Start with revenue
Payment providers supply observed orders, refunds, fees, and customer identities. Revenue is not inferred from traffic.
AI unit economics
Know what each visitor earns, what AI costs, and what is left.
Traffic alone cannot tell you whether an AI product is healthy. Grow or Die joins observed product activity, payment revenue, and provider-reported model usage so the denominator and the cost come from real sources.
Revenue
What customers paid
AI cost
Priced model usage
Profit
What the product keeps
Why this is different
Token totals are a cost input, not a business answer.
Price model usage on the server
The SDK reports token usage, cache usage, status, and timing. Versioned server prices turn those facts into cost.
Use honest denominators
Analytics supplies visitors and acquisition context. Missing coverage stays visible instead of silently becoming zero.
Move from site to customer
Site-wide profit works first. Customer-level profit appears only after your backend links the same account to payment and usage.
Cost coverage and reconciliation
A profit number is only as complete as its cost coverage.
Grow or Die separates priced, attributable model usage from unallocated cost. Missing identity or pricing evidence stays visible instead of quietly becoming zero.
Observed AI cost
Provider-reported usage is priced on the server and joined to a customer only when the same canonical account is present in the payment and model-call evidence.
Unallocated cost
Usage without a trustworthy account link remains in an explicit unallocated bucket. It still counts toward site-wide cost, but it is not assigned to a customer by guesswork.
Confidence label
Contribution profit is complete only when the selected window has compatible revenue, visitor, identity, and model-price coverage. Partial coverage produces a partial result, not a confident margin claim.
Provider-bill reconciliation
Compare priced SDK events with the provider bill for the same provider, project, currency, and time window. The difference is reported as a reconciliation gap and investigated before an automated cost decision is made.
The honest formula
Observed revenue − priced attributable AI cost = observed contribution profit.
Unallocated cost and reconciliation gaps stay next to that result. They are not subtracted twice, assigned to convenient customers, or hidden behind a zero. See AI cost attribution for the event-level method.
The next useful question
Which feature, model, or customer changed the margin?
Once the top-line formula is trustworthy, cost attribution shows where model spend originates and customer profitability shows whether the account paying the revenue also consumes the cost.