This calculator backs out the per-user price you need to charge from your real AI token cost and a target gross margin, then stress-tests that price against your heaviest users. Flat-rate plans quietly lose margin on power users who consume many times the average, and this tool shows exactly where. It is aimed at founders and product owners setting the price of an AI feature or an AI-native SaaS.
Loading calculator...
Pricing from a target margin is the easy part. The trap is a small share of power users whose token consumption dwarfs the average, turning a healthy blended margin into a loss on your most engaged customers. The calculator surfaces that before your billing does.
How to use it
The first section builds your cost of goods per user. Enter the average input and output tokens a user consumes per month, your input and output prices, and any other fixed cost per user such as infrastructure, support, or a vector database. Then set your target gross margin.
The power-user section stress-tests the result. The heavy user AI multiplier is how many times the average a power user consumes, and the share of heavy users is what fraction of your base they are. These two feed the blended price and the heavy-user margin check.
Get token-per-user figures from real usage if you have any. Take total monthly input and output tokens across all users and divide by user count for the averages, then look at your top few percent to set the multiplier.
The four tiles show cost of goods per user, the price needed for your target margin, the blended price accounting for power users, and the margin you would earn on a single heavy user at the base price. A negative heavy-user margin turns red.
Fields explained
Input tokens / user / mo – average monthly input tokens per user. Default 300000, step 1000.
Output tokens / user / mo – average monthly output tokens per user. Default 120000, step 1000.
Input price / 1M – your input rate. Default 0.5, step 0.01.
Output price / 1M – your output rate. Default 1.5, step 0.01.
Other fixed cost / user – non-token cost per user, such as infra and support. Default 0.5, step 0.01.
Target gross margin (%) – the margin you want on an average user. Default 70, step 1. Capped just below 100.
Heavy user AI multiplier – how many times the average a power user consumes. Default 5, step 0.5. Floored at 1.
Share of heavy users (%) – fraction of users who are heavy. Default 10, step 1.
Reading the results
| Result | What it means | How to act |
|---|---|---|
| COGS / user | AI token cost plus fixed cost per user | Your floor; price must clear it |
| Price needed | Per-user price for the target margin on an average user | Your starting list price |
| Blended price | Price that hits target across average and heavy users | Use if you cannot separate tiers |
| Margin on a heavy user | Margin a single heavy user earns at the base price | If negative, add caps or tiers |
At the defaults, AI token cost is $0.33 per user, cost of goods is $0.83 with the fixed cost added, and the price needed for a 70% margin is $2.77. Accounting for power users, the blended price rises to $3.21, and a single heavy user earns only a 22% margin at the $2.77 base price.
That heavy-user margin is the number to watch. A heavy user earns 22% margin, not the 70% target. They still make money here, but a higher multiplier or a bigger share pushes that figure toward zero and past it.
The blended price only protects you if heavy users stay at the assumed share and multiplier. If power users churn less and accumulate, or your multiplier is understated, the blend drifts and margin erodes over time. Recheck it as your usage data matures.
When the heavy-user margin turns negative, a flat price loses money on your most engaged customers, which is the worst group to lose money on. That is the signal to add usage caps, metered overages, or a dedicated high-usage tier.
The formula
Price is cost of goods grossed up by the target margin, and the heavy-user check reprices their higher cost:
aiCost = (inTok x inPrice + outTok x outPrice) / 1e6
cogs = aiCost + fixedPerUser
price = cogs / (1 - margin)
blendedAi = aiCost x (share x multiplier + (1 - share))
heavyCogs = aiCost x multiplier + fixedPerUser
heavyMargin = (1 - heavyCogs / price) x 100
Walk the defaults. AI cost is (300,000 x 0.5 plus 120,000 x 1.5) / 1e6 = $0.33. Cost of goods is $0.83. Price is 0.83 / (1 minus 0.70) = $2.77. Blended AI cost is 0.33 x (0.10 x 5 plus 0.90) = $0.462, so blended cost of goods is $0.962 and the blended price is $3.21. A heavy user costs 0.33 x 5 plus 0.50 = $2.15, giving a margin of (1 minus 2.15 / 2.77), or about 22%.
| Target margin | Price multiplier on COGS | Default price |
|---|---|---|
| 50% | 2.0x | $1.66 |
| 70% | 3.33x | $2.77 |
| 80% | 5.0x | $4.15 |
| 90% | 10.0x | $8.30 |
The price multiplier is 1 divided by (1 minus margin), so margin targets get expensive fast. Moving from a 70% to a 90% target more than triples the required price, from 3.33 times cost of goods to 10 times.
