|
Software companies such as HubSpot and Adobe have started pricing their AI tools based on the number of tasks the AI successfully completes rather than based on flat fees or the number of AI tokens they consume. Many more firms plan to follow suit, even as the practice is still being hotly debated. Some 31% of 230 enterprise software firms expect to primarily charge their customers for AI based on “outcomes,” or tasks their AI tools complete, by mid-2029, compared to 5% today, according to an email survey of software executives conducted by Kyle Poyar, a former operating partner at venture capital firm OpenView. Outcomes could be anything from an AI agent resolving a customer support ticket to generating an advertisement for a marketing campaign, though software providers don’t necessarily agree on how to measure such things. Among the firms adopting the model are big companies like Salesforce, whose CEO Marc Benioff recently told The Information that outcome-based pricing was coming to Salesforce’s AI customers. Still, it doesn’t seem like outcome-based fees will be the dominant pricing approach anytime soon. About 47% of the firms said that by mid-2029, they expected to charge customers for AI using a hybrid model of subscriptions and usage- or outcome-based pricing, Poyar said. That’s up from 37% of firms that said they use a hybrid model today and 25% a year ago. (A full report on the data is due to be released Wednesday.) The pricing changes are coming as AI threatens subscription revenues tied to customers’ employees, who have long paid for software “per seat.” AI providers at the bleeding edge of the technology, including Anthropic and OpenAI, are also moving toward more of a usage-based pricing approach for their tools and have discussed task- or outcome-based pricing as well, as they try to improve their margins. FedEx Chief Digital and Information Officer Vishal Talwar said he’s happy that a few of the shipping giant’s software and AI providers are already offering outcome-based pricing tied to the business metrics FedEx wants to achieve with the help of the tools. “It just keeps things simple,” he said. CEOs of AI providers and their customers “can look each other in the eye and say, ‘I trust you, you trust me, I'll make sure none of us really suffer a lot from this.’” The suffering he’s referring to likely reflects the fact that AI tools cost more to run than traditional software and are squeezing corporate profit margins. There are risks, however, most obviously in uncertainty about what customers will end up paying. A third of software firms polled by Poyar said the prices customers will pay using outcome-based models will be difficult for the customers to forecast. Meanwhile, 20% of the software providers said they don’t think outcome-based models will help them expand revenue fast enough. “If you go to the real world, there’s no way in a contract to figure out what [an] outcome would be, whether you can measure it,” said Amit Zavery, chief operating officer at ServiceNow, which provides software for IT and other corporate departments. (ServiceNow charges for AI as part of a subscription and levies extra fees if a customer goes beyond a certain AI usage limit.) Nor is it easy to figure out whom to blame when a customer feels an outcome wasn’t achieved, he said. In other words, a customer may object to charges if they believe the AI hasn’t completed a task to their liking, and the software provider could be left on the hook for those costs.
| 
