EXAM / AI research & technology

Reading AI revenue and profitability: cost per accepted task

Measure revenue against compute, review and rework—not just the price of an API call.

An AI processor and measuring instruments on a technical bench

Revenue growth can coexist with weak economics if each completed job consumes too much computation or human work. Start with an outcome a customer accepts, then measure revenue and the variable cost of delivering it. The contribution calculation below is a management example, not a standard accounting definition of gross profit or net profit.

A hypothetical task-level calculation

All figures are invented currency units per accepted task, after allocating failed attempts. No company results or market averages are implied.

Table scrolls horizontally

ItemAmount
Revenue10.00
Compute and model use2.00
Human review and rework3.00
Other variable delivery costs1.00
Contribution before fixed overhead10 − 2 − 3 − 1 = 4 (40%)

Stress-test the human workload

If human review and rework rise from 3 to 6 while other assumptions stay fixed, contribution falls from 4 to 1, or 10% of revenue. That remainder still has to support fixed payroll, research, sales, administration and other excluded costs. The business can therefore grow revenue while generating little or negative net profit.

Ask for the reconciliation

  • Reconcile paid tasks, accepted tasks, refunded tasks and cash collected. Explain how retries, unused capacity and customer support are allocated.
  • Keep billed revenue, recurring revenue estimates and cash receipts distinct. Compare customer cohorts over time; a pilot subsidy or promotional credit may distort early margins.

Further reading

General information only; not investment advice, an offer or a recommendation. Any participation is subject to eligibility, documentation and applicable permissions. Capital and returns are not guaranteed; some or all capital may be lost.