The heavy-user margin uses the base price on purpose, to show what happens if you charge everyone the same flat rate.
Worked examples
Light usage, high margin. 50,000 input and 20,000 output tokens, 0.5 / 1.5 prices, $0.20 fixed, 80% target, 3x multiplier at 5% share. AI cost is $0.055, cost of goods $0.255, and the price needed is $1.28. A heavy user costs $0.365 and still earns a 71% margin, so power users are not a threat at this usage.
Default assistant. The shipped inputs give a $0.83 cost of goods, a $2.77 price for 70% margin, a $3.21 blended price, and a 22% heavy-user margin. Power users are profitable but well below target.
Heavy usage, thin power-user margin. 800,000 input and 400,000 output tokens, 0.5 / 1.5, $1.00 fixed, 75% target, 7x multiplier at 20% share. AI cost is $1.00, cost of goods $2.00, and the price needed is $8.00. A 7x heavy user costs $8.00, so their margin is exactly 0%. A 7x power user hits 0% margin at the $8.00 price. The blended price of $12.80 is what actually protects the target.
Power user goes negative. Default usage but a 10x multiplier at 15% share. The base price stays $2.77, yet a heavy user costs 0.33 x 10 plus 0.50 = $3.80, giving a margin of about negative 37%. Each such user loses money on the flat plan, and the tool flags the tile red. A cap or overage is required.
Common mistakes
Pricing off the average only. The average user can look healthy while power users bleed margin. Always read the heavy-user tile, not just the price-needed figure.
Underestimating the multiplier. Power users often consume ten or twenty times the average, not five. A low multiplier hides the risk that shows up once real usage data arrives.
Do not launch a flat unlimited plan without checking the heavy-user margin. If it is negative, your most engaged users cost you money on every bill, and they are exactly the ones who stay and consume more. Add usage caps, metered overages, or a higher tier before, not after, they arrive.
Forgetting fixed costs. Infrastructure, support, and storage per user are real. Leaving the fixed cost at zero understates cost of goods and overstates every margin.
FAQ
Why is the blended price higher than the price needed?
The price needed covers an average user, while the blended price accounts for power users who cost more. Charging the blend keeps your target margin across the whole base.
At the defaults the blend is $3.21 against a $2.77 average-user price.
What multiplier should I use for heavy users?
Look at your top few percent of users by token consumption. Ratios of 5x to 20x over the average are common in AI products, and heavier for tools used all day.
Set the share to match how many users hit that level, since both feed the blended price.
What does a negative heavy-user margin mean?
Each heavy user costs more to serve than the base price collects, so you lose money on every one of them under a flat plan.
The fix is usage caps, metered overage charges, or a separate high-usage tier priced for their cost.
Does this include company overhead or acquisition cost?
No. It sizes unit economics at the per-user level: token cost, fixed per-user cost, and the price to hit a gross margin. It excludes fixed company overhead, customer acquisition cost, and churn.
Use it as an input to a fuller model, not as a full profitability picture.
How high can the target margin go?
The field is capped just below 100%, since a 100% margin implies an infinite price. Higher targets raise the required price sharply, following the multiplier table.
Most software targets 70% to 90%, but token-heavy AI products often run lower until usage is optimized.
Disclaimer
This is an educational estimate for planning, not financial or pricing advice. Results depend on the token usage, prices, fixed costs, and margin you enter, and on your provider’s current rates, all of which change. It sizes unit economics only and excludes overhead, acquisition cost, and churn.
The live calculator on this page is the source of truth for its exact fields. Confirm current token prices, measure real per-user consumption including your heaviest users, and validate pricing against a real cohort before setting or changing a plan.
