Cloud Infrastructure02 / 05

AWS FinOps Analysis

$700–760/month in optimization identified across a drifted 4-year AWS account

$700–760Monthly savings identified
18%Of monthly bill (P1 tier)
7,753Stale ECR images found
13-weekRemediation plan delivered

Overview

Engaged to run an end-to-end FinOps analysis of a B2B SaaS platform's AWS account β€” averaging roughly $2,830 per month with no cost governance, a 4-year-old drifted CDK codebase, and significant infrastructure sprawl.

The Problem

The account had no cost governance in place. A CDK TypeScript codebase four years old had drifted significantly from live infrastructure: some resources β€” ECS, ECR, CloudFront β€” were still IaC-managed, while others, including RDS, ElastiCache, the VPC, and multiple Elastic IPs, were entirely unmanaged with no audit trail. Leadership's initial expectation was that all optimizations could ship within 48 hours.

The Solution

I ran a full account inventory across more than 10 service categories and a drift analysis between the CDK codebase and live infrastructure. That surfaced the critical anomaly: a ScheduledFargateTask cron set to run every minute, launching up to 9 Fargate tasks per minute, 24/7, across all three environments β€” likely running unnoticed since 2022. I delivered a 5-report suite: an executive inventory, a FinOps assessment with a structured 13-week remediation plan, an engineering discovery guide, a technical CDK analysis, and a supporting addendum β€” resetting the 48-hour assumption into a realistic, phased plan.

The Results

The analysis identified $700–760 per month in actionable optimization, with the highest-priority tier alone accounting for $513–523/month β€” about 18% of the total bill. The full inventory covered 5 EC2 instances, 4 RDS instances, 3 ElastiCache clusters, 9 load balancers reducible to 3, 21 ECR repositories holding 7,753 stale images with zero lifecycle policies, 29 S3 buckets, 3 CloudFront distributions, and 23 Elastic IPs β€” 20 of them unattached, costing roughly $73/month for nothing. Leadership accepted the realistic 13-week plan in place of the original 48-hour expectation.

Key Takeaways

  1. Infrastructure drift compounds silently β€” a 4-year-old IaC codebase that's stopped tracking reality is itself a cost and risk finding, not just a technical debt note

  2. A single misconfigured cron can quietly burn resources 24/7 for years β€” automated scheduling needs the same review rigor as application code

  3. Resetting unrealistic timeline expectations is part of the deliverable β€” a credible multi-week plan beats a rushed promise that can't actually be met

  4. Lifecycle policies are the highest-leverage, lowest-risk fix available β€” cleaning up thousands of unmanaged container images carries zero operational risk and meaningful savings

Tools & Technologies

  • AWS CDK TypeScript
  • ECS Fargate
  • RDS MySQL
  • ElastiCache
  • ECR
  • CloudFront

02 / 05

Let's work together