FOR AI APPS WITH METERED COSTS
AI unit economics
calculator
See what one user actually costs you — including your heaviest ones — before you commit to a price.
What AI unit economics means
AI unit economics is the revenue and cost attached to one user of an AI product. Unlike classic software, a metered AI product pays for every request, so cost scales with usage rather than staying flat per seat. That is why a flat unlimited plan can look profitable on the average user and still lose money on the heaviest ones. This calculator models API cost, the heavy-user tail, payment fees and refunds, then reports your real gross margin, break-even point and the minimum price for a target margin.
Where the money goes
| Line | Per month | Per paying user |
|---|
Same numbers, every payment provider
| Provider | Fees | Gross profit | Margin | Min. price |
|---|
Merchant-of-record providers charge more and handle VAT and sales tax for you. The cheaper card rate is not automatically the cheaper option once tax work is included.
How it works
- 1
Enter your usage
Requests per user, tokens per request, and your model's price per million tokens.
- 2
Set the heavy-user multiplier
Your heaviest 10% of users rarely consume the average. Set the ratio and see whether they are still profitable.
- 3
Read the price floor
Break-even price, break-even users, and the minimum price that still hits your target margin.
METHOD / CHECKED 24 OCTOBER 2026
How the numbers are calculated
Cost is driven by tokens, not by seats. Every request you serve is metered by your provider, so a flat price can be profitable on the average user and still lose money on the tail.
API cost per user
input cost = requests × input tokens ÷ 1,000,000 × input price
output cost = requests × output tokens ÷ 1,000,000 × output price
Heavy tail
The top 10% of paying users consume h times the average. The remaining 90% is solved so the population average stays fixed:
light = average × (1 − 0.1h) ÷ 0.9
Refunds and fees
Refunds reduce revenue. They do not return the tokens you already paid for, and processing fees are usually not refunded either. Both cost lines are left intact.
net revenue = price × users × (1 − refund rate)
Price floor
The minimum price for a target gross margin solves for price with fees and refunds inside the equation, not added afterwards:
price = (api cost + fixed fee) ÷ ((1 − r)(1 − target) − fee%)
Reproducible example
With the default inputs — $19/month, 1,000 paying users, 500 free users, 200 requests each, 800 input and 400 output tokens per request, $3 and $15 per million tokens, heavy multiplier 5, Stripe at 2.9% + $0.30, 3% refunds, $300 fixed costs:
| Metric | Result |
|---|---|
| Input tokens per paying user per month | 160,000 |
| Output tokens per paying user per month | 80,000 |
| API cost per average paying user | $1.68 |
| API cost per heavy user (top 10%) | $8.40 |
| API cost per light user (remaining 90%) | $0.9333 |
| Total API cost, including free users at 30% usage | $1,932.00 |
| Net revenue after 3% refunds | $18,430.00 |
| Payment fees | $851.00 |
| Gross profit | $15,647.00 |
| Gross margin | 84.90% |
| Net profit after fixed costs | $15,347.00 |
| Break-even paying users | 19 |
| Break-even price at 1,000 users | $2.42 |
| Usage multiplier where one paying user stops being profitable | 10.46× |
These are arithmetic results, not an audit of your business. Real token usage runs higher than estimates once system prompts, conversation history, retrieval context and retries are counted. Model prices change. Check your own invoice.
Method version 0.1 · free users modelled at 30% of paying-user usage by default.
Questions people ask
Why can't I just use a normal SaaS margin calculator?
A normal SaaS has near-zero marginal cost per user. A metered AI product pays on every request, so cost scales with usage. That difference is what makes a flat unlimited plan dangerous.
What is the heavy-user multiplier, and how do I pick it?
It is the ratio between what your heaviest 10% of users consume and what an average user consumes. If you have no telemetry yet, 3x to 5x is a common starting range; 10x is the pessimistic bound this tool allows. Replace it with your own percentile data as soon as you have it.
Why does a refund hurt more than I expect?
Because the tokens were already paid for and the processing fee is generally not returned. A 3% refund rate removes 3% of revenue and none of the cost.
Are the built-in model prices correct?
No. They are illustrative reference values with a checked date. Public sources disagree with each other and providers change prices often. Every price field is editable — use your own invoice.
Is my data uploaded?
No. Everything runs in your browser. There are no accounts, no analytics, no cookies and no upload endpoint. Reloading the page clears everything.
Can I save or share my scenario?
Yes. Every input is written into the page address, so copying the URL from the address bar saves your scenario, and a Copy link button puts it on your clipboard. Your last scenario is also restored automatically on your next visit.