The Audit That Saved ₹3.1 Lakhs/Month

In 2023, I was brought in to audit a fintech startup's AWS bill. They were spending ₹4.2 lakhs/month and couldn't figure out why. After 2 weeks of analysis and 6 weeks of changes, we brought it to ₹1.1 lakhs/month. Same workload. 74% reduction. Here's exactly what we did.

Finding 1: Zombie Resources (₹60K/month saved)

Unattached EBS volumes, unused Elastic IPs, snapshots from servers deleted 8 months ago, stopped EC2 instances still getting charged for EBS. Run this to find them:

# Find unattached EBS volumes
aws ec2 describe-volumes \
  --filters Name=status,Values=available \
  --query 'Volumes[*].[VolumeId,Size,CreateTime]' \
  --output table

# Find unassociated Elastic IPs
aws ec2 describe-addresses \
  --query 'Addresses[?AssociationId==null].[PublicIp,AllocationId]' \
  --output table

Finding 2: Wrong Instance Types (₹80K/month saved)

They were running m5.xlarge instances averaging 8% CPU utilization. Classic over-provisioning. AWS Compute Optimizer gave us the recommendations. We moved to t3.large instances with burstable performance — same workload, 60% cheaper compute.

• Enable AWS Compute Optimizer (free service)
• Review recommendations weekly for 4 weeks
• Rightsizing rule: if average CPU < 20% for 2 weeks → downsize

Finding 3: No Reserved Instances (₹1.2L/month saved)

They were running every instance on-demand. Reserved Instances for 1-year commitments give 40% savings. 3-year gives 60%. We converted all stable, always-on workloads to 1-year Reserved Instances. This alone saved ₹1.2 lakhs/month.

Rule: Any EC2 instance running 24/7 for more than 3 months should be Reserved. No exceptions.

Finding 4: S3 Storage Classes (₹35K/month saved)

All data — including 2-year-old logs — was sitting in S3 Standard at ₹2.3/GB/month. We set up lifecycle policies to move data automatically:

# S3 Lifecycle Policy (via Terraform)
lifecycle_rule {
  enabled = true
  transition {
    days          = 30
    storage_class = "STANDARD_IA"  # 58% cheaper than Standard
  }
  transition {
    days          = 90
    storage_class = "GLACIER_IR"   # 68% cheaper than Standard
  }
  expiration {
    days = 365  # Delete after 1 year (adjust per compliance needs)
  }
}

Finding 5: NAT Gateway Overuse (₹45K/month saved)

Every byte passing through a NAT Gateway costs money. They had Lambda functions in a VPC making thousands of API calls per hour — all going through NAT. We moved those specific Lambdas outside the VPC and used VPC endpoints for S3/DynamoDB. ₹45K/month gone.