Software vendors are moving away from fixed subscription fees and billing by usage or by results instead. According to The Information, OpenAI is now doing the same.

Over the past few months the company has started giving some large customers the option to pay only once its AI actually finishes a task — handling a customer support request, for instance. The report cites a person with direct knowledge of the arrangements. The move had not been public before, and an OpenAI spokesperson declined to comment.

Startups pushed the model first

OpenAI is not inventing something here; it is adopting the model smaller companies pushed hardest: outcome-based pricing, where the customer pays only when the AI works.

  • Sierra and Fin charge only for tasks the AI completes without human involvement. Fin is currently being acquired by Salesforce for $3.6 billion.
  • Cognition promises corporate customers credits of up to $10 million if the software fails to deliver results worth at least the price they paid.
  • Older vendors like Adobe, HubSpot and Zendesk are heading the same way.

What is squeezing the subscription

There is pressure on AI companies to make this move. The products are expensive to run and, according to The Information, they have not so far accelerated revenue growth at software companies.

At the same time, customers' IT budgets are strained by competing tools. Anthropic's Claude recently pushed harder toward billing by usage, and that competes for the same budget.

Salesforce now lets companies choose how they pay for its own AI, Agentforce, including individually negotiated contracts tied to how much the AI lifts revenue or cuts costs. CEO Marc Benioff said at an investor conference that customers want to buy and price in different ways.

The logic behind the shift: a flat subscription made sense when software cost the vendor nothing extra per use. Opening a spreadsheet program ten times as often did not cost ten times more to serve. With AI it is the reverse — every request burns real compute. Keeping price detached from usage means heavy users are subsidised by light ones.

The hard part: who gets credited

The real difficulty in the model is not technical but accounting. Without clear rules it is tough to prove that good results really came from the AI.

What does it mean for a support request to be "resolved by the AI"? That the user considered it solved and left, or that they never came back? If sales rose, how much of that belongs to the assistant and how much to the campaign? A flat subscription never raised these questions, because the invoice did not depend on the outcome.

Outcome pricing therefore imposes a new burden on both sides: the vendor has to measure success, and the customer has to audit that measurement. If the model spreads, the industry's next argument will not be about which model is better but about how success is defined.