What would happen if one day, OpenAI told companies: AI doesn't do things, and you can't pay the money, what would happen?
This has already begun. According to a person with direct knowledge of The Information, over the past few months, OpenAI has been targeting some large customers with a new way of paying: businesses only need to pay when AI actually completes tasks such as customer service.
It looks like it's just a different way of billing, and it's behind the business model that has shaped the entire software industry over the past two decades.
In the past, it was good to buy software: an employee bought an account, how much money a month, and the company subscribed according to the number of people. In the era of generative AI, this set of rules has become increasingly unusable. An agent may answer a question for a minute, or may work for several hours continuously, completing a whole process for several people; more troublingly, the higher the number of model calls, does not mean the higher the value created for the company.
So more and more companies are starting to ask a more straightforward question: Since AI claims to be able to work, why not wait for it to actually charge after it's done?
AI is selling more and more.
Software companies don't know how to charge
In fact, the software industry has been working on this issue for the past two years.
The easiest way to think about it is to charge by usage, such as Token, the number of calls or how many tasks to complete. This approach is at least closer to the cost structure of AI than traditional account subscriptions, but another problem soon arises: using too much doesn't necessarily mean high value.
For example, it's obviously not the same thing that an agent makes 1,000 calls and how much sales it ultimately contributes to. A company's engineers call hundreds of millions of tokens a day and can't say how much money the tokens ultimately make for the company.
So it's easy for companies to feel that they don't know what they're spending their money on. They really care about the return on investment.
This is why "results-based fees" are rapidly entering the pricing table of AI companies. Sierra, a customer management company, uses a similar model: only AI can actually complete tasks without artificial intervention, and customers pay for them. Salesforce is acquiring Fin for $3.6 billion, which is also charging for problems AI successfully solves.
Cognition, a programming smartphone company, goes further. It pledges to enterprise customers that it can offer up to $10 million in credits to Adobe, HubSpot, and Z if the value of the final delivery of engineering results does not reach the cost paid by the customer.