Pricing when the AI does the work

The thesis of the week, well argued: software is no longer a tool that extends human capacity — it completes work itself. So pricing shifts from access to outcome. The number that explains everything else: AI companies see 50 to 60 percent gross margins against 80 to 90 for classic SaaS. Every model call costs real money — a cost line that did not exist in the SaaS era. Bessemer’s recommendation is refreshingly unromantic: hybrid. A base fee covering fixed costs, an outcome component for the upside.

The data behind the thesis, from 230 companies: hybrid is already the most common model at 37 percent, AI margins average 50 percent, and a third plan to introduce AI credits within six to twelve months. The finding that interests me most sits there almost in passing: AI spending is mostly cannibalising existing software budgets rather than unlocking new ones. Anyone who believes AI opens a fresh pot of money is in fact selling against the budget the customer already has.

The most usable toolkit in the debate: four models — per agent, per action, per workflow, per outcome — drawn from an analysis of 60-plus agent companies. The sharpest commercial observation in it: pricing per agent draws on the headcount budget, which is at least ten times larger than the tools budget. Outcome pricing has the strongest customer alignment but demands clean attribution. That is where most negotiations actually break down — not on the price, but on who measured the result.

The unglamorous part almost nobody reads amid the excitement: how do you actually book outcome pricing? Under ASC 606 revenue becomes variable consideration that must be estimated and constrained — operationally complex and volatile. And behind it the real question: do you owe continuous availability of the agent, or a specified quantity of successful outcomes? Without contractually clear success criteria, an elegant pricing model turns into an accounting problem. Pricing models rarely die in the market. They die in the back office.

And now the part Silicon Valley overlooks: manufacturing already ran this experiment. Rolls-Royce has been selling flight hours instead of engines since the 1960s. This study in the Journal of Business Research (2021) analysed 1,566 manufacturers and finds that providers of outcome-based services carry on average 4.4 percentage points higher gross margin — but the large ones earn less, unless they invest heavily in digital servitization. That is exactly the curve AI companies are driving onto: outcome pricing is attractive until it scales. Switching to outcomes today does not buy you a price tag, it buys you an operating model.