How We Cut Cloud Costs 42% for a Series B SaaS
A practical breakdown of our FinOps process: rightsizing, reserved instances, spot fleets, and the architectural changes that compounded the savings.
Cloud bills are like slow leaks — you don’t notice until you’re bailing water.
When a Series B SaaS company came to us spending $94k/month on AWS, they assumed it was just the cost of scale. Three months later, their bill was $54k — same traffic, same SLA, fewer headaches.
Here’s exactly what we did.
1. Audit first, cut second
The instinct is to turn things off. That’s wrong. We spent the first two weeks mapping every dollar: which service, which team, which product feature.
Tools we used:
- AWS Cost Explorer + Cost Allocation Tags
- CloudHealth for multi-account visibility
- Custom tagging enforcement via AWS Config rules
We found that 31% of spend came from 3 underutilised RDS instances running at <8% CPU — provisioned for a load test that happened 18 months ago.
2. Rightsizing compute
Rightsizing is obvious in theory; painful in practice because engineers are (rightfully) protective of their services.
We created a monthly rightsizing report with three tiers:
- Safe to resize — 60+ days of <30% utilisation
- Investigate — spiky traffic patterns, needs load testing
- Leave alone — stateful services, databases with burst requirements
Only the “safe” tier got resized without a load test.
3. Reserved Instances + Savings Plans
Once we had a stable baseline after rightsizing, we committed. We used Compute Savings Plans (not instance-type locked) for flexibility, buying 1-year terms which hit the sweet spot of ~40% discount vs. on-demand.
4. Spot Fleets for batch workloads
Their ML inference pipeline ran on On-Demand EC2. We moved it to a Spot fleet with fallback to On-Demand for the final 10% of capacity. Savings: ~68% on that workload.
The key is designing for interruption: checkpoints every 60 seconds, jobs that can restart mid-flight.
Results
| Metric | Before | After |
|---|---|---|
| Monthly AWS spend | $94,000 | $54,200 |
| Savings | — | $39,800/mo |
| Annualised | — | $477,600 |
| Effort | — | 3 months |
Cloud optimisation isn’t a one-time project — it’s a culture. We left them with dashboards, budgets with alerts, and a quarterly review process their team now runs without us.
Want a similar review for your infrastructure? Get in touch.