Cost per Task

Cost per task is what one completed unit of work costs you in model spend — a resolved ticket, a generated report, a reviewed document. It is the denominator that connects AI infrastructure to a business model, and it is the number most teams do not measure. Token counts are the wrong unit for two reasons. They do not include retries, tool calls, failed attempts or the second model that judged the first one's output, and they are not what the customer buys. A task that costs $0.04 in tokens but needs 1.6 attempts on average and a re-ranking pass costs closer to $0.09, and that is the figure to price against. Track it per workflow rather than per account. Averages across a product hide the one workflow whose cost per task exceeds the revenue per task — which, in seat-priced products, is how a heavy user quietly becomes unprofitable while the dashboard still shows healthy gross margin. The lever that moves it most is usually not a cheaper model. It is fewer retries, shorter context, and a cache that catches repeated work.