Your Success Rate Is a Margin Line Now

A founder I work with sells an AI support agent on outcome pricing. Six dollars for every ticket the agent closes on its own, nothing when it hands off to a human. Buyers love the deal because they only pay for work that got done. He loved it too. Then came the quarter his revenue slipped a fifth and his gross profit fell almost a third, on the same ticket volume and the same price. He asked the bookkeeper what changed. Nothing in the books changed. The number that moved wasn't in the books at all.

You priced the wins, but you pay for the tries

Outcome pricing quietly splits your P&L in two. Revenue comes from successes. Cost comes from attempts. Those aren't the same base, and the gap between them is the thing you can't see.

Every ticket burns compute the moment the agent picks it up. It calls the model, pulls context, tries a few tool calls, sometimes retries. Whether that ends in a clean resolution or a handoff to a human, you've already spent roughly the same tokens. You don't find out a ticket is going to fail until the agent has finished failing at it. So you pay to attempt every ticket and you get paid for only the ones that land.

Which means one ratio sets your margin: how often the agent wins. And that ratio doesn't sit anywhere near your finance system.

The number that moved lives in a dashboard finance never opens

Run the founder's math. Compute to attempt a ticket runs about $1.30. At a 75% resolution rate, a thousand tickets cost $1,300 to serve and produce 750 wins at $6, so $4,500 in revenue against $1,300 in cost. Gross profit of $3,200.

Now the rate drifts to 60%. A model update made the agent more cautious about closing, or a customer routed a harder queue into it. Same thousand tickets, same $1,300 of compute, but now 600 wins. Revenue drops to $3,600. Gross profit falls to $2,300.

Revenue down 20%. Gross profit down 28%. Because the cost is fixed against attempts while the revenue floats on successes, every point of resolution rate you lose comes straight out of margin, amplified. The founder never cut price and never lost a customer. His unit economics moved anyway, driven by a success rate that lived in the product team's dashboard and never once appeared in the close.

Under outcome pricing, resolution rate is a gross margin input. If it isn't on the same page as revenue and compute cost, your margin is being set by a number nobody in finance is watching.

Put the rate where the money is

Start by measuring cost per attempt and resolution rate in the same place you measure revenue, monthly, side by side. A two-point drop in resolution should read as a margin event, not a product footnote. When the product team ships a model change, the finance question isn't whether accuracy improved. It's what the change did to the win rate you're billing against.

Then price with a floor. If a customer's mix is dragging your resolution rate below the point where the economics work, that account isn't a good outcome deal, and a per-attempt minimum or a blended plan protects you. The founder now reviews resolution rate by customer every month, the same way he reviews revenue by customer.

You agreed to get paid only when the agent wins. Fine. Just remember you're funding every time it loses, and the scoreboard for that isn't in your accounting system.