AI Feature Unit Cost & Usage Allowance Calculator

What the calculator measures

This deterministic calculator estimates a monthly account scenario from an observed, mature AI-task cohort. It runs locally and never calls an LLM. Each successfully completed task must involve at least one observed model request; a task may include multiple requests. Failed and abandoned work contributes cost but not a successful completion. Use the same task definition for historical evidence and planned usage.

All money inputs use one currency. Use net revenue allocated to this feature or an explicitly defined package scope. Other account-level costs must cover that same scope and must not duplicate task costs. This is contribution after the listed costs, not accounting gross margin, net profit or a pricing recommendation.

Formulas and units

• Observed task cost: model cost + other task-variable cohort cost, in currency units.
• Unit cost: observed task cost / successfully completed tasks.
• Requests per completion: all model requests / completed tasks. This includes multi-step work and failures; it is not a retry probability.
• Projected task cost: planned monthly completed tasks × unit cost.
• Monthly contribution: monthly net revenue − account-level variable cost − projected task cost.
• Contribution margin: contribution / net revenue. Undefined at zero revenue.
• Task budget: revenue × (1 − target margin / 100) − account cost.
• Cost-based allowance: floor(task budget / unit cost), provided revenue and unit cost are positive and budget is nonnegative.

A negative budget means the target is missed even at zero tasks. Zero completions means unit cost is unknown. Zero observed unit cost means the model has no finite cost-based cap, not that future costs are zero. Caps above one billion tasks are explicitly outside the tool's range. Counts are whole numbers from 0 to one billion, money inputs are finite values from 0 to one billion, and target margin is 0–100%.

Reproduce the synthetic example

Inputs: 1,200 requests; 1,000 completions; EUR 23.40 model cost; EUR 4.68 other task cost; EUR 49 monthly net revenue; EUR 5 account cost; 500 planned tasks; 60% target. These are invented teaching inputs, not current API prices.

Results: EUR 0.02808 per completed task; 1.2 requests per completion; EUR 14.04 planned task cost; EUR 29.96 contribution; 61.14% displayed contribution margin; EUR 0.56 budget headroom; and 519 whole tasks at the target. The JSON export retains unrounded ratios. At 520 tasks, the cost of EUR 14.6016 exceeds the EUR 14.60 task budget.

Turn the output into a review, not an automatic limit

Compare a proposed allowance with the cap, then test whether that allowance supports a useful customer outcome. The cap assumes constant task mix, completion quality, request cost and failure burden. Run separate scenarios for materially different cohorts. Do not multiply an average by an invented statistical safety factor and call it a confidence bound.

The calculator does not fetch invoices, validate currency conversion, determine tax treatment, infer demand or enforce entitlement. Binary floating-point arithmetic and display rounding make it unsuitable as a billing engine; near-boundary operational decisions need an approved rounding policy and cost reserve. No input is sent to an AI service or saved after leaving the page.

Sources and scope

Stripe's 20 August 2026 pricing-leader report supplies a qualitative reason to revisit AI pricing operations, not a demand estimate. Anthropic's prompt-caching documentation separates uncached input, cache creation and cache reads; its pricing documentation distinguishes token and additional tool charges. Both living documents were accessed 20 September 2026; their publication dates are unknown. No provider prices are embedded in this resource. The worksheet, guardrails and examples are original Growthcraft synthesis, not vendor-endorsed guidance.