38% cloud cost reduction with zero downtime trade-off

The situation

A mid-market SaaS had a six-figure monthly AWS bill that grew faster than revenue. Infrastructure was a mix of long-lived EC2 instances, an over-provisioned EKS cluster, snapshot trails nobody owned, and forgotten data-egress patterns. Finance was applying pressure, but nobody internally had the time to dig in without risking production.

What we did

• Audit first. Two weeks of Cost Explorer, CUR data, and architecture interviews. Output: a written report with a prioritized list of optimizations, each with effort, risk, and dollar impact.
• Quick wins (weeks 3–5). Killed idle resources, attached savings plans where appropriate, fixed unused EBS / EIPs / NAT misuse, and tightened S3 lifecycle policies.
• Right-sizing (weeks 4–8). Resized over-provisioned services using actual usage data. Moved cold workloads to ARM (Graviton) where it was safe.
• Re-platforming (weeks 6–12). Migrated a few standalone EC2 services to Fargate. Replaced one custom Kafka cluster with managed SQS+SNS where the throughput fit.
• IaC adoption. Every change went through Terraform — no more drift, no more "who clicked that?".
• Cost guardrails. Built dashboards tagged by team and per-service. Anomaly alerts on budget overruns.

Outcomes

• Monthly AWS spend down ~38% within 90 days, sustained.
• SLO availability nudged up — fewer flaky long-lived EC2 nodes, more managed services with proper failover.
• Infra is now 100% Terraform-managed and reviewed via PRs.
• Finance gets a per-team cost dashboard every month without manually pulling reports.

Stack

AWS (ECS Fargate, EKS, RDS, S3, SQS, SNS), Terraform, GitHub Actions, Datadog, AWS Compute Optimizer, AWS Cost Explorer.