Cutting a SaaS platform's AWS bill by 38%

The challenge

The client's AWS spend had grown organically for three years without a cost owner. Engineering shipped fast, but nobody was accountable for the bill. The result was a familiar pattern: oversized EC2 instances, idle non-production environments running 24/7, unattached EBS volumes, old snapshots, and everything served on-demand with no Savings Plans or Reserved Instances.

Leadership wanted savings, but with one hard constraint: no customer-facing impact. Any change had to be safe, reversible, and invisible to end users.

What we did

We ran a structured FinOps engagement aligned to the AWS Well-Architected Cost Optimization pillar. Rather than a single big change, we sequenced low-risk wins first, then structural savings.

1. Visibility first

We enabled Cost Explorer, AWS Budgets and cost allocation tags across teams, so every dollar mapped to a service and an owner. This alone surfaced ~15% of spend that was untagged or clearly waste.

2. Kill the obvious waste

• Scheduled non-production environments to shut down nights and weekends.
• Removed unattached EBS volumes and stale snapshots.
• Deleted idle load balancers, NAT gateways and orphaned Elastic IPs.

3. Right-size compute and storage

Using Compute Optimizer and real utilization data, we right-sized over-provisioned EC2 and RDS instances, moved eligible workloads to Graviton, and tiered infrequently accessed S3 data to S3 Intelligent-Tiering and Glacier.

4. Commit strategically

With a stable baseline established, we layered in Compute Savings Plans and RDS Reserved Instances sized to the floor of usage, capturing discounts without over-committing.

5. Automate and govern

We codified guardrails: budget alerts, anomaly detection, and tagging policies enforced in CI, so savings would not erode over time.

The bill dropped every month for three months straight, and we finally understand what we're paying for, without asking engineering to slow down.

The results

• 38% reduction in monthly AWS spend within the first full billing cycle after changes.
• Zero downtime and no measurable latency change for customers.
• Ongoing governance so the savings stick instead of drifting back up.