AI Agency Project Pricing & Margin Calculator

What this AI agency pricing calculator helps you decide

An AI engagement can look profitable when scoped around setup hours alone, yet lose margin after thousands of production agent runs, repeated revisions or unnecessary premium-model calls. This calculator brings those costs into the quote. Use it to answer four practical questions: Is the proposed client price profitable under the expected workload? What price merely breaks even? How much room remains for LLM usage after labor and infrastructure? What model-spend ceiling should the team enforce during delivery? For a one-off project, enter totals for the complete delivery period. For a retainer, use one consistent billing period—usually a month—for revenue, runs, labor and every other input. Do not combine monthly revenue with annual infrastructure or lifetime run volume.

Inputs and how to estimate them

Project revenue is the fixed fee or retainer amount you plan to charge, excluding taxes unless taxes are genuinely part of your available delivery budget. Expected agent runs are the number of billable-period workflow executions. Define a run consistently. A single customer action may trigger several model requests, so historical request logs are usually more reliable than user counts. LLM cost per run should include all model calls typically required to complete one run. For a workflow that makes three calls costing €0.01, €0.02 and €0.03 on average, use €0.06 per run—not the cost of only the first call. Labor should include discovery, implementation, prompt work, testing, deployment, monitoring and account management. Use loaded internal labor cost rather than the employee’s take-home pay. Infrastructure can include hosting, databases, queues, observability and workflow platforms allocated to the project. Revision allowance reserves capacity for changes already included in the scope. Contingency covers less predictable delivery risk, such as longer outputs, retries, traffic variation or extra debugging. Keep these separate so expected revisions are not disguised as exceptional risk.

Formulas behind the estimate

Expected model spend = expected agent runs × average LLM cost per run. Pre-contingency delivery cost = model spend + labor + infrastructure + revision allowance. If contingency is entered as a percentage, contingency cost = pre-contingency delivery cost × contingency rate. Total estimated cost = pre-contingency delivery cost + contingency cost. If the calculator accepts contingency as a fixed amount, add that amount directly instead. Gross profit = project revenue − total estimated cost. Gross margin percentage = gross profit ÷ project revenue × 100. Break-even price equals total estimated cost: it covers the included costs but produces no gross profit. When contingency is a fixed amount, the break-even maximum model spend is revenue − labor − infrastructure − revision allowance − contingency. If contingency is a percentage q applied to the whole cost base, the equivalent formula is revenue ÷ (1 + q) − labor − infrastructure − revision allowance. This is an absolute break-even limit, not necessarily a prudent operating budget.

Worked example: checking a fixed-fee automation project

Suppose an agency is considering a €10,000 project with 20,000 expected agent runs. Average model cost is €0.03 per run, labor is €4,000, infrastructure is €400 and the revision allowance is €700. A 10% contingency is applied to the pre-contingency cost. Expected model spend is 20,000 × €0.03 = €600. Pre-contingency cost is €600 + €4,000 + €400 + €700 = €5,700. Contingency is €570, making total estimated cost and break-even price €6,270. At a €10,000 fee, estimated gross profit is €3,730 and gross margin is 37.3%. The model spend could theoretically rise to about €3,990 before the project breaks even: €10,000 ÷ 1.10 − €5,100. That figure is too high to use as the normal operating limit because it allows the entire margin to disappear. A safer ceiling should reflect the margin the agency intends to preserve.

Set a model budget from your target margin

The calculator’s break-even model allowance answers, “How much could we spend before losing money?” Agencies should also ask, “How much can we spend while preserving the margin required by the business?” For a target margin t and percentage contingency q, a useful planning formula is: maximum model spend at target margin = [revenue × (1 − t) ÷ (1 + q)] − labor − infrastructure − revision allowance. Using the example above with a 30% target margin, the model allowance is approximately €1,264: €10,000 × 0.70 ÷ 1.10 − €5,100. Expected spend of €600 is below that allowance, leaving room for usage variation without immediately sacrificing the target margin. If the result is negative, the engagement cannot achieve the target margin even with zero model usage. Increase the quote, reduce non-model delivery costs, narrow the included scope or change the target. Do not assume cheaper model routing can repair a project whose labor and revision commitments already exceed its cost capacity.

How to interpret the client-ready quote

Use the quote summary as an internal pricing check and as the basis for a clear commercial proposal. The client-facing version should state the fee, billing period, included usage assumptions, revision scope and what happens when usage exceeds the allowance. It does not need to reveal internal salaries or every supplier cost. For fixed-price work, define the expected run volume and a change-control rule. For retainers, specify the monthly service scope and whether unused usage carries forward. If demand is uncertain, offer a base fee with an explicit usage tier instead of silently accepting unlimited model spend. A positive gross margin does not automatically make a quote attractive. Compare it with your required contribution toward sales, administration and other overhead. Conversely, a high projected margin based on unrealistically low run volume is not a safe quote. Stress-test expected runs, cost per run and revision effort before approval.

Assumptions and limitations

The result is an estimate, not an accounting statement or a guarantee of profitability. Average cost per run can change with prompt length, output length, tool calls, retries, provider pricing, model selection and routing rules. Run volume can also differ materially from the sales forecast. Gross margin includes only the costs entered. Unless you add them to labor, infrastructure or an allowance, the result may omit payment processing, sales commissions, support, taxes, legal work, insurance, refunds, currency conversion and shared company overhead. It also does not value cash-flow timing or the risk of late payment. Recalculate when scope, traffic, provider pricing or model behavior changes. For uncertain workloads, test a base case, a likely high-usage case and a severe case. A quote is more defensible when it remains acceptable under a realistic high-usage scenario rather than only under the best case.

Turn the pricing limit into an operating control

A spreadsheet limit protects margin only if delivery systems follow it. After choosing a safe model-spend ceiling, translate it into a daily or monthly budget for the relevant client or agent. AgentCost sits between an AI application and configured model providers. Agencies can use dedicated client API keys, apply daily and monthly budget limits before requests are forwarded, route work toward cost-effective configured models, and review requests, spend and estimated savings. Configured limits can reject requests after the ceiling is reached. Keep a buffer between forecast spend and the hard limit so normal variation does not interrupt a client workflow. Review actual usage against the assumptions in the quote, then update future pricing with observed run volume and per-agent cost data.

Frequently asked questions

What is a good gross margin for an AI agency project?

There is no universal target. The appropriate margin depends on sales cost, overhead, delivery risk, payment terms and how much ongoing support is included. Set an internal target before quoting, then verify that the project remains above it under a realistic high-usage scenario.

Should I price an AI project per run or as a fixed fee?

Use a fixed fee when scope and usage are predictable and the client values a defined outcome. Use usage tiers or overage terms when run volume is controlled by the client or can vary widely. A hybrid structure can combine a base implementation or retainer fee with an included run allowance.

How do I calculate average LLM cost per agent run?

Add the average cost of every model request, retry and supporting call required for one completed workflow. Divide observed total model spend by completed runs when production history exists. If not, test representative inputs and include variation in prompt length, output length and failure rates.

Is maximum model spend the same as the recommended budget?

No. A break-even maximum allows model costs to consume all remaining gross profit. The recommended operating budget should be lower and should preserve your target margin plus a buffer for usage variation.

How often should an agency update the calculation?

Recalculate before quoting, after material scope changes, when provider or model pricing changes, and after enough production usage exists to replace assumptions with observed data. Monthly review is practical for retainers with variable traffic.