Free guide

Cut your cloud bill without breaking anything

The FinOps checklist I use on Azure, AWS and Kubernetes platforms. Ten steps, in the order that saves the most with the least risk.

What's inside

1. 1. See where the money goes

   • Pull amortized cost per resource and meter from the billing API, not just the monthly invoice.
   • Build a forward run-rate so you see the effect of a change within days, not at month end.
   • Tag every resource with product, team and stage. Untagged spend gets an owner within a week.
   • Send a short daily cost report to the team chat. What gets seen gets fixed.

2. 2. Check before you cut

   • For every candidate: what runs on it, who uses it, and the peak load (p95), never the average.
   • Anything a customer could notice is out of scope until proven safe.
   • One change per ticket, each with impact, risk and the way back.

3. 3. Right-size compute

   • Set Kubernetes requests and limits from real p95 usage, then let the node count follow.
   • Autoscale (cluster autoscaler or Karpenter) and use spot capacity for jobs that can be interrupted.
   • In one engagement: a production cluster went from 7 to 4 nodes with no customer impact.

4. 4. Downsize and delete

   • Idle VMs, orphaned disks and volumes, unattached IPs, old snapshots and forgotten test resources.
   • Database standbys and high-availability tiers only where the business really needs them.

5. 5. Write a data retention policy

   • Logs: debug off in production, retention per log class, cheap storage tiers for audit logs.
   • Backups: retention matched to your recovery goals and legal duties, not to defaults.
   • Object storage: lifecycle rules that move old data to cold tiers and delete what nobody needs.

6. 6. Fix the expensive queries

   • Find the top queries by cost, not by count. One fixed query cut the load on a Cosmos DB database about 40×.
   • Move small tenant databases into the cluster (for example with CloudNativePG) when it pays off, with row counts and checksums checked.

7. 7. Commit last, not first

   • Buy reservations or savings plans after right-sizing, so you don't commit to machines you are about to remove.
   • Treat commitments as hard to undo: Azure savings plans can't be cancelled, AWS Savings Plans can only be returned within seven days, and Azure reservation refunds are capped. When the footprint changes, use the exchange options that exist.

8. 8. Turn environments off

   • Dev and test off at night and on weekends by default, rebuilt automatically in the morning.
   • Preview environments with a time limit, and nightly sweeps for orphaned resources.

9. 9. Watch the network

   • Alert on cross-zone and egress traffic, and review NAT gateways and load balancers nobody owns.

10. 10. Keep it down

    • Budgets with forecast alerts per team, and a weekly one-page report with a two-month forecast for management.
    • A named owner per cost center, and the checklist rerun every quarter.