OpenAI has started offering some large customers contracts that charge only when its AI successfully completes a task, according to The Information. Customer-support work is one example, and an OpenAI spokesperson declined to comment on the previously undisclosed arrangements.
Outcome-based pricing is spreading beyond subscriptions and token-based billing. Startups Sierra and Fin charge for tasks completed without human involvement, while Salesforce lets some Agentforce customers negotiate fees tied to increased revenue or reduced costs. Adobe has also said that part of its CX Enterprise suite will be priced according to value created, such as completed advertising campaigns.
The approach can make an expensive AI service easier to justify because the buyer pays for a business result rather than access to software. It also transfers some performance risk to the vendor. The difficult part is deciding what counts as success and who caused it. A sale or cost reduction may reflect product changes, marketing, seasonal demand or employee work as well as the AI system.
Contracts therefore need measurable task definitions, rules for human intervention and an agreed attribution method before deployment. Without those details, a pricing model meant to align incentives can instead create arguments over credits, failed tasks and whether the software deserves credit for a favorable result.