OpenAI Lets Some Customers Pay Only When AI Works, Report Says

Some major customers can reportedly pay when AI completes a task, adding another pricing model built around results rather than access or usage.

OpenAI has begun giving some major customers the option to pay only when its artificial intelligence completes specific tasks, according to The Information.

The arrangement is another example of AI companies reconsidering what customers should actually pay for as the technology takes on more work.

The Information reported that OpenAI has offered the pricing option to some large customers in recent months. It cited customer-support interactions as one example of a task that could be priced this way.

An OpenAI spokesperson declined to comment to The Information.

OpenAI has not publicly announced the pricing model. Its published business pricing still includes per-user pricing for ChatGPT Business and credit- or token-based options for Enterprise customers.

The reported arrangement appears to be a pricing option offered to some major customers, rather than a broadly available OpenAI pricing model.

From paying for usage to paying for results

Much of AI pricing has been based on access or consumption.

ChatGPT Business, for example, charges by user. OpenAI's API pricing is largely based on the number of tokens processed.

Outcome-based pricing moves the billing point closer to the result.

Instead of paying because someone had access to the technology or because the AI consumed computing resources, the customer pays when an agreed outcome occurs.

OpenAI isn't alone in using pricing tied more closely to outcomes.

Intercom prices its Fin AI Agent based on defined outcomes, with pricing starting at $0.99 per outcome. Sierra says customers using its AI agents pay when the software achieves agreed outcomes.

Salesforce offers several ways to pay for Agentforce, including per-conversation pricing, action-based Flex Credits and per-user options.

As AI products move from helping employees do work to performing more of the work themselves, the old billing units don't always fit as neatly.

A seat makes sense when a customer is buying software access. Tokens measure how much computing capacity is consumed.

Neither necessarily measures whether the customer got the result they wanted.

The harder part may be defining the outcome

Charging for a result sounds straightforward until both sides have to decide exactly what counts as success.

Customer service may be easier to define. Did the AI resolve the issue?

Other work can be harder to measure. A task may involve several steps, human involvement or a result that isn't obvious immediately.

Outcome pricing depends on being able to define and measure the result clearly.

Stripe notes that outcome-based pricing connects charges more closely to business results, while creating more variability in what it costs the provider to deliver those results.

Sierra has also said the model becomes difficult when a result can't be cleanly attributed to the software.

Insider Take

OpenAI's reported pricing arrangement puts a different question behind the bill: What did the AI actually accomplish?

With usage pricing, customers can pay for activity even when the technology doesn't produce enough value. With outcome pricing, more of that risk moves toward the provider because unsuccessful work may not generate revenue.

That can make the value easier for customers to understand. It can also make revenue and delivery costs harder for the provider to predict.

And it depends on something deceptively difficult: both sides have to agree on what an outcome is and when it has actually happened.

For subscription operators, the bigger story is the changing billable unit.

As software does more of the work itself, companies are experimenting with whether customers should pay for access, consumption, completed work or the result itself.

Sources

• The Information: OpenAI Starts Letting Some Customers Pay Only When the AI Works
• OpenAI: ChatGPT Business and Enterprise Pricing
• OpenAI: API Pricing
• Stripe: A Framework for Pricing AI Products
• Intercom: Pricing
• Sierra: Product and Outcome-Based Pricing
• Salesforce: Agentforce Pricing