Why unit economics beat raw token spend — and how to track the cost of an outcome rather than a call.
Total token spend tells you how much you are consuming, but not whether that consumption is producing anything useful. Two systems can have identical token bills while one resolves most requests and the other fails half of them. As organisations bring AI into production at scale, unit economics — cost tied to a unit of value — become the more honest measure. Cost per successful task (CPST) is an emerging FinOps-for-AI practice that frames cost around the outcome rather than the raw call or token.
This shift matters because adoption is now widespread. The FinOps Foundation's State of FinOps 2026 found that 98% of organisations now manage AI spend, up from 63% in 2025 and 31% in 2024 — so the question is moving from whether to manage AI cost to how to measure its value.
CPST only works if you can say what success means. A task is a unit of work the AI is meant to complete — a resolved query, a correctly extracted field, a completed step in a workflow — and a successful task is one that met its acceptance criteria. The definition is yours to set, but it should be specific, observable and agreed with the people who own the outcome.
With a definition in place, CPST is the cost attributed to a use case divided by the number of successful tasks it produced over the same period. The cost side draws on the same allocation and attribution discipline used elsewhere in FinOps, so model, platform and supporting spend are all counted. The success side draws on whatever signal you use to confirm an outcome. Tracking the two together turns cost from an undifferentiated bill into a per-outcome figure you can compare across time, models and approaches.
Once CPST is visible, optimisation changes character. Instead of simply cutting tokens, you can ask which changes lower the cost of a successful outcome — a cheaper model that still meets the bar, a prompt or retrieval change that raises the success rate, or removing work that rarely succeeds at all. Sometimes the best move increases raw spend slightly while lowering CPST, because more tasks now succeed. Optimising on outcomes keeps cost decisions tied to value rather than to volume.
Our AI FinOps capability is aligned with this outcome-based view — bringing cost attribution together with a notion of success so spend can be read per outcome, not just per call.
Bring attribution and a definition of success together to read AI cost as unit economics.