OpenAI Reportedly Tests Pay-on-Completion AI Deals With Some Large Customers

Some large OpenAI customers can reportedly pay only when the company’s AI completes an agreed task, rather than paying solely through subscriptions or usage fees. The arrangement, which has not been publicly disclosed, would make successful task completion part of the product’s price. OpenAI declined to comment.

The Information reported that OpenAI began offering the option over the past few months, citing a person with direct knowledge of the arrangements. The reported example is handling a customer-support request: payment is triggered when the AI successfully completes the task.

The unit of sale becomes completed work

The reported structure changes what the customer is buying. A conventional subscription sells access, while usage pricing bills for consumption. Outcome pricing instead ties the charge to a completed result. It is a model already used by AI startups Sierra and Fin, which charge for tasks completed without human involvement, according to the report.

Cognition has taken a more explicit guarantee-based route. It offers corporate customers credits of up to $10 million if its software does not deliver results worth at least the purchase price. These approaches differ in contract design, but each asks vendors to connect an AI system to a business result rather than merely to model access.

Why vendors are trying it

The Information’s account places the experiment in a tougher software-selling environment. AI products are expensive to run and, according to the report, have not yet accelerated revenue growth at software companies; customers’ IT budgets are also under pressure from competing AI tools. A fee tied to an outcome offers a more concrete buying proposition when access alone is harder to justify.

Salesforce is offering its Agentforce customers a related choice. Businesses can negotiate contracts tied to revenue gains credited to the AI or to reduced customer-service costs. CEO Marc Benioff said customers want to buy and price software in different ways, a sign that vendors are testing more than one path beyond a standard software seat or fixed plan.

The contract has to settle causation

The difficult part is attribution. Stripe has warned that a sales conversion, cost saving or other business result may also stem from product changes, marketing campaigns or seasonality. Unless an agreement sets explicit attribution rules, a customer can dispute whether the outcome belongs to the software provider or to its own work.

Adobe is also applying a value-based approach to part of its CX Enterprise AI suite, including billing connected to the number of advertising campaigns completed, though it has not disclosed specific prices. The next test for OpenAI’s reported arrangements is therefore not simply whether tasks can be completed, but whether the parties can turn those completions into a shared and defensible measure of value.