Lesson 2: Finding the Value Metric

What a value metric is

A value metric is the unit you charge for: per seat, per document, per matter, per contract. Price list amounts are a separate decision. The metric decides how your revenue grows as a customer uses the product more.

In Level 1 you saw that every task the law-firm assistant runs has a real inference cost. That gives the metric two jobs:

1. Track value. When a firm gets more out of the assistant, it should pay more, and it should feel that this is fair.
2. Track cost to serve. When a firm makes you spend more on inference, your revenue should rise too. If it doesn't, your heaviest users eat your margin.

You want a metric where value, revenue and cost all move together.

Step 1: List your candidate metrics

Do this: Write down every unit you could plausibly charge for. For the law-firm assistant, start with:

• Seats: lawyers and paralegals with a login
• Documents reviewed: files the assistant reads and summarises or checks
• Matters: client cases or files the assistant is used on
• Contracts drafted: first drafts the assistant produces

Add two "tempting" options so you can practise rejecting them:

• Tokens processed: the raw unit your model provider bills you for
• Hours saved: the outcome the firm actually cares about

Step 2: Score each metric on four criteria

Do this: Score each candidate from 1 (poor) to 5 (strong) on:

1. Value alignment: does more of this unit mean more value for the firm?
2. Predictability: can a managing partner estimate next month's bill before it arrives?
3. Ease of measurement: can you count it reliably, with a definition both sides agree on?
4. Cost tracking: does your inference cost rise and fall with this unit?

Here is the scored sheet for the case. The scores are judgement calls to argue over, not facts:

Why some scores land where they do:

• Seats are easy to predict and count. But one lawyer who runs 500 document reviews costs you far more than one who runs 5, and both pay the same. Cost tracking is weak. (Level 3 covers this in depth.)
• Matters fit how firms already think and bill. But a simple probate file and a complex litigation matter can differ in cost by a factor of a hundred, so cost tracking is weak.
• Contracts drafted are valuable, but they cover only part of what the assistant does, and "drafted" is hard to define. Does editing a template count? A redraft?

Step 3: Apply the rejection gates

Totals can hide a fatal flaw. Do this: Before comparing totals, reject any metric that fails either gate:

• Gate A, predictability: If customers cannot estimate their bill, reject the metric. A partner has no idea how many tokens a lease review uses, so tokens fail, even though they track your cost perfectly.
• Gate B, measurement: If you cannot count the unit reliably and defend the count, reject it. Hours saved is the best value signal, but you cannot observe how long a task would have taken. Every invoice would turn into a dispute.

A metric that tracks only your cost is a cost-plus meter. A metric that tracks only value but cannot be measured is a promise you cannot bill.

Step 4: Tighten the definition of the winner

Even a simple unit needs a written definition. Two dated public examples, as illustrations only:

• Slack's fair billing policy (checked October 2026) treats a member who hasn't used Slack in over 28 days as inactive and credits the account. So even "a seat" needed a definition.
• Intercom's pricing page (checked October 2026) charges $0.99 per outcome for its Fin AI agent. It lists exactly what counts as an outcome and says a conversation is charged only once.

Do this: Define "document" for the assistant so it is countable and roughly cost-consistent:

1. One uploaded file counts as one document, up to a page limit (for example, 50 pages).
2. Longer files count as one document per started 50-page block.
3. Re-running a review on the same unchanged file within a set window is not counted again.
4. Show the running count in the product, so firms can see it.

The page band fixes the measurement problem: a 2-page letter and a 300-page lease no longer count the same.

Step 5: Add the metric to your model

Do this: In the spreadsheet from Level 1, add three rows per customer profile:

1. Documents per month: for example, a five-lawyer firm at 400 (an assumption to test with customers).
2. Inference cost per document: carried over from your Level 1 cost-per-task figure.
3. Monthly cost to serve = row 1 × row 2.

Now every change in usage moves cost and the future revenue line together. You'll set the actual price per document in later levels.

Your decision

Choose your primary value metric. For the case, pick documents reviewed, with matters kept as the backup in case firms insist on per-matter budgeting. Write down:

• the metric and its exact definition
• the gate each rejected metric failed
• which scores you are least sure of, to test in customer conversations

Key takeaways

• The value metric is the unit you charge for. It should rise with customer value and with your cost to serve.
• Score candidates on value alignment, predictability, ease of measurement and cost tracking.
• Reject any metric customers cannot predict (tokens) or you cannot measure reliably (hours saved), whatever its total.
• Seats and matters are predictable but track AI cost poorly.
• Write a precise definition of the unit and show customers the count.