Ramp CEO Eric Glyman: Using AI to Build “Self-Driving Money”

0:00Look, I think the most important thing beyond this empathy, if you're trying to make great products, you need to have great taste. You need people who are making to so deeply understand really the experience of building the pain that people are going through. What they're actually, the customer at the end is actually doing it or running your business that you understand it not just decently, but in some cases, better than the customer has. and I think only then can you actually build products that are so well designed that they can actually automate the task. They can do it more efficiently and I think part of why you know as we're releasing products engineers are on the call with customers when we ship they're accountable for metrics and how it ultimately performs and I think you really won't find people at RAMP who haven't talked to customers with any level of recency and I think that's to say core part of what makes great product cultures.

1:03[♪ OUTRO MUSIC PLAYING [♪

1:11Joining us today is my friend Eric Gliman, co -founder and CEO of RAMP, which offers companies AI -powered financial tools to manage their spending and expense processes. many companies rush to add chatbots after the chat GPT moment, but Eric has a different take. He doesn't think that chatbots are the right form factor for everything. I mean, who wants to chat with Eric's Benz report? He thinks zero touch automation that works invisibly in the background can be much more valuable in many cases. In fact, he thinks a better analogy is to self -driving cars, or in this case, self -driving money.

1:41We'll talk about ramps approach to building an AI and how Eric sees the space evolving. Welcome to the show. Eric, we are We're so happy that you're here. We, you know, the original invitation for this podcast, as you know happened over dinner, and we were so happy that you decided to do it with us. So we have tons of questions for you about the future of AI and how it's gonna impact finance and businesses. But for the very few people that don't know, can you mind just saying a few sentences on what ramp is and what you all do? For sure. I mean, like we're dedicated to making companies more profitable and operate more smoothly.

2:19The way you can think about Ramp is, you know, we built a command and control system for company finances. So from one place, you can issue cards, manage approvals, make payments of all kinds, and even automate closing your books. And so for your finance teams, it means that your operations are simpler. You can automate a lot of business processes, and its services update an intelligence on how your company can spend less. And so the upshot is the average company using Ramp is able to save about 5 % per year on their expenses, which is pretty material. You know, over the past five years, we've been in business.

2:53This is added into billions of dollars of savings and the equivalent of thousands of years of labor that's been saved. And so, whether it's large, publicly traded companies like Shopify, Virgin Voyages, Boys and Girls Club in America to 25 ,000 other businesses in spring and benefit today. And a lot of what we're trying to do is really answer the question of how do you make people more productive with their time and with their money. I love it. Well, one thing that I was surprised by before we get into how ramp is using AI, you and Kareem, I'd tell you guys posted this the other day, you built an AI agent in 2015 at Parabas.

3:33I mean, this is way before it was cool. What did you build back then? And maybe tell us about that and then we wanna hear about what you've built now and how it's changed over the last nine years. Sure. Yeah, yeah, so Parabas was a much It was a very simple tool and I think it is fair to call it an agent basically Go back to a decade ago if you were buying something online Whether it's at Amazon best by whatever all these stores would guarantee that if you bought some TV for a thousand bucks The next day it went on sale for $900 You could get the difference back if you ask them these were price adjustment policies And so we built it was an app that integrated with your email software, your Gmail, Yahoo, whatever you used, scan your inbox for receipts, track the prices of what you bought.

4:20And if there was a price drop, you were eligible for something different. It generated an email to sound like you, sent it to the store. The store customer service responded and you know, you wake up to $100 back the next day. And so that was it. It was this agent that lived in your inbox and helped you save money. And, well, I guess I would say like a couple of things. I mean, first, I think it, you know, many people are thinking about a genetic AI and how is it going to change tools in this new use case? I would argue it's actually been around for a long time. This is a decade ago. You know, millions of people use this in order to get price drop, refund from stores and retailers, but it was a very narrow use case.

5:04When I think about AI today in 2024, I think that the sets of use cases where computers can be doing work on your behalf is far easier. It's not just going to be these narrow surfaces like price adjustment guarantees, it's really intensive and hard to do that, but more generalized reasoning, helping your finance team run more efficiently. Maybe not just asking for price refunds, but maybe even helping you negotiate and running more complex analyses. But I think this primitive of software that does things on your behalf has been your for a while and I think is going to be growing pretty quickly.

