A cloud cost target turns a measure into a commitment a team holds for a period: which measure, from which starting value, to which target value, by when, and who checks the result. The measure on its own only describes; the target says whether a result is good and who has to act, normally the scope's cost owner. For the bill, the shape of the target decides which behaviour is rewarded. A fixed amount also brakes growth that brings revenue, and a savings target that counts estimates rewards promises rather than effect.
How it works
In the cost pillar of its Well-Architected Framework, AWS distinguishes goals, which give the direction, from targets, which put a measurable amount and a deadline on it. Goals should cover usage and cost together and the relationship between them, such as falling cost per business outcome. The FinOps Foundation covers the subject in the KPIs & Benchmarking capability of its Quantify Business Value domain (CC BY 4.0). Its activities run from picking measures that matter to the business, through agreeing improvement targets and who answers for them with the Leadership and Finance personas, to revisiting the measures over time. Its KPI library collects measures with definition, formula and data sources. How each measure is calculated is in cloud spend KPIs; this page covers how they become targets.
A few measures, one question each
A team does better with two or three targets than with ten. Each measure answers one question, belongs to one person and has a typical way of misleading:
| Measure | Question | Target form | Held by | Wrong incentive |
|---|---|---|---|---|
| Unit cost | Does the service get cheaper per unit of output? | relative: cost per unit by a date | the product owner | a unit that grows with cost |
| Allocation coverage | Does every euro have an owner? | share of allocated cost | whoever maintains allocation | split the remainder by a key and count it as allocated |
| Budget variance | Does the scope keep to its plan? | a band around the budget | the cost owner | generous budgets, spending before period end |
| Realised savings | Does what was decided take effect? | verified amount per quarter | the cost owner | counting estimates instead of evidence |
Whoever buys commitments also holds a target for the effective savings rate; a value set too high tempts them to commit beyond the stable baseline.
Baseline, target, period, review
The baseline is set before a measure starts, on the same cost basis and for the same scope as the later result. In the same capability, the FinOps Foundation counts it as a step up in maturity when the comparison case for avoided cost is written down before the action rather than assembled afterwards, and it keeps avoided cost (spending that would have come and does not) apart from savings, which lower what is already being spent. Which cost basis produces which figure is explained in list, billed, effective and amortized cost; whoever plans commitments amortized also measures them amortized.
The target is relative where the scope grows and absolute where it is meant to stay stable. AWS gives as an example of a goal that a platform's usage rises sharply while its cost rises only a little. For non-production environments or a fixed platform, an amount fits instead. For budget variance, the FinOps Foundation's Budgeting capability gives reference values by maturity: at most 20 percent variance from actual spend at Crawl, 15 percent at Walk and 12 percent at Run (the FinOps maturity model). These are points of orientation, not rules; according to the Unit Economics capability, the expected target values may differ per scope even when the measure is the same.
A quarter is a workable period for a target, read monthly (reporting cadence). It is measured on closed months, because provisional data still changes (freshness, restatements and finalisation). The level above reviews it: the cost owner reports, and management or finance assess.
OKRs as one frame
Objectives and key results (OKRs) mirror the split between goal and target: the objective states the direction in words, for example "the order service grows without its cost growing with it", and the key results make it measurable, for example unit cost per order, the service's allocation coverage and realised savings in the quarter. In its Executive Strategy Alignment capability, the FinOps Foundation names product team OKRs and performance reviews as places to anchor cost efficiency, with accountability resting with engineering and product. Microsoft's cost pillar likewise wants each team measured against expectations that fit that team's own goals. Where teams should not see raw cost data, Google suggests sharing totals, for example per environment or feature, or relative figures such as the average cost per transaction or per user.
What it means for your cost
Targets steer the decisions cost owners take every month: which recommendation to accept, whether a service scales, whether to request an exception. A relative target lets a successful service grow; an absolute one forces it into a cut or an exception. Every target also needs data. Unit cost needs a business volume next to the cost data (unit economics and unit costs), realised savings a baseline per action (from estimated to realised savings), allocation coverage a rule for the remainder. A measure without that data serves as an observation, not as a target.
Where it goes wrong
- A fixed amount for a growing service. Symptom: the team misses its cost target although unit cost is falling. Cause: the target does not know the driver.
- Estimated savings counted. Symptom: the savings target is met, the invoice is unchanged. Cause: the target measures assumed rather than realised savings.
- Inflated allocation coverage. Symptom: the rate jumps without new tags or rules. Cause: the remainder was split by a key and counted as allocated.
- A baseline chosen afterwards. Symptom: every action looks successful. Cause: the baseline was picked after the action rather than documented before it.
- Two cost bases. Symptom: the target is missed in exactly the month a commitment was paid upfront. Cause: the plan used amortized cost, the measurement uses billed cost.
- Underspending punished. Symptom: spending rises shortly before the period ends. Cause: unused budget cuts the next period's budget.
Example
In the baseline quarter a service has cost K0 and volume V0, so unit cost S0 = K0 / V0. The target is to lower unit cost by the share r by the end of the target quarter: S1 ≤ S0 × (1 − r). If volume grows by the factor g, the service may cost K1 ≤ K0 × g × (1 − r). An absolute target that lowers cost by the same share, K1 ≤ K0 × (1 − r), instead demands unit cost of at most S0 × (1 − r) / g: the more the volume grows, the more each unit would have to save. If volume does not grow (g = 1), both targets say the same, and an amount is the simpler form.
Sources
Read on 2026-10-16. FinOps Foundation KPIs & Benchmarking, FinOps KPIs, Budgeting, Unit Economics and Executive Strategy Alignment (CC BY 4.0); AWS COST02-BP02 implement goals and targets; Microsoft recommendations for creating a culture of financial responsibility; Google Cloud foster a culture of cost awareness.