Genius CEO Reveals AI is Doing $1 Trillion in Office Work

There's probably a trillion dollars of work towards jobs that I think are beneath people's potential. About a year ago this time, the entire world was spending on the order of $30 billion a year on inference across the major labs. Today, estimates are something closer to 150 to $170 billion. In other words, the amount spent on AI is equal to about 2% of all white collar wages.

It's growing over 100% per year. Growing faster this year than it did last year.

A typical business running on Ramp is able to cut their expenses by about 7 to 12% per year. I think a lot of companies get stuck in this idea that we're going to go sell you a product. They forget really at best companies should be selling outcomes.

Joe, couldn't be happier to be here. It's a ton of fun.

So Paribus, now the idea is much more popular, but it was a prehistoric AI agent is how I would describe it. The premise was, let's say you bought something online, let's say a TV, and it dropped in price a week after you bought it. Stores, Amazon, Best Buy, whatever, would guarantee that you get the difference in price if you'd asked them. So we built an app that would ask for you. You would link it to your Gmail. It would scan your inbox for receipts, structure the data, and if you were eligible, just wrote an email to them. Stores would respond, and you'd wake up to the difference, and we charged a percentage.

Well, one, it helps to have a really simple, easy to understand value proposition. Stores owe you money. We're going to go get it for you. And having very aligned incentives. That was super helpful. And a lot of the learning truly that we picked up and led into Ramp happened post-acquisition. So it was a very fast process. We launched in May of 2015. A year later we had sold, and Capital One bought the company and wanted to scale us from, you know, a million users to, you know, 10x more, put us in the credit card division. When we were there, we really got a deep lesson in how financial institutions work, what makes them great and very profitable, but I also got a deep education in incentives, and actually how does it affect what people really do. We found an organization that said good things, but somehow people were working to grow the amount that people spent on their cards. Then there was other teams trying to grow net margins. Sometimes it meant convincing people the points were worth a lot, and then there was another division that would go devalue the points.

Try to make more money. We were in the savings divisions, and so we'd call customers and ask them, you know, and we had to sell cards, too. We'd ask them like, what do you want? Do you want points, cash back, something different? And if you really listen to people, people would say, you know, I actually just want more in my bank account.

And we're like, huh, if you really think through that and you help people not spend $100 on some subscription that they're paying for but forgot about, no longer using, cancel that at the end of a month, that's $99 more in their account than if they got like a dollar back on that purchase.

And we got obsessed with this idea of like, why isn't there a card or software designed to help you go and spend less? But I think that financial institutions, a lot of organizations are filled with these misaligned incentives, you know, from the bank that could pay you more interest. Maybe they have a high yielding savings account, but they don't want to tell you about it. You know, loans that could be at lower rates. And not only did we learn a lot about incentives, but it led us to really question what if you tried to create a financial institution, a company that inverted those incentives, that actually was aligned with the idea of what if we tried to make our customer better off? What if we tried...

Yep, that's exactly right. And so that's where a lot of the idea came from. But we got a lot of education in that. We also got a lot of education too in what happens in large organizations. You know, they go from these lean, mean, fighting machine startups that everybody loves to work at to what happens when there's 50,000 people and how do you fight really the slowdown in bureaucracy that can go and set in, which has helped us become a fast-moving company. We can get into that too.

Take something as simple as, you know, an expense report or paying an invoice that you get, right? If you kind of think about it, there's too many steps to all of it. Probably most people listening, if you've ever taken a business trip or taken a client out to dinner, maybe you had a credit card issued by some company perhaps based in New York. Your expense policy doesn't drive or the thing actually works. Swipe it, goes through, you get a piece of paper, and then you're supposed to go onto another website, maybe made by a company out in Germany. And you are supposed to manually enter all the details of the transaction that was digital, by the way, in the first place. And you're going to enter all that. Maybe you do it the next day, you probably do it the next month.

Then your manager is supposed to look at it a week later, they probably look at it a month later. Then maybe a center of excellence in India reviews all this stuff. Then eventually it gets reconciled. And, you know, a month or two has passed. There's so many unnecessary steps. Paying an invoice. It's not just sending the check. When someone in accounts payable goes and gets this, they have to go and see, do we have a contract with this vendor? Is this really them or is it just someone trying to pretend that they're them? Did we pay this invoice already? To the descriptors breaking apart, how do we book this and allocate the expense? It can be 50 steps to do any of this stuff. The payments are the easy part. The hard part is actually going and asking how do you get this done with less bureaucracy, with less time. And I think that what our initial problem we tried to solve is, you know, if you could go and help make the company run on less, every dollar go further, every hour go further. You start asking about these different types of steps that banks never really thought to look at. They were in the business of selling money, and we wanted to be looking at how do you go and sell people time. And it turns out AI that can reason can go and not just digitize these processes but redesign the process to make it such that you could tap a card and your expense is done. You get an invoice and it's paid correctly on the last day that you owe the actual money. So you're maximizing your earn and you can just eliminate this bureaucracy in the first place. And so we're really focused on this.

