arrow_backAll posts
Cloud & FinOpsJune 3, 2026· 7 min read

How We Cut $1.4M in Annual Cloud Costs Without Touching a Single Application

Most cloud cost problems look like engineering problems. They're not. They're governance problems wearing an engineering mask. Here's how we fixed one — and what we found underneath.

Most cloud cost problems look like engineering problems. They're not. They're governance problems wearing an engineering mask.

The symptoms are always the same: engineers blaming overspending on business requirements, finance blaming engineers for waste they can't explain, and leadership unable to answer the basic question: what does the cloud actually cost us, by product, by team, by business unit?

I ran into this head-on managing a large enterprise AWS account. Cloud spend was growing faster than business outcomes. Stakeholder confidence in cloud ROI was eroding. And nobody had a clear picture of where the money was actually going.

The Problem

When I dug in, the core issue wasn't the infrastructure — it was the operating model around it.

Tagging compliance was below 40%, which meant the majority of spend had no cost center attribution. Finance couldn't chargeback. Engineering couldn't identify who owned what. Business units were receiving rolled-up cloud bills with no ability to connect spend to outcomes.

On the compute side, instances migrated from on-premises were sized for peak load and running at single-digit CPU utilization. GP2 EBS volumes that should have been GP3 were everywhere. Reserved Instance coverage on stable, predictable workloads was under 25% when it should have been pushing 70%.

None of this was malicious. It was the natural result of a migration that optimized for speed and risk reduction over cost efficiency — which was the right call at the time. But nobody had gone back to optimize afterward, because nobody owned "cloud cost" as a business outcome.

The Approach

I didn't start with the infrastructure. I started with ownership.

First: I established a monthly FinOps cadence with a cross-functional working group — finance, engineering, and a business representative from each major unit. Each session had a fixed agenda: attribution review, anomaly investigation, and a savings action item with an owner and a due date. The cadence alone changed behavior. When engineers knew their team's spend would appear in a monthly review, idle resources started getting cleaned up without anyone asking.

Second: I built a tagging governance framework with teeth. We defined the required tags — cost center, product, environment, team — and implemented AWS Config rules to flag non-compliant resources. New resources without required tags triggered an alert. Within 90 days, tagging compliance moved from below 40% to above 85%.

Third: I ran a systematic right-sizing analysis using CloudWatch utilization data across all EC2 instances. Any instance running below 15% average CPU for 30 consecutive days got flagged for review. We migrated every GP2 EBS volume to GP3 — a configuration change that takes minutes and saves 20-30% per GB at no performance cost.

Finally: for stable, predictable workloads, I modeled Compute Savings Plans purchases using 12 months of historical usage data and presented the business case to each Unit CIO. Savings Plans are commitments, not optimizations — getting executives to commit requires showing them the math in business terms, not cloud pricing terms.

The Results

  • $1.4M+ in annualized savings delivered across the portfolio
  • $572K in Compute Savings Plans commitments locked in
  • $875K in EDP savings identified and executed
  • Tagging compliance improved from below 40% to above 85% in 90 days
  • Monthly FinOps cadence adopted as the permanent operating model

The Real Lesson

None of the technical changes were complicated. Right-sizing, GP3 migration, Savings Plans — these are well-documented, low-risk optimizations that anyone with AWS experience can execute. The engineering part took weeks.

What took months was the governance part: getting a cross-functional working group stood up, getting tagging policies enforced at the account level, and getting executives to understand and commit to Reserved pricing on workloads they'd been running at on-demand rates for two years.

The lesson I carry from every FinOps engagement: cloud cost is not an infrastructure problem. It's a decision rights problem. Until someone owns the number — has their name on it, stands up in front of leadership every month and explains it — the infrastructure optimizations are table stakes that will never be executed.

Give someone the number. Everything else follows.

Peter Olson

Peter Olson

Senior technology leader. 20+ years across AWS, Amazon, Fidelity, Wells Fargo & American Express. Building at the intersection of AI strategy and enterprise execution.

Resume
More in Cloud & FinOps