Spending on artificial intelligence has overtaken a larger share of corporate budgets as companies seek labor efficiencies and process improvements.
As a result, tech players have piled into the field of providing AI gateways and routing for companies that want to monitor and optimize the tasks their employees send to AI model providers, such as Anthropic, DeepSeek, Google, Open AI and others.
Ramp, which issues corporate payment cards and makes software for corporate expense management and to pay their bills, is among the latest with an AI-routing tool.
New York-based Ramp is focusing on an area of potential rapid growth as companies gradually bring their spending on AI tokens – the unit of text an AI language model processes – under the same controls that cover their other corporate payments.
AI use is among companies’ fastest-growing expenses “and the one they can least measure,” Ramp’s chief technology officer, Rahul Sengottuvelu, said in an Aug. 19 press release introducing that company’s offering in the field, called Router.
Ramp is among about a dozen players in the nascent AI routing industry, with the company introducing an AI routing product publicly last month, three years after it began using the AI product internally. Stripe last month disclosed a reported $7.5 billion acquisition of OpenRouter, one of the largest startups in the routing field.
As a result, finance teams are working to optimize corporate AI use to send simpler, easier tasks towards cheaper AI tools, with more complex work directed to the more sophisticated and expensive models. These newer routing tools can also impose AI spending limits, much the same as corporate finance departments set travel-and-expense curbs for their corporate payment cards.
Ramp’s head of applied AI, Veeral Patel, spoke in an Aug. 20 interview about the company’s views of AI expense management and how routing may change as corporate managers exert greater controls.
Editor’s note: This interview has been edited for clarity and brevity.
PAYMENTS DIVE: How did Ramp decide to morph this internal AI tool into a new product to sell customers?
Veeral Patel
Permission granted by Ramp
VEERAL PATEL: A lot of companies were telling us that their AI spending had grown. Ours internally, obviously, had grown as well. I think it was something like 21 times since summer of last year, and we didn’t have visibility into what was happening. It’s just tokens. You see dashboards that aren’t updated in time. Filtering is not super easy, and so the initial product that we launched here was more just like the visibility layer. We called it token spend management. The whole goal of token spend management was just to have one single place to see all of your AI spend. Obviously, people are spending on OpenAI, Anthropic. There’s new providers, Meta, Grok, Microsoft. All these providers are coming out all the time, and the companies that are at the forefront of this want to test all these models, but don’t want to do it at the expense of losing this (cost) visibility.
It sounds like Ramp’s product is focused on expense management and financial guardrails. What about the technical aspects of who’s choosing which model is best for a particular job?
That’ll be our router product. You can control how aggressive we go and the guardrails there but we give you a choice. You can do it based off of benchmarks. You can do it based off of (speed) latency. You can do it based off of the pricing that you’re seeing. Our opinion is that there’s going to be 100 kinds of different levers that people want to pull to help route. And you’re not going to want to use 10 different tools to do this. You might as well have everything in one place. And at the end of the day, when you actually make the routing decision, we’ll show you inside the product what the rationale was, and obviously what the output was. For example, we chose this model because the latency was this way, and you said it needed to pass this score on these benchmarks, and because of that, we routed it to Model A versus Model B.
But if Ramp or others wanted to monetize on traffic routing, that could happen in the future?
I’ll be honest, we have not thought about it enough. Our goal so far has just been to make sure we get a stable router out. We addressed all the bugs that we saw. Thankfully, there’s not that many. But our goal is to just make sure that both sides of this two-sided marketplace are happy, whether it’s the providers providing these models or the users using them. And the immediate focus has been, ‘People love the coding use cases. How do we make it even easier to get started?’ So that’s what we’ve been focused on.
Where do you think the industry moves over time as AI routing becomes corporate financial infrastructure? Does that neutrality remain?
Ramp’s main business right now is our expense-management software. We have no interest in upending that anytime soon. I do think companies that use routers and companies that use AI products will actually think, hopefully with the help of Ramp, more deeply about the ROI that they’re getting from their AI spend. And so I think there’s going to be more guardrails on AI spend that people put on internal AI usage.
Will this become a large product for Ramp?
It’s definitely growing. I think we are solving a very important problem for a growing number of customers. With that said, I think you kind of have to look at it in slices. There are companies that are not spending on AI, and they will have different problems than the people that are on the opposite end, spending millions, if not hundreds of millions [of dollars]. Our challenge as a product engineering and design team is just to make sure that this product scales for every type of customer.