It adds up. And look, when we launched the company with this opposite idea of we're going to help you spend less, you know, we did our very best and we could help the average business cut their expenses by about 2% per year, which is quite good. It's pretty significant. Today, a typical business running on Ramp is able to cut their expenses by about 7 to 12% per year.

Yeah. Yeah, we look both at what are the dollars going in, what was the company paying for on day one, to what vendors do they actually keep, what costs can we go and negotiate down, what waste can we eliminate. And so there's a whole side focused just on tracking dollars going out, what spend do we block that was out of policy that would have gone through if not for our technology. There's a whole segment focused on time. We actually go and try to measure clients' calendars, how much time do they actually spend in the app. And unlike social media companies that are incentivized to get you to spend more time looking at the feed longer, we're obsessing over how do we get you out of Ramp? How do we make sure you never have to go and spend time again?

That's right.

For sure. I mean, I think we're stepping into a future that is, look, it's very exciting. Like, when I think about the background of part of what made America great was, you know, we took on manufacturing and we were able to kind of go out, build, and outproduce. And I think that this machine intelligence is going to allow us to have firms that are radically more competitive. And I think there's a whole host of questions that need to be solved related to this. But, you know, a lot of the spend is going through cards and bill payments today. So, we're very attuned to it. But we fell into it in a funny way. Our, you know, back to the mission, help people spend less time and money. We were interested in these very narrow problems of how do we go match an image of a receipt so you don't need to go and be like this receipt goes here or this receipt goes there, or you get a, you're an accountant and you get a list of 200 transactions, you have to go click, click this, this invoice or this payment to Amazon, you know, is in supplies, which is accounting code, you know, 60,635. How can you go and have autocomplete for that?

And while we were building this, I mean, we just saw so many incredible moments. Like, we were one of the first customers to use GPT-3 when it was a research product, you know, months before it. And, you know, I remember it couldn't do arithmetic. To now, you know, within the last month, you know, AI is solving math problems that no mathematician alive has ever been able to solve. And so some of this is just feeling the expansion along the way. But, you know, our goal is, I think a lot of companies get stuck in this idea of we're going to go sell you a product, we're going to sell you a card, or we're going to sell you a car or a computer, and they forget, you know, really at best companies should be selling outcomes.

Yeah, 100%. Like, it doesn't make the world better if I can, you know, sell more credit cards and more wallets or sell more subscriptions. I think it makes the world better when you can actually look someone in the eye and say, like, if you use these products, your business will be better. And if there's other tools you can use that are better than mine, fantastic. But we need to compete by understanding who you are and how do we actually build products in service of that. And AI was this huge accelerant.

I grew up in Las Vegas.

It's a good question. I think some of this, I mean, even before right and wrong, just building things. I think a lot of people grow up in towns that have been established where the people who created things were people you read about in the history books.

Vegas 100 years ago had 500 residents, maybe. It didn't really work until the air conditioner was invented. And, you know, I'm in my mid-30s. Like, when I was born there, I think it must have had 400 or 500,000 people. And when I left to go to college, there was almost 2 million. And so my whole childhood was just parts of the city were getting built as you were there. You kind of didn't just hear about people whose names are on the buildings. You might know them. You might be able to meet them. And so there was just something to growing up there that I feel really lucky with, which is it did not feel crazy to be a kid who wanted to build things, to want to create businesses, to want to build. And I feel really fortunate for that.

You know, I think as to doing the right things, some of this is just maybe how I was raised. Like, you know, I think my family always had tried to emphasize this idea of humility, of trying to leave the world, you know, a better place and actually working hard. You know, I don't know how long we all have, but hopefully it can be worth it and meaningful. And so some of it was that, too. I think as I eventually got into doing more work for more people, I don't know, it feels better to be able to work super hard, help people, and just like actually hear, like, you know, see people who you worked hard for do better and succeed. And so I find it pretty infectious.

You know, just the history of the world, like it's never been zero sum. You know, people come up with an idea and everyone gets to be, you know, hopefully a little bit better off, and you can have small conveniences that help people live a better life. And if you just keep on doing that, like, you know, you just think about how much the world has advanced from, you know, thousands of years ago to hundreds of years ago, where most people were farmers. You kind of got by by how much could you lift up or what limited skills you have, to you have people flying around in tin cylinders through the sky, you heat up sand and you have intelligence.

I think it came back to incentives at the core of it. I think there was good ideas that came out of that company, to be clear. But in many ways, I think it was a continuation of how financial institutions have been built. Sign up for a card and we'll give you points and multipliers that if you buy on this category, we'll give you 4x points. This one, 3x points. This one, 2x. They had figured out a way to reach a new audience, but it was still at the core, what was best for the business was not best for the customer.

