Lesson 1: Why AI Changes Pricing: The Law-Firm Case

The Running Case: An AI Assistant for Small Law Firms

In this course you will price one product from start to finish. The product is an AI assistant for law firms with 2 to 10 lawyers. It does three jobs:

• Summarise a contract into key terms and risks.
• Draft a client letter from a short brief.
• Prepare a research memo on a narrow legal question.

Every one of these tasks calls a language model, and every call costs money. This is the first fact that separates AI pricing from traditional SaaS. With classic software, serving one more user costs almost nothing. With AI, each task has a real bill.

In this level you will (1) build the base spreadsheet model, (2) learn the three forces that make AI pricing different, and (3) decide which force matters most for this product.

Task 1: Build the Base Spreadsheet Model

Open a blank spreadsheet. Do these steps in order.

1. Column A: Users per firm. Start with 5 lawyers.
2. Column B: Tasks per user per month. Estimate 120, which is about 6 tasks per working day.
3. Column C: Cost per task. This is what you pay the model provider for one task. Use $0.08 as a working assumption. You will refine it with real token counts in a later level.
4. Column D: Monthly cost per firm = A × B × C. Here: 5 × 120 × $0.08 = $48.
5. Column E: Monthly price per firm. Assume a placeholder of $50 per user, so 5 × $50 = $250. This is not a recommendation. It is a starting point to test.
6. Column F: Gross margin % = (E − D) ÷ E. Here: ($250 − $48) ÷ $250 = 80.8%.

Now add a second row for a heavy-use firm: the same 5 lawyers, but 600 tasks each per month.

• Cost: 5 × 600 × $0.08 = $240
• Price: still $250
• Margin: ($250 − $240) ÷ $250 = 4%

Do this now: add both rows and look at column F. Same price, same number of users, and a margin that falls from about 81% to 4%. In traditional SaaS the heavy firm would be nearly as profitable as the average one. Here it is close to break-even.

Note: column C covers model cost only. Hosting, support and other costs come later.

Task 2: Learn the Three Forces That Change AI Pricing

Force 1: Marginal cost per task

Each extra task adds cost. Effects on pricing:

• Margin depends on how much customers use the product, not just how many customers you have.
• A small share of heavy users can consume most of your costs.
• Margins are often thinner than founders expect. Bessemer's State of AI 2025 report put the average gross margin at about 60% for steadily growing AI companies ("Shooting Stars") and about 25%, often negative, for hypergrowth ones ("Supernovas").

Force 2: Falling model prices

The price of a given level of model capability has dropped very quickly:

• Andreessen Horowitz (a16z) estimated in November 2024 that, for an LLM of equivalent performance, cost was falling about 10x per year. They called this "LLMflation".
• Epoch AI found that the price of reaching a fixed benchmark score fell between 9x and 900x per year, depending on the benchmark and threshold.

What this means for pricing:

• Your $0.08 cost per task may not hold for long. Don't lock in a price that only works at today's costs, and don't bet that costs will fall on schedule.
• Lower costs do not always reach your bill. Customers want the newest, most capable models, and cheaper tasks lead people to run more of them.
• If you price close to cost, competitors can undercut you as costs fall. Prices tied to value hold up better.

Force 3: Value tied to outcomes

AI often completes a piece of work rather than just providing a tool. That makes it possible to price on the result.

• Law-firm example: say a lawyer takes 1 hour to summarise a contract by hand and the assistant cuts that to 10 minutes. The value is about 50 minutes of lawyer time. That is far more than the $0.08 the task costs you.
• Public illustration, not a template: as of October 2026, Intercom prices its Fin AI agent at $0.99 per outcome, meaning per customer conversation resolved, not per seat.
• The hard part is agreeing what counts as an outcome and measuring it reliably.

Task 3: Decide Which Force Matters Most

Work through these questions for the law-firm assistant:

1. Cost: Does usage vary widely between firms? (Yes: 120 vs 600 tasks changed margin from 81% to 4%.)
2. Price trend: Would falling model prices change your price in the next 12 months, or mainly your margin?
3. Value: Can you name a clear, measurable result, such as a contract summarised or a letter drafted, that firms would pay for?

Do this now: add a notes cell to your spreadsheet. Write which force matters most and one sentence on why.

Worked answer for this case: marginal cost per task matters most at launch. A flat price with uneven usage can wipe out the margin on your best customers. Value per outcome is the biggest opportunity, and you will pick it up when you choose a value metric. Falling prices are a tailwind to plan around, not something to build the price on.

Your answer may differ. What matters is that you write down the reasoning so you can check it later.

Key Takeaways

• The base model has six columns: users, tasks per user per month, cost per task, monthly cost, monthly price and gross margin %.
• Monthly cost = users × tasks per user × cost per task. Gross margin % = (price − cost) ÷ price.
• AI tasks have a real marginal cost, so heavy users can erase margin at a flat price.
• Model prices for a fixed capability are falling fast, but your costs may not fall at the same rate.
• AI often delivers outcomes, which opens up pricing on value rather than access.
• Write down which force matters most for your product and why.