AI Feature Margin Calculator

Price, payment fees, inference and infrastructure into a gross margin — on a typical user, on a heavy user, and blended.

Price per user per monthWhat you charge, before fees.

$ / user / mo

Payment processingCard fees and platform cuts, as a share of revenue.

AI cost, typical userPer month. The cost-per-user calculator produces this figure if you do not have it.

$ / user / mo

Other cost of goodsHosting, storage, egress, per-seat vendor licences.

$ / user / mo

Heavy usersShare of paying users on far above-typical volume.

Heavy user multiplierTheir AI cost as a multiple of a typical user's.

Paying users

users

Blended gross margin

71.6%

78.5% on a typical user, -59.0% on a heavy one. A user stops being profitable above 7.28× typical usage.

Revenue per user: $20.00
After payment fees (3.0%): $19.40
AI cost — typical user: $2.50
AI cost — heavy user: $30.00
AI cost — blended: $3.88
Other cost of goods: $1.20
Gross profit per user (blended): $14.32
Margin — typical user: 78.5%
Margin — heavy user: -59.0%
Usage multiple at which a user breaks even: 7.28×
Heavy-user share at which blended margin hits zero: 57.1%
Monthly gross profit: $143,250
Annual gross profit: $1,719,000

What this assumes: gross margin only — no salaries, no marketing, no support, so the number here is a ceiling and not a profit; payment fees are treated as cost of goods, which is the conservative reading and differs from how some finance teams book them; margins are expressed against list price rather than net revenue, so discounts and annual plans need the price field adjusted rather than a separate line; two usage populations only, which understates a real long tail; and every AI cost is assumed to scale with usage, when in practice a share of it is fixed and improves with volume.

Software margins are a habit, and AI breaks the habit. A seat that costs nothing to serve can be sold at any price above zero and still make eighty per cent; a seat with a variable inference cost attached has a floor, and the floor moves with how much the customer uses the thing you sold them. The consequence is that the interesting figure is not the margin, it is the usage multiple at which the margin disappears — the row above that reads "a user stops being profitable above N× typical usage". If N is under 10, you are one enthusiastic cohort away from selling dollars for ninety cents.

The blended margin hides that completely. It averages a large majority of customers who barely use the feature against a minority who use it constantly, and it stays healthy right up until the mix shifts. That shift is not a risk, it is the expected outcome of the feature working: the people who get value from it use it more. The heavy-user share at which the blended margin hits zero is therefore a forecast, not a worst case, and it is worth watching monthly against your actual distribution.

The fixes are the ordinary ones and they are all easier before launch than after. Include an allowance in the price and meter visibly beyond it. Route the heavy tail to a cheaper model where quality permits. Cache aggressively, because a cache hit is pure margin. And price the feature as a separate line if its cost behaves nothing like the rest of the product. What this calculator does not tell you is whether the feature earns its keep in retention — a negative-margin feature that halves churn can still be the best line on the sheet, and that is a judgement no arithmetic here will make for you.

AI Feature Margin Calculator

The reasoning behind this

Related tools