5:39That's a really good segue into what you're building now at ramp. I'd love to learn more about your vision for ramp by AI. Yeah, for sure. I mean, first, like, forget even just this ramp AI. We're dedicated to this mission of helping people spend less money and spend less time. I think we're very excited about AI because it's a new set of tools that enables us to do this, but ultimately comes down to like, where is the pain point? Are there processes in your business that result in your company spending more time than you need to? Or are there areas in your business where you're spending too much money?

6:17And so when we think about where things are going to go over time, so much of running a finance team and Rubba, you know this from your time as being CFO to Instacart, there's a lot of tedium and monotony. You know, you're trying to grow your business and yet when you look deep in your finance team, you'll find people who are auditing expense reports by hand. They're downloading spreadsheets in order to be able to tag and classify vendors. They're re -running analysis time and time again. And what we're trying to do at Ramp R vision is how do you take these tedious and monotonous tasks and either through better design integrated tools automate this.

6:54And so, you know, the simple way to understand this is instead of eating two apps to buy one thing. your American Express and your concur, where people are going in and out and adding a receipt to it. If it's one tool, you tap your ramp card and through zero touch where we will pull the receipt from the merchant, from your email, your expense report, is not just easier to do, but it's done for you. So there's a lot of cases of zero touch expenses, better categorization. We sort of apply large language models to the transaction data itself, as well as your general ledger. so we can auto complete expenses for you, certainly faster, and for the vast majority of customers more accurately, so your books are effectively doing themselves.

7:35And over time, ramps should be able to point to ways for your business to operate vastly more efficiently. Same sets of vendors, better prices, same business processes done automatically for you. And so there's a lot in just the manifestations of how ramp helps your finance seem run. More efficiently, there's also ways we're trying to experiment with even how should people interact with ramp? Should it be something where you're prompting ramp to do these things or should it be taking care of these things for you and doing working? You're happy and so we can go either way, but that's that's a loose framework of how we think about it.

8:10And how do you decide what's in scope like anything that makes a finance team more efficient? Is that what's in scope for you or what what's in scope and what's out of scope? Yeah, definitely. I mean, I think I I think so, you know, that's probably the right way to put it. You know, it really is, will, can we create a product that should save time and money for, for our end customers? If so, I think it's in scope. You know, today, ramp is largely used for payments out of a business. So I think hard expenses, bill payment expenses, business processes on top of that. But when you think about, you know, over time building towards self -driving money, you you know, funds department that's improving itself, I think, thinking about higher yield, more efficient collections, to even more efficient record keeping.

9:01I think are all things that we're thinking about over time. I would argue there's very little out of scope and I would say in the same way, we think about ourselves more as a productivity company than we do is just a money company. And so anything that makes the productivity of capital and your business go higher are things that we should be thinking about. Derek, what is the intuition then on your side? You mentioned you're thinking about how folks are actually going to use AI within Ramp, right? And you said them prompting Ramp versus things happening in the background on their behalf. How do you think it's going to play out?

9:34I get that you guys are going to test it and you're going to listen to customers, but where do you think it's going to go? So I mean, maybe just like a couple of notes stepping back. They think about the first phase. is what happened after a chat GPT came out. Suddenly, there was a chatbot slapped on everything. And people say, finally, it's your app and you can talk to it. And I've personally never met anyone who says, I just wish I could chat to my bank account. I wish I could chat with my expense report. And so I would say, I totally get the skepticism of the first set of people who finds this really silly.

10:11And I think that we've tried to classify internally, and when we think about the sets of user experiences, probably into two buckets. I mean, the first, we think a lot about this paradigm of really zero touch AI. And I would argue even Parabus was an early example of this. The interaction was actually, no UX. You would sign up, you'd link your email, and that was it. And you'd wake up the next day, and it was done for you. And there's a whole category of experiences like this. You tap your card, your expense report is done for you. Your accounting is done on your behalf. We're suggesting memos for you to click 1, 2, 3.

