What is Savings as Code?
Every saving is a change you approve, and every change is measured on the bill.
In 30 seconds
Savings as Code is a way of cutting cloud and AI spend in which every saving reaches your team as a pull request: what changes, why, the expected saving and how to undo it.
- Your team approves it like any other change.
- About 30 days later, the saving is measured on the bill.
- Each saving gets one line in a savings ledger.
Why it exists
Finding savings is the easy half, and most cost reports stop there: a list of findings, each with an estimate. The hard half is making the change safely and proving it worked. Savings as Code puts that work where engineering already happens: in a pull request, with review and a history.
How it works
- Find the savings on the bill, each with an estimate.
- Propose each one as a pull request, with the rollback.
- Approve: your team reviews it like any change.
- Merge through your usual pipeline.
- Measure it on the bill and record it in the ledger.
The same five steps work for model and token spend: see Savings as Code for AI.
| Change | Estimate a month | Measured a month |
|---|---|---|
| API service to a smaller instance family | $1,150 | $1,080 |
| Delete unattached volumes and old snapshots | $620 | $640 |
| Savings Plan sized to steady usage | $2,300 | $2,210 |
| Support summaries routed to a smaller model | $900 | $840 |
Illustrative example.
For engineers: what the pull requests look like
- Cloud: a Terraform change that moves a service to a smaller instance family, removes unattached volumes, sets lifecycle rules on old logs, or adds a Savings Plan sized to steady usage.
- AI: a change that routes a task to a smaller model, turns on prompt caching, moves overnight jobs to batch processing, or caps output tokens. See Savings as Code for AI.
- Principles: one saving per pull request; no change without an owner, a reason and a rollback; estimated before, measured after.
Questions
Is Savings as Code the same as managing Savings Plans in Terraform? No. A Savings Plan purchase is one kind of saving. Savings as Code covers every saving, cloud or AI, and adds the approval and the measurement on the bill. AWS says Savings Plans save up to 72% compared to On-Demand prices (AWS, checked November 7, 2026).
What if we don't use infrastructure as code? The pull request goes to a small change-log repo instead. It describes the console change, and your team approves it there before anyone touches the account.
Why measure every change? A saving you cannot measure is an opinion. The measured line is what finance can audit and what makes a fee on savings fair. It also shows when an estimate was wrong, so the next one is better.
What if a change goes wrong? The pull request carries its rollback. Revert it and the previous configuration comes back on the next deploy. That is why each saving is its own pull request: one revert undoes one change, and the others stay in.
Which savings come first? The ones with the largest estimate and the lowest risk. Idle resources and unattached storage go before resizing, and resizing goes before commitments: a Savings Plan sized to an oversized fleet locks the waste in for one or three years.
How we run it
MTO FinOps runs Savings as Code as a service on AWS. $0 upfront: we start with a free review of your AWS bill, and during the review we only look. We find the savings and implement them. You approve every change. You pay 50% of what you save for 6 months. Then 100% is yours.
A company spending about $42k a month on AWS can save about $6k a month*.
*Illustrative example.
Once the savings are in, Spend as Code is how they stay in.