AWS cost optimisation: a pragmatic playbook for growing teams

Most AWS bills carry 20–40% of waste that nobody chose on purpose. It accumulates: a test environment left running, an over-provisioned database, logs retained forever, traffic crossing availability zones for no reason. The good news is that cost optimisation is mostly disciplined housekeeping, not heroic re-architecture. Here is the order we work through.

1. Make the bill visible before you touch anything

You cannot optimise what you cannot see. Turn on Cost Explorer and AWS Budgets, and tag resources by environment and team. The single highest-leverage habit is a weekly five-minute glance at spend grouped by service. Most overspend is obvious once someone is actually looking.

Set a budget alert at 80% of expected monthly spend. It turns a nasty end-of-month surprise into a Tuesday afternoon adjustment.

2. Kill idle and oversized resources

The fastest savings come from things you are paying for but not using:

• Idle environments. Schedule non-production EC2 and RDS instances to stop overnight and at weekends. A dev environment running 168 hours a week when it is used for 40 is paying for 75% nothing.
• Oversized instances. Use Compute Optimizer's right-sizing recommendations. The default instinct is to provision for peak; most workloads run comfortably one size down.
• Unattached resources. Orphaned EBS volumes, old snapshots and idle load balancers bill quietly forever. Sweep them quarterly.

3. Commit to what you actually run

Once your baseline is stable, Savings Plans and Reserved Instances cut 30–60% off steady-state compute in exchange for a one- or three-year commitment. The trick is to commit only to your reliable baseline, and leave spiky or experimental workloads on on-demand pricing.

Rule of thumb:
- Predictable 24/7 baseline  -> Savings Plan / Reserved
- Variable daytime load      -> On-demand or Spot
- Fault-tolerant batch       -> Spot instances (up to 90% cheaper)

4. Fix the architecture-level leaks

Some of the biggest line items are architectural:

• Data transfer. Cross-AZ and cross-region traffic adds up fast. Keep chatty services in the same AZ and cache at the edge with CloudFront.
• NAT Gateway. A surprisingly common top-five cost. Route S3 and DynamoDB traffic through VPC endpoints instead.
• Log and storage retention. Ship logs to cheaper tiers and set retention policies. Move infrequently accessed S3 data to Intelligent-Tiering or Glacier.

5. Make it a habit, not a project

The teams that stay cheap treat cost as a standing metric, not a once-a-year panic. A lightweight FinOps rhythm — tagging discipline, a weekly cost review, right-sizing each quarter — keeps the bill honest as you grow.

A first pass through this list on a mid-sized account typically recovers 25–35% within a few weeks, with no impact on reliability. If your AWS bill has been climbing faster than your traffic, get in touch and we will run the audit with you.