Cloud bills have a way of creeping up. A startup launches on a handful of services, the product grows, test environments multiply, and one day the monthly invoice is several times what anyone expected — with no clear idea where the money goes. The good news: most cloud accounts contain significant savings that do not require rebuilding anything. Here are the steps we use to cut AWS costs for clients, roughly in the order they pay off. Most of them apply equally to Google Cloud and Azure.
1. Find out where the money goes
You cannot optimise what you cannot see.
- Open Cost Explorer and group the last three months by service, then by account or region. A handful of line items usually make up most of the bill.
- Tag everything by project, environment (production, staging, development), and owner, so costs can be traced to a team or feature.
- Set up AWS Budgets with alerts at 50%, 80%, and 100% of your monthly budget, and enable cost anomaly detection, so a runaway service is spotted in days, not at the end of the month.
2. Delete what nobody uses
Every account we review has forgotten resources:
- unattached EBS volumes and old snapshots;
- idle load balancers and unused Elastic IP addresses (public IPv4 addresses are now billed by the hour);
- test environments and databases left running after a project ended;
- old logs kept forever — set retention periods on CloudWatch log groups.
This step alone often saves 10–20% of the bill.
3. Right-size your servers and databases
Teams usually start with generous instance sizes "to be safe" and never revisit them. Check CPU and memory usage over a few weeks: an instance that never goes above 20% CPU can usually be one size smaller. AWS Compute Optimizer gives specific recommendations. Do the same for databases, which are often the most over-provisioned resources.
4. Switch to Graviton (ARM) instances
AWS's own ARM-based Graviton processors offer noticeably better price-performance than comparable x86 instances for most web applications, APIs, and databases. Many workloads — Node.js, Python, Java, Go, PostgreSQL, MySQL — move over with little or no code change. Test in staging first, then switch.
5. Stop paying for idle time
- Schedule non-production environments to shut down outside working hours and at weekends. Development servers that run 50 hours a week instead of 168 cost about 70% less.
- Use auto scaling so production capacity follows real demand instead of the peak.
- For spiky or low-traffic workloads, serverless options (Lambda, Fargate, serverless databases) charge only for actual use.
6. Commit to what you will use anyway
Once your baseline is stable, Savings Plans or Reserved Instances cut compute costs substantially in exchange for a one- or three-year commitment. Commit only to the steady base load you are confident about, and keep the variable part on demand. Spot instances offer even deeper discounts for work that can be interrupted, such as batch jobs, CI pipelines, and data processing.
7. Watch the hidden network costs
Data transfer is the line item that surprises teams most:
- NAT gateways charge per hour and per gigabyte processed. Traffic to S3 and DynamoDB can go through free gateway endpoints instead.
- Cross-region and cross-availability-zone traffic costs money; keep chatty services close together.
- Serve static files through CloudFront, which is often cheaper than serving them directly from servers or S3.
8. Tidy up storage
- Use S3 lifecycle rules to move old objects to cheaper storage classes, or Intelligent-Tiering to do it automatically.
- Delete incomplete multipart uploads and old object versions you do not need.
- Switch EBS volumes from gp2 to gp3, which is cheaper and lets you set performance independently of size.
9. Make cost part of engineering
Savings do not last if nobody owns them. Review the bill monthly, show each team its own costs, and include cost in architecture decisions — the same way you include performance and security. Infrastructure as code (Terraform, CloudFormation, or CDK) makes environments reproducible and easy to tear down.
What to expect
For an account that has never been optimised, a combination of these steps commonly reduces the monthly bill by a third or more — sometimes much more — without any change users would notice.
Our cloud & DevOps team runs cost reviews, migrations to Graviton and serverless, and ongoing infrastructure management on AWS, Google Cloud, and Azure. Send us a rough breakdown of your bill and we will tell you where we would start.
