How to cut your AWS bill by 20–45% without touching your architecture

Cloud bills grow quietly.
You add a feature, spin up a database for testing, forget to delete it. You launch on On-Demand because you're moving fast. Your S3 bucket fills up and nobody sets lifecycle rules because "we'll do it later.
Two years later you're paying $8,000/month and your CFO is asking questions.
Here's a systematic way to audit your own bill in under an hour.
Step 1 — Compute: find your zombies
Pull a utilization report for the last 30 days. Any instance averaging under 10% CPU and 20% memory is a zombie. Either rightsize it or kill it. This alone typically saves 15–25% of compute spend.
Step 2 — Reservations: stop paying on-demand tax
If you've been running instances for 6+ months and they're stable, you're paying the most expensive rate possible. 1-year Reserved Instances or Compute Savings Plans cut this by 30–45% with zero architecture change.
Step 3 — Storage: tier everything
Data you haven't accessed in 90 days should not be in S3 Standard. Set lifecycle policies to move to Infrequent Access at 30 days, Glacier at 90 days. S3 Intelligent-Tiering does this automatically.
Step 4 — Databases: dev environments
Stop running your staging RDS 24/7. Use Lambda schedules to shut them down at 7pm and restart at 8am. That's 65% of the hours eliminated.
Step 5 — Spot for batch
Any workload that can tolerate interruption — CI runners, ML training, data pipelines — should be on Spot. 60–80% cheaper. AWS rarely reclaims Spot for most instance types.

I built a free tool that walks you through all of this systematically — kloudaudit.eu. 18 checkpoints, takes 15 minutes, no account access needed.
What am I missing? Drop your favorite cost-saving trick in the comments.