AgentCostAI

AI SaaS Unit Economics Calculator

Model how LLM requests, agent activity and other variable costs affect the economics of your AI SaaS product. Enter your subscription revenue and usage assumptions to estimate cost per active user, margin sensitivity and the minimum price needed to cover variable delivery costs.

Interactive LLM cost calculator

Estimate monthly model cost from request volume, token usage, and your provider's per-million-token prices. Enter the prices from the provider you actually use.

What this calculator helps you measure

AI products can look profitable at the subscription level while losing margin through model calls, multi-step agent runs and usage-heavy customers. This calculator connects monthly revenue to the workload required to serve active users. Use it to estimate total AI spend, variable cost per user, gross or contribution margin, and break-even revenue per active user. Scenario comparisons also show what happens when usage, model cost or customer volume changes.

Inputs and how to choose realistic values

Use monthly figures from the same period. Subscription revenue should be recognized revenue for the users included in the calculation, not annual contract value unless you divide it by 12. Active users should represent customers or seats that generated meaningful product activity during that month. Requests per user is the average number of billable model requests attributable to each active user. Model cost should be an average cost per request, including input and output usage where relevant. Non-AI variable costs can include infrastructure, transaction fees or usage-based services that rise with customer activity. Keep fixed expenses such as salaries and rent separate unless the calculator explicitly provides a fixed-cost field.

How agent runs relate to model requests

An agent run may trigger one model request or fan out into several calls for planning, tool selection, retries and final generation. Avoid counting the same workload twice. If one user averages 40 agent runs and each run produces two billable model calls, use 80 requests per user. If you already have total request volume from provider logs, divide it by active users and use that result directly. For mixed agents, calculate a weighted average based on the share and cost of each workflow rather than using the most expensive run as the default.

Formulas used for the unit economics estimates

Let U be active users, R be average model requests per user, C be average model cost per request, S be monthly subscription revenue and N be monthly non-AI variable costs. Estimated monthly AI spend is U × R × C. Total variable delivery cost is estimated AI spend + N. Variable cost per active user is total variable delivery cost ÷ U. Revenue per active user is S ÷ U. Simplified contribution margin is (S − total variable delivery cost) ÷ S × 100. The variable-cost break-even price per user is total variable delivery cost ÷ U. If fixed costs must also be recovered, use (total variable delivery cost + monthly fixed costs) ÷ U outside the calculator.

Gross margin versus contribution margin

The difference depends on how your company classifies costs. Model usage and delivery infrastructure are commonly treated as cost of revenue, making gross margin equal to (revenue − cost of revenue) ÷ revenue. Contribution margin goes further by subtracting other variable expenses associated with serving or acquiring usage. Because accounting policies differ, treat the displayed result as an operating model rather than a financial statement. For board, tax or statutory reporting, map each input to your company’s approved cost classifications.

How to interpret the results

Cost per active user shows the delivery burden behind each paying or included user. Compare it with revenue per active user to see whether pricing leaves enough room for support, product development and fixed overhead. A positive contribution margin means revenue exceeds the variable costs entered; it does not mean the business is profitable overall. Break-even price is a floor under the selected assumptions, not necessarily the right commercial price. Pricing also needs to account for fixed costs, desired profit, unused plan capacity and the risk that a small group of customers generates disproportionate AI usage.

Run scenarios before changing pricing or model selection

Test a baseline, a high-usage case and an efficiency case. In the high-usage case, increase requests per user or agent fan-out without changing subscription revenue. In the efficiency case, lower average model cost to represent routing suitable work to a more cost-effective configured model. Also test growth carefully: adding users improves the picture only when revenue and costs scale consistently. If a flat-price plan permits unlimited usage, model active-user growth and request growth separately because request volume may rise faster than subscriptions.

Worked example

Consider an illustrative product with €6,000 in monthly subscription revenue, 200 active users, 80 model requests per user, an average model cost of €0.012 per request and €400 in other monthly variable costs. Estimated AI spend is 200 × 80 × €0.012 = €192. Total variable delivery cost is €592, or €2.96 per active user. Revenue per active user is €30, and the simplified contribution margin is about 90.1%. If requests per user double while revenue stays constant, AI spend rises to €384, variable cost per user becomes €3.92 and the margin falls to about 86.9%. The example is illustrative; actual results depend on provider pricing, token mix, retries, caching, routing and workload design.

Frequently asked questions

What is a good gross margin for an AI SaaS product?

There is no universal target. The appropriate margin depends on growth stage, support requirements, infrastructure, model intensity and pricing strategy. Compare your result with your own operating plan and test whether it remains sustainable under heavier usage.

How do I calculate average LLM cost per request?

Divide model spend for a representative period by the number of billable model requests in that period. Use a period that captures your actual mix of prompts, outputs and models. If workloads differ substantially, calculate separate scenarios or a weighted average.

Should free users be included as active users?

Include them if the revenue and cost inputs cover the same population. Free users can materially increase AI spend without adding subscription revenue, so separating paid and free cohorts usually produces more actionable unit economics.

Does break-even price include salaries and other fixed costs?

Not when the calculation uses only AI and non-AI variable costs. To estimate full operating break-even, add the monthly fixed costs you need to recover before dividing by active users or paying accounts.

What are the main limitations of this calculation?

Average values can hide expensive customers and workflows. The estimate may also omit retries, failed calls, provider price changes, discounts, taxes, support time, fixed infrastructure and annual-plan revenue timing. Validate assumptions against actual usage and rerun scenarios regularly.

Turn a one-time model into ongoing AI cost control

Once you know the cost per active user and your acceptable margin, AgentCost can help operationalize those limits. Route AI requests through an OpenAI-compatible endpoint, apply daily or monthly budgets, and review requests, spend and estimated routing savings by client or agent. Actual savings depend on workload, provider pricing, model choice and routing configuration.

Explore AgentCost →