You Priced on Outcomes. You Can't Book the Revenue Yet.

A founder I work with sells an AI support agent. Last quarter she moved off per-seat pricing and onto per-resolved-ticket, four dollars for every ticket the agent closes without a human touching it. Sales loved it. Deals that had stalled for months closed in weeks, because the buyer only pays when the thing actually works. Then her controller came back from a call with the auditor. The revenue she thought she'd booked in Q2 wasn't all hers to book. The agent had resolved the tickets. The cash had cleared. A chunk of it still had to sit in deferred.

She changed her pricing and, without meaning to, changed when she was allowed to recognize a dollar.

Outcome pricing is variable consideration, and that comes with a brake

Deloitte put out a technology spotlight in June on exactly this, accounting for outcome-based pricing in an agentic AI product. The short version: when your fee is tied to a result the agent achieves, that fee is variable consideration under ASC 606. And variable consideration comes with a constraint. You can only recognize the amount where it's probable a significant reversal won't happen later.

Read that constraint slowly, because it's the whole problem. It isn't "book it when the outcome occurs." It's "book it only to the extent you're confident it sticks." If your definition of a resolved ticket includes a customer satisfaction check that lands three days later, or a clawback if the issue reopens inside a week, then the outcome isn't final when the agent finishes. Your revenue waits for the outcome to stop moving.

Your cost shows up first, your revenue waits for proof

Here's the part that hits the P&L. The compute cost lands the instant the agent runs. Every attempt burns tokens, resolved or not, refunded or not. But the revenue only clears the constraint once the result is locked. So cost leads and revenue lags, and they don't land in the same period.

For one month that's a rounding difference. Across a quarter where you're growing fast, it's a gross margin that reads worse than the business actually is, because you're expensing this month's effort against last month's recognized wins. Your cash can look fine while your recognized revenue trails what you collected.

Per-seat pricing let you book the whole contract on day one and spread it evenly. Per-outcome pricing makes you earn every dollar twice, once when the agent does the work and again when the result proves durable enough to keep. Most founders price the first event and forget the second one gates the revenue.

Write the definition of "success" before you write the price

The lever is the contract, not the accounting. Deloitte's guidance is blunt on this: if "success" isn't objective and measurable, you lose the practical expedients that would otherwise let you recognize as you invoice, and you're stuck estimating total variable consideration at inception. Vague success terms are the expensive ones.

So define the outcome tightly enough that it's final when the agent says it's final. Decide whether success resets each month or accrues across the whole contract, because a monthly reset lets you allocate revenue to the period it happened and a cumulative measure forces you to estimate the entire term up front. Do that work with your controller before the pricing goes on the order form, not after the auditor reads it.

Outcome pricing is a genuinely better way to sell an agent. It's also a promise that you'll get paid for results, and your books take that promise literally.