The cloud spend KPIs worth an engineering leader's attention are the ones that connect cost to a decision a team can make. Four carry most of the weight: unit cost, which is spend divided by a business driver such as cost per transaction or per active user; commitment coverage and utilization, which show how much steady usage is on discounted rates and whether those commitments are being used; waste rate, the share of spend on idle or oversized resources; and forecast accuracy, how close the prediction was to the actual bill. Absolute spend going up is not a problem if unit cost is falling and the business is growing; absolute spend flat while unit cost rises is the real warning. The point of these KPIs is to separate growth from inefficiency.
Here is each KPI, how to define it, the target to aim for, and the behaviour it is meant to change.
Why not just track total spend?
Total cloud spend is the least useful number for an engineering leader because it conflates two things that demand opposite responses. Spend rising because the business is growing is healthy; spend rising because resources are oversized and commitments are idle is waste. A single total cannot tell them apart, so it generates either false alarms or false comfort. Unit economics solves this: by dividing spend by a business driver you see whether each unit of value is getting cheaper or more expensive to serve, which is the question that actually informs an engineering decision.
What are the four KPIs and their targets?
- Unit cost. Spend divided by the driver that matters for your product, such as cost per thousand requests, per active user, or per order. Target: trending down quarter over quarter as efficiency work compounds. This is the headline metric for the board and the one that survives growth.
- Commitment coverage and utilization. Coverage is the share of eligible steady usage on Savings Plans, Reservations, CUDs, or Universal Credits; utilization is how much of what you committed to you actually used. Target: high coverage of the defensible base load with utilization near full, so you neither pay on demand for steady usage nor pay for commitments you waste.
- Waste rate. The share of spend on idle, orphaned, or oversized resources. Target: low single digit percent and falling, tracked so a spike triggers a specific cleanup rather than a vague concern.
- Forecast accuracy. How close last period's forecast came to actual spend. Target: tight enough that finance can plan on it, because forecast accuracy is what makes commitment decisions and budgets credible.
How do these KPIs change engineering behaviour?
Each KPI is chosen because it points at an action. A rising unit cost prompts a rightsizing or architecture review of the service whose cost per unit climbed. Low commitment coverage prompts a purchase decision against a defensible forecast; low utilization prompts trimming overcommitment. A climbing waste rate prompts a specific cleanup of idle and orphaned resources. Poor forecast accuracy prompts better tagging and demand modelling, because you cannot forecast what you cannot attribute. KPIs that do not map to an action become wallpaper, so the discipline is to track few metrics and tie each to an owner and a response.
Who owns each KPI?
KPIs without owners drift. Unit cost belongs to the product or service team whose driver it tracks, so the people who can change the architecture see the number. Commitment coverage and utilization usually sit with a central FinOps or platform function that has the portfolio view. Waste rate is shared: central reporting surfaces it, individual teams clear it. Forecast accuracy belongs jointly to finance and engineering, because it depends on both a good demand model and clean attribution. Reviewing these in a regular spend meeting, with each number attached to a name, is what turns measurement into reduction.
A worked example
A scaling fintech reported only total cloud spend to its leadership, and the number rose every month, which read as a problem. Introducing unit cost told a clearer story: cost per transaction was actually falling as the platform matured, so the rising total was growth, not waste. The same KPI set surfaced the real issues, a waste rate inflated by orphaned test environments and commitment utilization well below full because an old reservation no longer matched usage. Assigning each KPI an owner and reviewing them monthly drove the cleanup and a recommitment to a defensible forecast. Unit cost kept falling while the genuine waste was removed, and the work formed part of the program that left the company materially lighter on cloud spend. Figures are verified against billing data and anonymised.
Frequently asked questions
What cloud spend KPIs should engineering leaders track?
Why is total cloud spend a poor KPI?
What is a good commitment utilization target?
Talk this through with us
We help engineering leaders define the small set of cloud spend KPIs that drive decisions across AWS, Azure, GCP, and OCI, and stand up the reviews that turn them into reductions. We take zero provider commissions and answer only to you. Our guarantee: we reduce your cloud spend or we reimburse our service fee, on either a Fixed Fee scoped up front or a no risk Gainshare basis. Book a strategy call to scope it for your estate, and follow more analysis in The Cloud Spend Navigator.
Put a defensible number on your cloud spend.
No provider in the room, no published price list. Tell us your footprint and we will scope the savings against your billing data — we reduce your cloud spend or we reimburse our service fee.
The Cloud Spend Navigator: what changed in cloud pricing, commitments, and FinOps — no vendor spin.