How We Reduced Cloud Costs by 35% in 60 Days With Finoptic
A real customer story: where the waste was, what we fixed first, and how we made the savings last after we left.
A growing SaaS customer came to us. They were spending about $90K per month on AWS. Their costs were growing 40% every quarter. They had no clear view of where the money was going. The CFO had asked Engineering to reduce cloud costs three quarters in a row. Nothing had worked.
Sixty days later, their monthly spend was $58K - a 35% reduction. There were no production incidents and no team layoffs. Here is what was wrong and how we fixed it.
Day 1: Connect Finoptic, find the obvious waste
Finoptic is our FinOps platform. The first thing it does on a new account is build a cost map across services, accounts, and tags. In the first 24 hours it found these problems:
- $8K/month in EBS volumes still attached to instances that were terminated more than 90 days ago.
- $4K/month in NAT Gateway egress from an internal service that should have used a VPC endpoint instead.
- A development EKS cluster running 24/7, but only used during business hours - $2.5K/month wasted on nights and weekends.
Week 2: Right-size the important workloads
After removing waste, we moved to right-sizing. Finoptic compares CloudWatch metrics with billing data. This shows which instances are too large without any guesswork.
The biggest finding: a group of 24 m5.4xlarge instances running the API layer. They peaked at only 18% CPU during the busiest hour of the week. We moved them to m6a.2xlarge (smaller, newer, cheaper). The p99 latency stayed the same. Savings: $11K/month.
Week 4: Reserved capacity and Savings Plans
After right-sizing the workloads, we added commitment-based discounts. We always wait until after right-sizing. If you buy Savings Plans before right-sizing, you lock yourself into paying for oversized resources for a full year.
For this customer, a 1-year Compute Savings Plan covering 70% of the right-sized baseline was the best choice. They did not commit on the remaining 30% that has unpredictable spikes - that stays on-demand for flexibility.
Week 6-8: Make the savings last
The risky part of a 60-day project is that costs can grow back. Six months later, someone adds a new service without tags, and the savings disappear.
We left Finoptic running on the account with three protections: a tagging policy enforced with Service Control Policies, a per-team budget alert connected to Slack, and a weekly cost report sent to Engineering leadership. When monthly spend rises more than 8% above the previous average, someone is notified the next morning - not at the end of the month.
“The technical work was not the hard part. The hard part was building the habits to stop the team from undoing it.”
What we did NOT do
Some things customers ask about that we chose not to do on this project:
- No Spot Instances for production workloads. The savings would have been about $3K, and the added complexity was not worth it for this team.
- No multi-cloud strategy. The team is small. Splitting workloads between AWS and GCP would have created more cost than it saved.
- No database re-architecture. RDS was 18% of the bill and the database is healthy. Changing it would have taken 6 months for a small savings.
Cost optimization is not exciting work. It is mostly removing unused things, choosing the right sizes, and setting up alerts so the savings stay. Want to see what your bill could look like? We do free assessments. Most customers see 30-40% savings in the first 60 days.