How do you calculate AI margin per customer?
AI margin is the share of customer revenue left after AI usage cost: (revenue − AI cost) ÷ revenue × 100. The calculator also subtracts other variable costs to show contribution margin. MarginFuse’s dashboard focuses on revenue minus AI cost; it does not automatically include every cost of running your business.
For example, $99 revenue and $30 AI cost leave $69, or a 69.70% AI margin. If serving that customer also costs $5 in payment fees and infrastructure, contribution is $64, or 64.65%. These are illustrative inputs, not measured customer results. Salary, acquisition spend and other fixed costs still need to be covered.
At zero revenue, a percentage margin is undefined. The dollar loss remains meaningful: a free user who consumes $8 of AI has an $8 cost to recover elsewhere. We show that loss without inventing a percentage.
Which revenue number belongs in the calculation?
Use revenue attributable to the same customer and service period as the cost. A $1,188 annual subscription does not supply $1,188 of revenue every month; use the share earned during the period you are measuring. Account for refunds and credits, and avoid comparing an entire year’s payment with a single month’s AI bill.
For mobile subscriptions, start with store proceeds after commission and tax. Do not subtract those fees again under other variable costs. RevenueCat provides the store data MarginFuse uses; the app-store revenue guide explains the distinction. A declared plan price is useful for planning, but does not prove that a payment was collected. Keep that uncertainty visible.
What counts as AI cost?
Include all billable requests, including retries and failed calls that consumed usage. Separate fresh input tokens, cached input and output tokens. Provider-reported charges take precedence over a catalogue calculation; discounts, batch pricing, image quality and gateway fees can change the bill. Missing usage is an incomplete measurement, not evidence of a free request.
A new model can be tracked even when its list price is absent from the catalogue: send the provider’s reported costUsd. Otherwise the cost may be partial or estimated. Check the integration guide and gateway cost reporting before trusting a margin figure.
How much AI can a flat subscription afford?
For a chosen AI margin target, the AI budget is revenue × (1 − target margin). At $99 and a 60% target, that is $39.60. If the 60% target must also cover $5 of other variable costs, the AI budget drops to $34.60. A target is your planning assumption, not an industry benchmark or a guarantee.
Look at expensive individual customers as well as averages. A healthy blended margin can hide subscribers who cost more than they pay. Before blocking anyone, compare a cheaper model, a usage allowance, a top-up, and a different plan price. Measure answer quality and retention alongside spend.
Can a guardrail guarantee this margin?
No. MarginFuse evaluates recorded usage and an estimate of the next request. Concurrent calls and delayed reporting can overshoot a threshold. If the service is unavailable, the SDK allows the request to proceed. A downgrade may change answer quality; historical savings do not prove that customers will accept it. Read the protection limits before enabling live policies.
The paid plan makes economic sense when the savings or operational value you verify exceed its subscription price and integration cost. Start with a dry-run on real traffic, review individual decisions, then test a narrow policy. Compare the plans once the evidence supports paying.