Our AWS bill was growing 15% month-over-month with no corresponding increase in traffic. After a two-week deep dive, we reduced monthly spend by 40%. Here's exactly what we did.
Step one was visibility. You can't optimize what you can't measure. We enabled Cost Explorer, set up budget alerts, and tagged every resource by team and environment. This alone revealed that our staging environment was costing 30% of production — running 24/7 with full production-sized instances.
Right-sizing was the biggest win. We analyzed CPU and memory utilization across all EC2 instances and found that 60% were over-provisioned. Switching from m5.xlarge to m5.large for our Django workers saved $2,000/month with no performance impact.
Reserved Instances and Savings Plans provided predictable baseline savings of 30-40% for workloads that run consistently. We committed to 1-year reservations for our database instances and core services.
Architectural changes delivered the remaining savings. Moving static assets to CloudFront, implementing S3 lifecycle policies for old backups, and switching from NAT Gateway to NAT instances for development VPCs all contributed.