|
Greetings! AI payments are getting complicated fast—spurring the development of new billing technology. Providers of payments and corporate expense services built big businesses helping companies process bills and track and control major costs like travel and software subscriptions. Now, as spending on AI soars, there’s a new opportunity to help companies manage an increasingly complicated expense. That’s prompting companies including Brex, Adyen and Stripe to develop new features or even make acquisitions to add new capabilities. Many sellers of AI tools have moved away from subscription pricing to charging by usage or tokens, the small units of text that models process when handling a prompt or generating a response. Companies can also buy AI through application programming interfaces, cloud providers and developer tools like coding assistants, which connect to multiple models. Meanwhile, the prices of AI models themselves can change frequently. That’s a lot for AI-using companies to manage at once, especially as many are trying to better monitor and control their employees’ AI costs. For one, they have to trace spending back to a particular employee or project and verify charges against actual usage. “What you see in your usage dashboard needs to match what the bill says. That’s not always right, and so we spend a lot of time reconciling,” said Carter Busse, chief information officer at automation firm Workato, whose employees use a range of AI services from companies such as OpenAI and Anthropic. At the same time, AI vendors’ own systems are evolving quickly to account for different variables, and usage data doesn’t always line up with the invoices they send. That means companies using AI are setting budgets and cost controls around spending they might not have a number for until the bill arrives. “What’s in their API and their bill is kind of messed up right now because they’re trying to figure it out. There are different products and different models billed at different rates. I think everyone’s struggling with that,” Busse said. “It’s a complex billing invoicing model that we talk about every Wednesday at 2 o’clock” with the company’s chief financial officer and head of engineering. “We look at invoices.” Some corporate payments and expense-management providers are trying to tackle that problem by adding tools that can track AI spending and check charges against usage. One example is Brex, the corporate card and expense management company Capital One acquired for $5.15 billion this year. “We are spending a lot of time on the AI side, helping really rethink how finance teams should operate” in areas such as expense management, auditing and accounting, Brex co-founder and CEO Pedro Franceschi said. In particular, “token spend is a big topic because everyone is thinking about it now, and we put a lot of thought into it.” Brex said it plans to launch an AI spend management tool called Magpie later this year. Magpie pulls usage and billing data from AI providers’ APIs, tools like Cursor and invoices from AI vendors, and puts that data into a single format so companies can see where their AI spending is going. Some corporate payments and expense-focused firms say they are poised to benefit from the AI boom as more spending flows through their systems. Some companies might have employees expense AI tools on their corporate cards, and card providers earn some of their revenue by taking a small slice of card transactions. “A lot of the company-level expenses on AI go through credit cards, believe it or not, so we’re a big beneficiary of that,” Franceschi said. Other companies handle payments more directly through centralized contracts with their AI providers. For instance, Workato doesn’t allow individual employees to pay for AI tools using their credit cards. The company has direct agreements with AI companies and offers the tools at the enterprise level, so that it can better control security and data governance. “We provide every employee in the company access to every model that we subscribe to,” Workato’s Busse said. Keeping Tabs on Tokens In addition to helping consumers of AI, there’s opportunity for innovation in helping sellers of AI models and tools better manage their billing and understand any costs they’re passing on to their customers. Billing for AI is a complex task to manage when it is usage based or a hybrid model, as opposed to a flat monthly per-seat fee. It involves potentially millions of API calls a month, different models with different pricing, and frequently changing prices. And some API usage data doesn’t show up in billing systems in real time. Payments companies have already made some acquisitions in this area. Stripe earlier this year finalized its $1 billion acquisition of Metronome, a startup that specializes in usage-based billing, and Adyen agreed to buy AI pricing and billing startup Orb for $335 million in June. At the same time, companies are grappling with understanding how AI impacts the price and profitability of their own products. Some companies might be losing money on sales without realizing it, because matching their own compute costs to individual customer revenue can be difficult. “There’s a lot of these companies that are selling tokens and reselling tokens that are probably not making that much money on a gross profit basis, and they may not even know about it,” Brex’s Franceschi said. “Because this attribution between where the revenue is coming from and where is the cost associated with that specific revenue is much harder to do now.” Workato, for instance, this year launched AIRO, an agent team for workflow automations that makes use of AI, and is closely monitoring the related costs for customers. “The AI [cost] in our product is going up quite a bit right now,” said Workato’s Busse. “We are trying to figure out the finances of that—the cost versus what they’re paying right now.…It looks good right now, but we’ve got to constantly monitor that.” Solving for Visibility One of the hottest areas in AI-related payments right now is routers, marketplaces where developers can choose different AI models. Last week, Stripe announced that it’s buying OpenRouter in a reported deal of up to $7.5 billion. Corporate card and expense startup Ramp, meanwhile, recently launched its own router for AI models that it initially provides to clients for free. Still, some corporate expense companies say routers don’t solve the underlying issue of knowing if the money they’re spending on AI is worth paying. Brex, for its part, argues that the bigger challenge is helping companies measure the performance of the AI models, just as they would evaluate their employees. “Routing to different models is actually not the problem,” Franceschi said, adding that Brex is not building an AI router. “The problem is, how do you know the performance of the model that you’re using is actually equivalent to [that of] a more expensive model?” Brex plans to help clients measure performance of AI models. To do that, Brex is partnering with AI data-labeling startup Mercor and in discussions with customers to help build evaluations to understand how well the models are doing. “What we’re really trying to solve for is: How do you give the finance team a very fine-grained visibility into every dollar of AI spend that goes out?” Franceschi said.
| 
Gift Subscription