A multi-region fintech platform came to me with a familiar problem: the AWS bill kept climbing, margins were shrinking, and nobody on the team could say with confidence why. Spend was sitting at roughly $20-25K a month.
- ~58% lower monthly AWS spend overall
- 75% lower Redis costs
- 64% lower RDS costs
- 86% lower S3 spend
- No performance degradation
The work ran over three months in four phases. The fixes weren't discount programs or a one-off cleanup that drifts back within a quarter. They were structural changes across Redis, RDS, S3, Glue, ALB, and ECS. Here's the way I approach this kind of work.
1. Get visibility before touching anything
Most "cloud bills are out of control" problems are really visibility problems. Before changing a single resource, break spend down by service, environment, and workload, so every big line item has an owner and a reason to exist. The cost work then becomes a ranked list instead of guesswork.
2. Fix the architecture, not the invoice
Reserved capacity and savings plans have their place, but they lock in whatever you're paying for today, including the waste. The bigger wins come from asking whether each component is shaped right for the load it actually carries. The usual places to look:
- Caches (Redis): clusters sized for a peak that never comes, or holding data that doesn't need to be in memory.
- Databases (RDS): instance classes and storage chosen at launch and never revisited, plus replicas and environments nobody uses.
- Object storage (S3): data kept forever in the most expensive tier, with no lifecycle rules.
- Data jobs (Glue): jobs allocated far more capacity than they need, or running more often than the business requires.
- Load balancers and containers (ALB, ECS): sprawl from services that could share infrastructure, and tasks sized by guesswork instead of measurement.
3. Prove performance doesn't regress
Cost cuts that hurt latency or reliability aren't savings, they're deferred incidents. Every change should ship with before-and-after metrics for the things users feel, and a way to roll back. In this engagement the bill came down by more than half with no performance degradation.
4. Make it stick
The last step is keeping the savings: infrastructure as code (Terraform here) so changes are reviewed, clear ownership of each cost center, and alerts when spend moves. A one-time cleanup decays. A structural fix holds.
Is this worth it for you?
If your bill is in the tens of thousands per month and nobody can explain the biggest line items, there's almost certainly structural waste. I start with a short preliminary scan, and implementation is priced as a share of the savings, so the work pays for itself.