Unit Economics

Unit economics in the FinOps sense relates technology cost to a defined unit of activity or outcome. FinOps Unit Economics distinguishes resource-efficiency metrics, such as cost per GB or token, from business-unit metrics, such as cost per customer or resolved case. The distinction helps select useful measures; it does not reserve each kind for one job title. A unit cost supports a decision but does not, on its own, establish value or identify waste.

Define the cost and the unit together

State the service scope, currency, period, and cost basis. Specify whether the numerator includes shared services, idle resources, failed work, and commitments, and how prepaid charges are amortized if the comparison uses an economic-period view. Reconcile that view separately with invoice cash payments. Define the denominator’s eligible population, successful output, retries, duplicates, and missing records. With zero qualifying output, show the expenditure and zero count rather than a fabricated finite ratio.

For video processing, actual processed hours and unique useful output hours answer different questions. In a hypothetical example, processing 100 hours once costs 100: both ratios are 1. Repeating the same work six times at the same rate costs 600 across 600 processed hours, still 1 per processed hour. If useful output is still 100 distinct hours, cost per useful hour is 6. Reporting both exposes repeated work without mistaking it for new output. Quality, freshness, and recovery checks determine whether the repeat work is unnecessary.

Cost per training run should say whether failed runs and idle infrastructure are included. Cost per promoted model also depends on promotion criteria and the difficulty of the work. FinOps notes that AI measures often begin with cost per token and develop toward outcomes such as an assist or resolved case. Those outcome definitions need evidence too: fewer tokens or more reported actions do not themselves establish a better service.

Read a trend with its total and its context

The fictional media company’s cloud cost rises from 400,000 for 100,000 active subscribers at month 0 to 800,000 for 120,000 at month 12. With the same normalized monthly period and cost scope, cost per subscriber rises from 4.00 to 6.6667, about 66.7%. This warrants investigation. It does not show that most of the increase is waste: customer activity, quality, product mix, service requirements, and prices can change. At unchanged efficiency and price, 20% more subscribers each requiring 5/3 as much useful work would double cost.

At month 24, cost is 510,000 and subscribers are 138,000, 15% more than at month 12. The unit cost is about 3.6957, roughly 7.6% below month 0 and 44.6% below month 12. Total monthly expenditure is nevertheless 27.5% above month 0. These figures describe a fictional observation, not the causal effect of any individual change. A cost/revenue ratio also needs the other relevant costs before it can support a profitability claim.

FinOps recommends trends within a defined scope and metrics suited to the organization’s goals. Keep total spending, workload composition, quality, and the cost of improvement visible. Compare observed changes with the chosen baseline and distinguish them from attributed effects or cash savings. A lower unit cost can coexist with higher expenditure or an unacceptable service.

The Bill Doubled: Cloud Cost Operations and FinOps develops the allocation, commitment, and operating decisions in this case. The Case Said Four Million and the Cost Was Nine distinguishes lifecycle cost, financial savings, and a public-service cost-per-transaction measure.

References:

• FinOps Foundation, Unit Economics (CC BY 4.0)
• FinOps Foundation, FinOps Principles (CC BY 4.0)