AI Unit Economics Calculator
Cost per completed task, margin per task and AI cost per customer, from the three numbers you already have.
Everything one attempt spends: model calls, retrieval, reranking.
What one completed task earns. 0 for internal tools.
Against a written success bar: validated, not reopened, actually done.
Refine the inputsHide the detail
Review, cleanup, the manual fallback.
For AI cost per customer, the subscription view.
Cost sits well below revenue. Optimisation is an improvement project here, not a rescue.
Cost per completed task, as quality changes
Cost per call is flat across this whole chart. The honest metric is not, and the left side is where it quietly eats the margin.
Why cost per task and not cost per call
The property that makes one metric honest and the other flattering.
Failures stay in the numerator
Three billed attempts, one finished task. The honest metric charges all three to the one, which is why it rises when quality falls.
Per-call dashboards flatter you
Cost per call improves as retries rise, because each retry is another cheap call. It rewards the exact regression you need to catch.
The verdict is relative
A cost per task is not good or bad on its own. It is good or bad next to what the task earns, which is why the margin is the headline.
The method behind this tool, including the five fields to log and what "completed" has to mean, is in cost per task, and the wider sequence is in LLM cost optimization.
AI unit economics FAQ
What is AI unit economics?+
The cost and margin of AI work measured in business units: cost per completed task, margin per task, AI cost per customer. It differs from traditional software economics because the marginal cost of serving one more unit is not near zero; every task consumes paid compute, and failed attempts consume it without producing anything.
How do I calculate cost per completed task?+
Divide total spend over a period by the number of tasks that finished successfully in that period. Failed attempts stay in the numerator, which is the point: three billed attempts that produce one good answer give a cost per completed task of three attempts, not one. The calculator adds retries and human review time on failures for you.
What margin should an AI feature have?+
There is no universal bar, but a useful reading: cost well below revenue per task is healthy, cost close to revenue is an engineering problem with known levers, and cost above revenue at any plausible efficiency is a pricing or product problem that optimisation cannot fix. The verdict card applies exactly that reading.
What is AI cost per customer?+
Cost per completed task multiplied by the tasks a customer consumes per month. It is the number to put next to your subscription price, and it is the difference between a customer who is profitable and one who quietly is not. Heavy users on flat plans are where AI margins go to die.
Why not just track cost per API call?+
Because it improves when quality gets worse. Failures and retries make each call look cheap while making each finished task expensive, so a per-call dashboard rewards exactly the regression you most need to catch. Cost per completed task moves in the honest direction, rising when retries rise.
Cost per task instrumented, not estimated
The calculator models it. We instrument it: the five fields, the task_id join and a written success bar, before the model choice. Book a free cost review.