Cost Optimization
The ongoing practice of finding and killing waste without degrading performance.
Why
Cloud spend creeps 5-15% per quarter even without feature growth — idle resources, over-provisioned instances, forgotten dev environments. Cost optimization is the monthly discipline of finding this waste and removing it, which is the highest-ROI engineering work there is: killing $2k/mo of waste takes an afternoon and pays $24k/year forever. Teams that don't do it end up at 3x their necessary bill, at which point finance mandates a painful freeze that delays features. Regular optimization prevents the crisis.
When
Monthly, 2-4 hours per review. Skip only if monthly cloud spend < $500 (the waste is below the effort to find it).
How
1. Pull the cost explorer, group by service, sort descending 2. For the top 5 services, look for: - Idle resources (EC2 < 5% CPU for 30 days, unattached EBS) - Over-provisioned (RDS with 2x the CPU it uses) - Forgotten environments (dev/staging running 24/7) 3. Kill or downsize; tag each saving in a cost-reduction log 4. Right-size via 2 weeks of metrics, not guesses (use recommendations API) 5. Set up anomaly alerts (spend > 20% of forecast) to catch future creep early 6. Review reserved-capacity utilization; re-balance if < 85%
Example
Monthly review findings: - 3 dev EC2 running 24/7 (only used 9-5) → schedule stop → $540/mo saved - RDS db.r5.large at 8% CPU → downsize to .medium → $310/mo saved - 200GB unattached EBS from deleted instances → delete → $23/mo saved - CloudFront serving 50GB that's also in S3 directly → tighten cache → $40/mo Total: $913/mo saved in one afternoon = $11k/year recurring. Logged in cost-reduction.md for the next review to verify savings persisted.
Common Problems
- ⚠Right-sizing based on a single peak day — you downsize, hit a real peak, and page; use p95 over 2 weeks.
Best Practices
- ✓Make cost optimization a named monthly task with an owner, not 'when someone has time' — without an owner it never happens.
- ✓Track the cost-reduction log publicly so the team sees the cumulative savings — gamification sustains the discipline.