Cost and cloud economics / March 2026
Four decisions that set the ceiling on a cloud cost program
Most of the money a cloud cost program recovers is determined by executives in the first month, not by engineers in the third. Four decisions fix the upper limit on the result. Make them early and in writing, or the technical work will stall against appeals it was never given the authority to settle.
The Framing
This is a decision problem wearing an engineering costume.
Executives usually ask us the wrong opening question. They ask which savings exist. In almost every mid-market estate we review, the savings are already visible to anyone holding read access on the billing account, and the platform team can name them within a week. The data is there. The capacity to act is there. What is absent is a set of decisions that let an engineer terminate something without negotiating with the team that provisioned it.
The scale of the opportunity is not in dispute across the industry. Flexera reported an average of 27 percent self-assessed cloud waste in its 2024 State of the Cloud study, with respondents estimating further reduction available under stronger governance. Note what that finding measures. It is what organizations already believe they are wasting. The constraint is rarely discovery.
So the executive question changes. Instead of asking what to cut, ask which four rulings you are prepared to sign in the first month, because every one of them will be tested by a team with a reasonable-sounding objection.
This work is usually labeled FinOps, and the label matters less than the four decisions underneath it.
The Four Decisions
Sign these in the first 30 days, in this order.
- 1. Discipline before tooling. Below a few million dollars of annual cloud spend, a capable engineer with billing access and a spreadsheet will surface most of what a licensed cost platform would. It costs no seat fee. Defer the purchase until the manual work has shown you which taxonomy you actually need. Buying first produces a tool configured around a vendor's model of your business.
- 2. Set the default to off. Decide, before anyone is asked, that an unused environment is terminated after a fixed quarantine window and that the burden of proof falls on the team asking to keep it. Without that ruling, month one produces a very good list and no reduction, because every candidate becomes a debate the engineer cannot win.
- 3. Commit the floor. This is the CFO decision. Lock the stable base of compute on the longest commitment terms available, layer shorter commitments over the next band, and let the top of the demand curve run at on-demand rates. Provider discount instruments differ in scope and flexibility, so a single blanket commitment policy applied across two providers is a mistake, not a simplification.
- 4. Name the owner and the unit-cost measure. One person accountable to cost per active user, cost per transaction, or cost per gigabyte processed. Without that, the recovered spend reappears within two quarters and nobody is responsible for noticing.
The Objection You Will Hear
Finance will say the architecture moves too fast to commit.
The objection is sincere and usually wrong. The stable base of compute at a mid-market business is far steadier than executives expect. The volatility is concentrated in the top of the demand curve, which is precisely the band the commitment ladder leaves uncommitted. The exposure from over-committing is real and bounded by the term. The exposure from under-committing is paying full list price every month with no end date.
The second objection comes from the platform team and is about spot capacity. Interruptible capacity belongs in batch and stateless work. It is not a primary commitment lever for an organization without the engineering depth to absorb an interruption cleanly, and treating it as one converts a cost decision into a reliability decision that the executive team did not intend to make.
Reporting Line
Where the cost owner reports changes what the cost owner optimizes.
An owner reporting into finance will optimize against budget variance and miss waste created by engineering choices. An owner reporting into engineering will under-weight commitment posture, because commitments feel like a procurement problem. The arrangement that works pairs a dotted line into both with a solid line into whichever function the CFO and the technology leader jointly trust to enforce a decision. In practice that is usually platform or infrastructure engineering, with finance supplying the unit-cost framing.
Put the unit-cost measure on the same review as reliability and delivery. If cost per transaction is absent from the dashboard that shows latency, engineers will improve latency at the expense of cost, exactly as instructed.
A logistics operator paid for legacy hosting, three support contracts and a recovery site that had never been tested. We took the four decisions to the executive team in the first fortnight, wrote the termination default and the commitment floor into policy, and gave one platform owner a unit-cost target reviewed alongside reliability. Run cost fell 36 percent and the reduction held through the following year.
Scope Discipline
What does not belong in the first 90 days.
Executive attention is the scarce input, and bundling adjacent programs into a cost effort spends it badly. Each of the following may be worth doing on its own merits. None of them belongs inside this window.
- Migrations and container modernization. Both have their own business cases and their own timelines. Attaching them dilutes accountability for the cost result and pushes the finish line past the point where executive attention stays on it.
- A second cloud provider as a price lever. The theoretical arbitrage rarely survives egress charges, duplicated tooling, and the engineering cost of maintaining parity. If you are on one provider today, the stronger move in most cases is a harder negotiation with that provider.
- Cost platform procurement. Covered above, and worth repeating because it is the most common way a cost program becomes a software project.
- Chip architecture migration for stateless services. The price and performance case is genuine. The dependency validation work generally exceeds the window you have left, so park it on the roadmap that follows.
What To Do Next
Five moves for the executive team.
- 1. Write the four decisions down and circulate them before the technical work starts. Verbal support does not survive the first appeal.
- 2. Require tag coverage above ninety percent across compute, storage and data services as the one technical precondition you personally track.
- 3. Set the termination default and the quarantine window yourself, so no engineer has to defend it alone.
- 4. Approve the commitment ladder as a CFO decision with a stated floor, and revisit utilization monthly so expirations do not slip.
- 5. Put the unit-cost measure and a quarterly waste sweep into the standing operating review before the program ends, not after.
Where the executive team makes these four calls, the discipline stays in place and the following year's bill grows in step with the business. Where it does not, the savings still show up by the third month and most are gone again by the ninth.
Related Service
Technology Cost and Cloud Economics
We put the cost decisions in front of the executive team, in the order they have to be made.
Start A Conversation