Before you start
Give the AWS bill an owner. Someone needs the access to see costs, the authority to switch things off, and fifteen minutes a month to look. Then work through the checks below roughly in order: the first four cost nothing and need no commitment, and they make the later ones more accurate.
| Check | Effort | Where to start |
|---|---|---|
| See where the money goes | Low | Cost Explorer, tags and AWS Budgets |
| Delete what nobody uses | Low | Unattached volumes, old snapshots, idle IPs |
| Right-size servers | Medium | Compute Optimizer recommendations |
| Switch off non-production | Low | A start and stop schedule for test servers |
| Commit for steady workloads | Medium | Savings Plans recommendations in Cost Explorer |
| Spot, Graviton, storage, data transfer | Medium to high | One workload at a time |
1. See where the money goes
Start with AWS Cost Explorer and break the bill down by service and by account. Then tag resources by project, environment and owner, so every rupee has a name next to it. Finally, set up AWS Budgets alerts so a sudden jump reaches someone the same day, not at the end of the month.
2. Delete what nobody uses
Every account collects leftovers. Look for:
- storage volumes no longer attached to any server
- old snapshots and machine images from projects that have ended
- load balancers with no traffic
- public IPv4 addresses you are not using, which AWS bills by the hour
- test environments that were never switched off
3. Right-size your servers
Right-sizing means matching each instance to what it actually uses. AWS Compute Optimizer and Cost Explorer both recommend smaller instance types based on utilisation history. Do this before you buy any commitment, or you will lock in a discount on capacity you don’t need.
4. Switch off non-production at night
Development and test servers rarely need to run around the clock. Running them for 12 hours on weekdays instead of all week cuts their compute hours by roughly two-thirds. AWS Instance Scheduler or a simple automation can start and stop them on a timetable.
5. Commit for steady workloads
For servers that run all the time, Savings Plans and Reserved Instances trade a one- or three-year commitment for a lower rate. AWS advertises discounts of up to 72% off On-Demand prices. Real fleets usually land lower: one analysis of a 20-instance production fleet found one-year plans saved 27% to 34%.
- Compute Savings Plans are the most flexible. They apply across instance families, sizes and regions, and to Fargate and Lambda.
- EC2 Instance Savings Plans and Standard Reserved Instances give the largest discounts, but tie you to one instance family in one region.
Commit in stages rather than all at once, and only after right-sizing. Any usage the commitment doesn’t cover is billed at the normal On-Demand rate.
6. Use Spot for work that can be interrupted
Spot Instances use spare AWS capacity at a steep discount, up to 90% off On-Demand, but AWS can reclaim them at short notice. They suit batch jobs, build servers, data processing and anything else that can pause and resume.
7. Try Graviton instances
AWS Graviton instances run on Arm-based processors and often cost less for the same work. Most Linux workloads, containers and managed databases run on them with little or no change. Test one service first.
8. Tidy up storage
Use S3 storage classes and lifecycle rules to move older data to cheaper tiers automatically, such as S3 Intelligent-Tiering or Glacier. Delete snapshots and logs that are past their retention period. Keep the logs you are legally required to: in India, the CERT-In directions require many organisations to keep ICT system logs for 180 days within the country.
9. Watch data transfer and region choice
Moving data between regions, between availability zones and out to the internet all cost money. Keep services that talk to each other in the same region, and use a content delivery network such as CloudFront for files your users download. For users in India, the Mumbai and Hyderabad regions keep latency low and data in the country. Prices differ by region, so compare them before you build.
10. Make it a monthly habit
Cost control is not a one-off project. Fifteen minutes a month stops the bill from creeping back. A simple monthly routine:
- Compare this month’s bill with last month’s, by service, and explain anything that grew.
- Check that every new resource is tagged with a project and an owner.
- Review Compute Optimizer’s latest right-sizing suggestions.
- Look for new idle resources: stopped servers with attached storage, unused IPs, old snapshots.
- Each quarter, check how much of your steady usage is covered by Savings Plans, and top up in small steps.
Mistakes that cost the most
- Buying a three-year commitment before right-sizing, then paying for capacity you don’t use.
- Moving servers to AWS exactly as they were on-premise, sized for peak load, and leaving them running.
- Keeping every snapshot and log forever because nobody set a retention rule.
- Spreading chatty services across regions and paying for the traffic between them.
Frequently asked questions
How much can we save on AWS?
It depends on how the account has been run. Savings Plans and Reserved Instances advertise up to 72% off On-Demand prices, but real fleets usually save less from commitments alone. The quickest savings often come from deleting idle resources and right-sizing.
Should we buy Savings Plans or Reserved Instances?
Compute Savings Plans suit most businesses because they apply across instance families, regions, Fargate and Lambda. EC2 Instance Savings Plans and Standard Reserved Instances give bigger discounts but lock you to one instance family and region.
Is AWS cheaper than running our own servers?
Not automatically. AWS saves money when resources are sized to real demand, switched off when not needed, and committed for steady use. Servers moved across unchanged and left running can cost more than before.
Can you review our AWS account?
Yes. AltF9 reviews usage, right-sizes resources, recommends commitments and sets up monthly cost reports as part of our AWS services.