By how much will FinOps reduce the cloud bill, and from which quarter? A finance department asks the question in these terms when it sees the cloud line of the budget rise from one closing to the next.
A cloud bill falls in two ways: consuming fewer resources, or paying less for each resource consumed. FinOps practice adds no third. It provides the measures that show where each is open, and whether the reduction obtained is real. Four are enough to answer: the share of allocated spend, the unit cost, the commitment utilisation rate and the amount of unused resources. None reads as a saving achieved, and each rests on a calculation convention to be known before commenting on the figure. The discipline itself is presented in our entry on the definition of FinOps. This article keeps to its indicators.
The FinOps measures behind the objective of reducing the cloud bill
| Indicator | Common calculation convention | What it lets you say | What it does not say |
|---|---|---|---|
| Share of allocated spend | Spend attached to an owner, divided by total spend for the period | What share of the bill has someone responsible who is able to act | Whether the attached spend is useful |
| Unit cost | Spend attributed to a product, divided by a unit of activity | Whether spending follows activity or exceeds it | Whether the total bill will fall |
| Commitment utilisation rate | Commitment consumed, divided by commitment purchased | Whether the discounts bought are being used | Whether the committed usage was necessary, and what share of consumption remains at the on-demand rate |
| Unused resources | Cost of resources matching a defined waste pattern, as an amount or as a share of spend | Where a reduction in usage is possible | Whether deletion is risk-free |
The FinOps Foundation framework places these measures in several of its capabilities: allocation, unit economics, rate optimisation, usage optimisation. For some it publishes a formula, which we quote below where we have read it on its website, finops.org. The framework presents itself as non-prescriptive: these formulas are proposals, which each organisation adapts and documents.
Without a share of allocated spend, no reduction target can be verified
The common convention divides the spend attached to an owner (team, product, cost centre) by total spend for the period. Among its allocation indicators, the FinOps Foundation framework proposes the percentage of costs allocated directly to organisational units and, as a mirror, the percentage of costs that cannot be categorised.
Several choices make the result vary and are written next to the figure. Does the numerator count only directly attached costs, or also shared costs distributed by an allocation key? Is spend taken as billed, or after amortising prepaid commitments? Is a tagged resource considered allocated, when the tag may designate a team that has been disbanded?
This indicator limits what can be discussed with someone. A reduction in usage is decided with the owner of the resource, and the share without an owner escapes this discussion. It says nothing about whether the spend is well founded: allocated spend is not justified spend. A rate can also rise because the remainder has been attached to a “shared” account, without anyone being any more accountable for it.
Unit cost separates growth from drift
Unit cost divides the spend attributed to a product by a unit of activity: transaction, case processed, active user. The FinOps Foundation framework distinguishes two families. Resource efficiency measures (cost per gigabyte stored, per virtual processor) are in the engineers’ hands. Business measures (cost per transaction, per customer) speak to senior management and require reliable activity data.
This is the indicator that speaks to a finance department, and it disappoints if an amount is expected. A unit cost falling on a rising volume gives a bill that rises: efficiency improves, and so does the budget line. The useful reading is a trend on the same scope. The same framework notes that comparison between unrelated products is less usable.
The figure depends on what goes into the numerator. With or without shared costs, it does not tell the same story, and a change of allocation key shifts the curve without any resource having moved. The denominator has its own pitfalls: when activity declines, the fixed share of spend pushes the unit cost up although no waste has appeared.
The commitment utilisation rate is read with coverage
Providers grant discounts in exchange for a consumption commitment over a period. The utilisation rate compares the commitment actually consumed with the commitment purchased. The FinOps Foundation publishes this formula under the name Commitment Utilization Score, and its complement, the unused share of the commitment, under that of Percentage of Commitment Discount Waste.
A low rate means that the organisation is paying for capacity it does not use: the correction involves reassigning the commitment when the contract allows it, otherwise the next renewal. A high rate is falsely reassuring if read alone. It does not say what share of eligible consumption is covered by a commitment. That is the purpose of the coverage rate, whose denominator (eligible consumption, or total consumption) varies from one tool to another. A fully used commitment on low coverage leaves most consumption at the on-demand rate.
Neither tells you whether the committed usage was necessary. A commitment taken on oversized machines shows an impeccable utilisation rate. To measure the overall effect of discounts, the Foundation proposes an Effective Savings Rate, with two possible formulas: the convention chosen is specified each time the figure is published.
Unused resources, an estimated amount
This indicator adds up the cost of resources that match a waste pattern defined in advance. Under waste reduction, the FinOps Foundation framework cites orphaned storage volumes, excess backups or snapshots, and unused test environments. To these are added active but little-used resources, identified by a utilisation threshold over an observation window.
The amount obtained depends on these definitions. A raised utilisation threshold or a shortened window inflates it. The same framework recommends examining long cycles, month end or quarterly peak, before reducing a resource. A little-used machine may also be a backup required by a continuity plan. The figure therefore identifies candidates, which each owner confirms or rules out, which brings us back to the first indicator.
It has a blind spot: a resource constantly used by a process whose result nobody reads does not appear in it.
Answering the finance department without putting forward an amount
Read together, these indicators make it possible to break down the change in the bill into three effects. This is a reading convention, close to the variance analysis used in management accounting: a volume effect (activity has changed), a usage effect (consumption per unit of activity has changed) and a price effect (the price paid per unit consumed has changed). FinOps acts on the last two. The first belongs to the business, and a finance department has no reason to want to reduce it.
The answer to the initial question then takes the form of a four-line statement, each line accompanied by its calculation convention, its scope and the name of the person accountable for it. A reduction target is set on the usage and price effects, the former within the limit of the share of spend that has an owner.
Putting these measures in place, together with the reviews that keep them alive and the rules for pricing commitments, is the purpose of our FinOps offer. After a cloud migration, the same statement is used to compare observed costs with the estimates on which the decision was based.
The first figure to ask for remains the share of allocated spend: until it is established, the other three concern an unknown fraction of the bill.