actually uses.
- Schedules. A week has 168 hours; a 10-hour working day, five days a week, is 50. Stopping non-production instances outside those hours removes about 70% of their running hours. Storage is still charged while they are stopped.
- Snapshots. Check your records policy before deleting. A snapshot nobody restores may still be the copy your retention rules require.
These are the classic idle and orphaned resources: small individually, and often hundreds of them.
Structural checks: where the larger savings are
| # | Check | Where to look | What it can save |
|---|---|---|---|
| 8 | Oversized instances | AWS Compute Optimizer | The difference between the current and recommended size |
| 9 | Steady compute at On-Demand rates | Cost Explorer → Savings Plans → Recommendations and Coverage | Up to 72% off On-Demand, per AWS |
| 10 | S3 data on the wrong storage class | S3 → Storage Lens; bucket lifecycle rules | Large savings on cold data, after retrieval rules |
| 11 | Data transfer and NAT Gateway charges | Cost Explorer, grouped by usage type | Per-GB charges that a route change can remove |
| 12 | Log groups kept forever | CloudWatch → Log groups, Retention column | Storage for logs past the period you must keep |
- Rightsizing first, commitments second. Compute Optimizer is free and reads your own utilisation data. Commit after you have resized, or you lock in a discount on capacity you don’t need.
- Savings Plans. AWS says Savings Plans offer savings of up to 72% compared with On-Demand, for a one- or three-year commitment. Savings Plans vs Reserved Instances covers which to buy and how much to commit.
- Storage classes. The cheaper classes have minimum storage periods and retrieval fees. S3 Storage Classes Explained shows where they pay off and where they don’t.
- Data transfer. A private workload reaching S3 through a NAT Gateway pays a per-GB processing charge that a gateway endpoint removes, as AWS’s VPC pricing shows. AWS Data Transfer Costs Explained walks through the usual culprits.
- Logs. CloudWatch log groups are kept indefinitely unless a retention period is set. Set one per log group, matched to how long you must keep the logs, and archive anything needed for longer to cheaper storage.
What regulated firms should not rush
Three findings look like waste and often aren’t.
- The recovery environment. A pilot light or warm standby in a second Region exists to be idle most of the time. Its cost is the price of meeting your recovery targets, so review its size against those targets, not against its utilisation. Our post on AWS disaster recovery strategies explains what each pattern needs running.
- Retained data. Backups, snapshots and logs may sit under record-keeping or client contract terms. Move them to cheaper storage rather than deleting them.
- Commitments before a migration or exit decision. A three-year commitment is hard to unwind. If a system may move, be replaced or be part of an exit plan, a one-year term or a smaller commitment keeps your options open.
A worked example
(Illustrative. Rates are assumed; price your own Region in the AWS Pricing Calculator.) A 40-person investment manager spends about $9,000 a month on AWS in London. Running the checklist finds:
| Finding | Monthly cost now | Change | Monthly saving |
|---|---|---|---|
| 8 development and test instances running 24/7 | $1,200 | Run them 50 hours a week | about $840 |
| Steady production compute at On-Demand rates | $3,000 | One-year Compute Savings Plan at an assumed 25% discount | about $750 |
| 1 TB of unattached EBS volumes from an old project | $116 | Snapshot what must be kept, then delete | about $116 |
| 3 TB of gp2 volumes | $348 | Change to gp3 at 20% lower per GB | about $70 |
| 8 idle public IPv4 addresses | $29 | Release them | about $29 |
| Total | about $1,805 a month (20%) |
The order matters. The schedules and clean-up go first because they change what the Savings Plan should cover. The commitment comes last, sized on what is left running.
Turning the checklist into a habit
Most of these checks drift back within a quarter. A tag goes missing, a test environment is left on over the holidays, a new volume is created as gp2 from an old template. A monthly half-hour with Cost Explorer and the anomaly alerts, and a quarterly look at Savings Plans coverage, catches most of it.
If you would like the whole bill reviewed for you, with every saving priced in pounds a month and checked against your recovery and retention needs, see our AWS cost optimisation service.