And I think that with Ramp, there was two big things we focused on. One, we should want the same things that our customers want. We should be happy to make less money in a given month so our customers can be healthier, have stronger balance sheets, and we can grow with them. And so we inverted the incentives to go and help businesses spend less, to do better, and to line our incentives with theirs. And so it led us to not just have happier customers who genuinely liked us, but, you know, it led us to ask the next right question. Not just what can we sell, but what do our customers want? What will make them better off?

One, just the scale of it all is hard to comprehend. I mean, truly, we talked about this, this business is doing multiple billions of dollars a year in revenue. It's growing over 100% per year, growing faster this year than it did last year. And the product maybe Ramp classically was most known for, of the corporate card world, we're on the order of 3% of corporate card transactions in the US, which is a really nice way of saying we have 97% to go. We haven't done much of anything yet. And by the way, there's more forms of payments. By the way, most businesses are earning almost nothing on their deposits. Most businesses aren't getting paid on time. Collections are a really hard problem and are having to deal with how do I go and adopt AI. And so we're just in the very earliest days of it all. But, you know, to give you a sense of the type of customers, we've got really every industry is represented. There's over 85,000 businesses that use this, from farms to nonprofits to companies launching rockets into space to manufacturers to healthcare. To, you know, I think about, you know, from the nonprofits, like there's organizations working hard to make sure kids have somewhere safe to go after school, and they're obsessing over this problem, and yet what drags them down is just the cost of bureaucracy of operating that organization. And so we're focused on helping make our dollars go to that mission. You know, and we're focused on making dollars go to whatever mission people are working towards.

Yes. Yeah. Well, we should talk about just AI spend in general because it is in some sense this unbelievable opportunity. I'll put it this way, like, I mean, we've seen just the advancements in intelligence. These models can be incredibly useful and help people think faster, access skills that they don't have or don't know about, to simply do tasks for them, make them more productive. And so there's a huge amount of promise. And yet I think about most of the businesses that I meet and that we serve, most don't have a single software engineer at the company, let alone one going and trying to make their finances better. And so when we're seeing AI usage explode and some companies getting all the benefit, a lot of what we think about is how do we bring this...

That's exactly right. And some of this is just simple tools embedded where I'm not trying to sell you AI. It's tap card and your expense does itself. Might be AI in the background that does it for you. It might be you get an invoice and it's paid, it's reconciled. We show you lower costs. We do a lot of the research in the background that you would want, and it's a part of how it works. And so a lot of this is putting it in the background. And then next, there are more and more companies starting to buy these services and use them. And I think this is an unbelievable but also very hard to manage category of spend. I think classically if you were a small business owner or if you were a CFO of an organization, you could build infrastructure to manage basically two types of spend. One, the people you hire. Maybe you have frameworks to assess how much they cost, how much are you making, and you can kind of predict those types of costs out. Or the vendors that you're doing work with. You have a contract, people follow the contract. If they go above it, you can decline the spend. It's fairly linear. Hard but fairly linear. With AI, it's quite different in some sense. One, it's growing very, very quickly. About a year ago this time, the entire world was spending on the order of $30 billion a year on inference, you know, across the major labs. Today, estimates are something closer to 150 to $170 billion per year depending on how you measure it. To put that in perspective, I think in the United States there's something like $8 trillion a year spent on white collar wages, sort of back office wages. In other words, the amount spent on AI is equal to about 2% of all white collar wages. And it just went up 5x over the last year and could go up who knows, but some another multiple again. And we're a knowledge economy and information ecology. It doesn't strike me as so crazy if you were to say, hey, we spend 10% of what we bring in on thinking. You might even say perhaps that's low and these machines are getting smarter. And so one, it's growing at a rate that other categories have never grown at.

Second, most of the spend, you know, in the industry they call it agentic spend. But there's really two types. One is maybe what many people are used to, of you have a chatbot, you ask a question, there's a cost of that question and it comes back and forth. Then there's another called agentic where you give it a complicated task and the model will spin up many agents who perhaps hire other agents who perhaps have their other agents to go break up the work and come back to you. So that's already 70 to 80% of all the spend is that. And what it means is that it's growing exponentially. Most companies don't really have visibility into how to go and manage this. And then there's also, back to incentives, very misaligned incentives. I think many of the people at the labs have good incentives, but right now they're racing towards building the smartest possible AI at almost any cost. I know they're thinking about safety, but one thing that they're not obsessed with is, well, sometimes there are some models that could answer your question for a dollar, and other times there's models that could go and answer it for a penny.

And if you ask the model that costs a dollar for it, they're not going to route you to the penny model. They have no incentive to. They're very compute constrained. And back to what we started Ramp with, we want to help organizations, well, pay the penny instead of paying the dollar.