28 AWS Cost Optimisation Tactics We Use for Every Client

The average company using AWS overpays by 30–60%. We know this because we run cost optimisation engagements as a standard phase of every managed IT onboarding, and the findings are remarkably consistent regardless of company size or maturity.

The biggest single lever: Reserved Instances and Savings Plans. Switching from On-Demand to 1-year reserved pricing on predictable workloads — databases, baseline application servers, internal tools — immediately reduces compute costs by 40%. This is table stakes. If your organisation isn’t doing this, stop reading and start here.

Right-sizing is the second lever. AWS defaults to the instance type you provisioned, not the one you need. Running CloudWatch metrics for 30 days and then right-sizing under-utilised instances — particularly RDS and ElasticSearch clusters — typically yields another 15–25% reduction. AWS Compute Optimizer automates the analysis; the work is in reviewing and applying the recommendations.

Data transfer costs are the most underestimated category. Egress between availability zones, between regions, and out to the internet adds up fast. Centralising your VPC architecture, using Gateway Endpoints for S3 and DynamoDB (free), and caching aggressively at CloudFront reduces transfer costs by 30–50% in most architectures we audit.

Storage hygiene matters more than most teams realise. Unattached EBS volumes, forgotten snapshots, S3 buckets filling with logs that nobody reads, RDS automated backups retained for 35 days when 7 would do — these accumulate silently. A quarterly storage audit with automated lifecycle policies applied to S3 and EBS snapshots pays for itself in the first month.

Finally: turn things off. Dev and staging environments running 24/7 at full production spec is one of the most common and avoidable costs we see. Automated start/stop schedules for non-production environments — running 10 hours/day, 5 days/week instead of 168 hours/week — cut those environment costs by 73% overnight.