Cloud Cost Optimization: How We Saved a Client $2M Annually

The Problem: $3M Annual AWS Bill, No Visibility

When Meridian Logistics brought us in, they were spending $3M per year on AWS with no clear sense of where the money was going. Their FinOps practice was a spreadsheet updated monthly by one overworked engineer.

By the time we were done, they were spending $960K annually — a 68% reduction — with fully automated cost monitoring and a team that understood every dollar.

Step 1: Build a Cost Map Before Touching Anything

The worst mistake in cloud optimization is optimizing before you understand. We spent the first two weeks doing nothing but mapping:

Which services generate which costs
Which teams own which resources
Which resources are idle, underutilized, or orphaned

Using AWS Cost Explorer and a custom tagging enforcement policy, we found:

23% of compute spend went to resources that hadn't served traffic in 30+ days
41% of storage costs came from uncompressed, non-tiered S3 data that was accessed fewer than 3 times per year
18% of network costs were cross-region data transfer that could be eliminated with topology changes

That's 82% of waste identified before we wrote a single line of infrastructure code.

Step 2: Right-Size Before You Optimize

Everyone wants to use Spot Instances and Reserved Capacity immediately. That's the wrong order.

Right-sizing first:

Analyzed actual CPU and memory utilization across 340 EC2 instances over 90 days
Found 60% were running at under 15% CPU utilization
Downgraded 200+ instances to smaller types, maintaining headroom

This alone saved $420K annually.

Step 3: Storage Tiering — The Overlooked Win

Most teams treat S3 as a single-tier flat storage system. That's leaving significant money on the table.

We implemented a tiering strategy:

S3 Standard — active data accessed frequently
S3 Intelligent-Tiering — data with unpredictable access patterns
S3 Glacier Instant Retrieval — archives accessed occasionally
S3 Glacier Deep Archive — compliance archives, rarely accessed

Automated lifecycle policies move data through tiers based on access patterns. The savings: $380K annually on storage alone.

Step 4: Reserved and Savings Plans

Only after right-sizing and tiering did we touch commitments.

With 12 months of reliable baseline data, we purchased:

1-year Compute Savings Plans for predictable baseline load
Spot Instances with intelligent interruption handling for batch workloads
Reserved Instances for databases that run 24/7

Combined savings: $540K annually.

Step 5: Architecture Changes for Long-Term Efficiency

Some optimizations require architectural changes, not just configuration tweaks:

Moved scheduled batch jobs from ECS to AWS Batch — 40% cheaper for workloads that don't need constant containers
Replaced polling with EventBridge — eliminated expensive Lambda invocations that ran every minute to check for work that wasn't there
Moved static assets to CloudFront — reduced origin server load and network egress significantly

The Lasting Change: FinOps Culture

The technical changes saved $2M. But the cultural change is what makes the savings permanent.

Every service now has:

A cost allocation tag linking it to a team and a product
A monthly budget alert at 80% of expected spend
A dashboard visible to every engineer on that team

When engineers see the dollar impact of their architectural decisions in real time, they make better decisions. That's not a technology problem — it's a visibility problem. And it's the most important infrastructure you can build.