Step 1: see where the money goes
You cannot trim what you cannot see. Start with visibility, because every later step depends on it.
- Tag every resource with an owner, a product and an environment
- Split the bill by team and product, not only by service
- Look at untagged spend each month, and make it someone's job
- Include costs that sit outside the cloud invoice: licences, SaaS, support contracts
Step 2: trim the waste
Most bills contain waste that nobody chose. These are the usual places to look.
| Where | What to look for | What to do |
|---|---|---|
| Compute | Idle or oversized instances | Right-size to real usage, or switch off |
| Non-production | Test and staging running all night and weekends | Schedule them to switch off out of hours |
| Storage | Unattached disks, old snapshots, data nobody reads | Delete or move to a cheaper tier |
| Networking | Unused addresses, heavy data transfer between zones | Clean up, and review architecture |
| AI and GPU | Experiments left running | Budgets and alerts, and shut down what is not in use |
Test each change in staging first and keep a rollback ready. Savings are worthless if production breaks.
Step 3: commit carefully
Reservations and savings plans give a discount in exchange for a commitment. They help only for usage that is stable.
- Trim first: committing to waste locks the waste in
- Commit to the steady base, and leave the variable part on demand
- Review coverage every quarter, because usage changes
- Check licence benefits, such as Azure Hybrid Benefit, where they apply
Step 4: make it a habit
- Budgets per team and product, with alerts before the limit
- Anomaly alerts, so a spike is found in days, not at invoice time
- A monthly review with owners from engineering and finance
- Cost checks in your release pipeline, so new resources are sized with cost in mind
A one-off clean-up helps once. A routine keeps the savings.
Frequently asked questions
What is FinOps?
A practice that brings finance, engineering and the business together to manage cloud spend. It follows three phases: inform, optimize and operate.
How much can we save?
It depends on how your cloud is run today. A read-only review of your bills shows what is possible, and we do not promise a percentage up front.
Will cost changes affect production?
Not if they are tested in staging first and applied in agreed windows with a rollback ready.
Where should we start?
With a read-only bill review: spend by service and team, idle resources, commitment coverage and tagging gaps.