Last quarter, we completed a cloud cost optimization engagement for a Series B SaaS company spending $180,000/month on AWS. After a 6-week engagement, their monthly bill dropped to $68,000 — a 62% reduction — without any degradation in performance or reliability.
Here’s exactly what we did.
The Starting Point
The client had grown from a startup to 50,000+ daily active users over 4 years. Their infrastructure had been built iteratively, optimizing for speed of delivery rather than cost efficiency. Nobody had ever done a systematic cost review.
Monthly spend breakdown:
- EC2 instances: $95,000
- RDS databases: $38,000
- Data transfer: $22,000
- S3 storage: $12,000
- Other services: $13,000
What We Found
1. Massively Over-Provisioned Compute ($45K savings)
Their EC2 instances were running at an average of 12% CPU utilization. They’d provisioned for theoretical peak load that never materialized.
Actions taken:
- Right-sized 45 instances using AWS Compute Optimizer recommendations
- Converted 60% of workloads to Spot Instances with graceful interruption handling
- Purchased 1-year Reserved Instances for baseline steady-state workloads
2. Zombie Resources ($8K savings)
A systematic audit of every AWS resource revealed:
- 23 EBS volumes not attached to any instance (orphaned from terminated instances)
- 7 EC2 instances that hadn’t received traffic in 90+ days
- 12 Elastic IP addresses not associated with anything
- 3 load balancers routing to zero targets
- 15 old RDS snapshots past retention policy
3. Database Over-Provisioning ($18K savings)
Their RDS PostgreSQL instances were db.r6g.4xlarge running at 8% memory utilization and 5% CPU.
Actions taken:
- Downgraded primary instances to
db.r6g.xlarge(with monitoring to validate) - Enabled Aurora Serverless v2 for development and staging environments
- Implemented connection pooling with PgBouncer (reducing connection overhead)
- Enabled storage autoscaling (they’d pre-allocated 10TB, using 1.2TB)
4. Data Transfer Costs ($14K savings)
Data transfer is insidious — it’s easy to miss in cost reports.
Key finding: Their application was reading large datasets from S3 in us-east-1 but their processing compute was in us-west-2. Cross-region data transfer at scale is expensive.
Actions taken:
- Co-located data processing with data storage (migrated to single region)
- Implemented CloudFront CDN for static assets (reducing origin data transfer)
- Enabled S3 Transfer Acceleration only where needed
- Compressed data before transfer (reduced volume by ~60%)
5. Storage Optimization ($6K savings)
- Implemented S3 Intelligent Tiering for objects not accessed in 30+ days
- Deleted 8TB of old log files past retention policy
- Compressed backup archives
The ROI Calculation
| Category | Monthly Savings | One-time Work |
|---|---|---|
| Right-sizing | $45,000 | 40 hours |
| Zombie cleanup | $8,000 | 8 hours |
| Database optimization | $18,000 | 24 hours |
| Data transfer | $14,000 | 32 hours |
| Storage | $6,000 | 8 hours |
| Total | $91,000/month | 112 hours |
The engagement paid for itself in the first month.
The Ongoing Framework
We left the client with:
- AWS Cost Anomaly Detection — Alerts on unusual spend spikes
- Weekly FinOps review — 30-minute ritual to review cost trends
- Tagging policy — Every resource tagged with team, environment, and project
- Rightsizing schedule — Quarterly review of compute utilization
- Reserved Instance plan — Calendar-based renewal schedule
Want to know how much you could save on cloud infrastructure? We offer a free 2-hour cloud cost assessment. Book a session.