Your $20 AI Plan Has a Power-User Problem

A plan can look healthy for its normal user and still miss its margin target. The question is not whether one prompt is inexpensive. It is how much accepted work the plan promises, and who uses that promise most intensively.

Open this example in AI Plan Stress Test. Its default assumptions match the calculation below. Replace them with your own numbers before making a pricing decision.

Start with the budget the price can support

This example uses $20 in monthly net revenue per user, after discounts and taxes, a target gross margin of 70%, and $2 of other variable costs. Payment processing, support, or infrastructure belongs in that $2 only to the extent you actually included it. Fixed operating costs are outside this gross-margin calculation.

The entire variable-cost allowance is $20 multiplied by 30%, or $6. Subtracting the other $2 leaves $4 for accepted AI actions. An accepted action costs $0.03 in this scenario, including its failed attempts, tools, and review. Dividing $4 by $0.03 gives 133.33 actions, so an integer included allowance rounds down to 133.

const revenue = 20;
const targetMargin = 0.70;
const otherCost = 2;
const actionCost = 0.03;
const actionBudget = revenue * (1 - targetMargin) - otherCost;
const includedActions = Math.floor(actionBudget / actionCost);
console.log(includedActions); // 133

This allowance is a cost constraint, not a recommendation to impose exactly that limit. Your product may need more generous usage to be useful. The calculation shows what must change elsewhere when you offer it.

Five percent of users can change the average

Assume normal users complete 100 actions per month while heavy users complete 1,000. The normal user costs $5: $3 of actions plus $2 of other costs. Their gross margin is 75%. The heavy user costs $32 and produces a negative $12 gross profit at the same price.

At a 5% heavy-user share, the target is missed by $0.35 per user. Holding the usage mix and costs fixed, the net revenue required for a 70% gross margin is $6.35 divided by 30%, or about $21.17. Alternatively, a lower action cost or a different included allowance could restore the target.

The share that exactly reaches the target is about 3.70% under these assumptions. That threshold is a sensitivity result, not a prediction that your product will attract that many heavy users.

Change a product rule, then recalculate

A hard cap, an overage price, a more expensive high-usage tier, and a cheaper implementation solve different problems. A cap limits included work. An overage can finance additional work but changes the customer’s bill. Routing can lower cost only when the cheaper path meets the same acceptance criteria.

Do not silently assume all four improvements at once. Start with one change, keep the others fixed, and inspect the resulting margin. Otherwise the model can hide how much of the apparent improvement depends on an untested quality or behavior assumption.

The Cost per Accepted Task calculator helps establish the action-cost input. Raw per-call cost can understate that input when several attempts are required to produce one usable result.

What this example cannot tell you

Two user groups are not a measured distribution, and they do not produce a valid P95 forecast. Real workloads can also have multiple action types with different costs. Segment them or calculate a defensible blended accepted-action cost before using a single input.

A positive gross margin does not establish company profitability. Salaries, product development, acquisition costs, and other fixed expenses still need funding. Conversely, one unprofitable heavy user does not prove the whole plan is unprofitable; the usage mix determines the average.

Sources and method

This is an original, modeled example using the default AI Plan Stress Test inputs. No customer sample, provider quota, or measured success rate is claimed. Calculations use weighted averages and the gross-margin identity; table values are rounded only for display.

• AI Plan Stress Test and editable assumptions
• Cost per Accepted Task methodology
• Per-User Gross Margin calculator