10:53And so we're trying to predict and understand what are the user interaction inputs we may need. Can we get to a level of competence where we can do this all the way? And can we expand the surface area of these? And so there's a whole set of things that happen from the administration to the analyses to the operational closure of tasks that you can do. I think there's a large category there. We can unpack and talk about the expansion surface area. There's this next surface area. Internally, we think about it as a genetic AI, where you're going to want to effectively prompt some kind of outcome, maybe monitor what's being done along the way, or once you get a level of trust have that done for you.

11:41Well, it's been a use case. A few weeks ago, we launched, right out there, the launch of GPT -40, the new multi -model for Mocha AI, which had vision, audio, and text understanding capabilities. We created a way for customers to just ask Ramp what it wanted done. In Ramp, we would show you how to do it and do that on your behalf. So you could say ramp, you know, I'd like to issue a card with the budget of $50 that only can be used at Starbucks go and then Effectively the large language one could read everything that's on your screen everything you're seeing and would guide you through steps in order to how to do this we call this the tour guide, so which I you click here Enter in this text and you could monitor this and you can see with extreme accuracy Again, we didn't build into the interface.

12:34Click here, this wasn't a managed demo. The reasoning of large models are now capable to go and do that for you. And I think there's a whole class of services. And I would argue the majority of businesses actually done this way. The interface that most people think about this today is I think as you've joked once before, Revy, you know, you hire an analyst and you say, hey, I want this thing done. Go figure this out for me. And you know, that's your interface. I think there have been forms of AI that have been around a long time as I mentioned to you. Exactly. You've had a large language model before.

13:10It's an annual spin. No, but I think that's actually a good way of thinking about these things, right? I'd actually say it's kind of strange that the status quo today is, people are trained to learn how to use your app. Instead, I think it makes more sense to me that people should be thinking about how to run their business, is how to sell the customers, how to find ROI versus how to manage the intricacies of how your app is designed. And I think done right, you should be able to say, this is what I'm trying to accomplish, go get it done in the early era. You may monitor that, but over time as you build trust in that process and the reasoning capabilities increase, you'll see it go.

13:50And I might draw an analogy even to you two, the progression in self -trying cars. When you think a decade ago around the hype of self -driving, you would have people sitting in the driver's seat ready to take over with a steering wheel and now there are waymos all around the Bay Area. It's strange when there's someone in the seat because the capabilities are actually more accurate. I think there's going to be sets of tasks over time that will get handed off. You can have an agent, it's owned drivers, driving parts of your organization where there's high fidelity. I'll pause there. By the way, there's so many jokes I could make about Ravi handing off works to others, but I'll hold myself back You know some of us have been focused on ROI for longer I'm not gonna apologize for that Eric if you had to guess Like a lot of what you just described was almost more conversational the way you kind of get an agent to do something Like is the way you go the way that Ravi would go talk to talk to a teammate What does, do you think that means that kind of software interfaces as we know them in terms of, you know, buttons and knobs, etc.

14:58Kind of go away and fade into the background or how do you think that all plays out and like what does it like AI first user experience look like? First, yes. I do. I do think so very, very much so and I think that great design actually is about understanding, you know, the job to be done so well. that you can reduce steps, you can make things easier, more intuitive, and effortless to get things done. And, you know, I think balancing, you know, not only really the range of powerful tools that you can do against, and outcomes that you want to drive to the simplicity of the interfaces is always going to be attention.

15:42But I would say we think about that quite a bit internally. I would say, I think you will want to be able to audit, understand, functionally, what are these powerful tools doing for you? But done right. Yes, I think that it's going to be more, in many cases, just prompting, here's the outcome I want to drive. and the tool will go and drive it, which just is like a brief aside like Thera's, you know, I'll put it this way, there's over 25 ,000 businesses using ramp. Some are run by sophisticated finance teams. You know, others are running small businesses. There's lots of complexity and baggage that prevent them from focusing on the areas that really generate a lot of value from them, meeting new clients, writing better pitches, investing in the parts of their business that generate return, building the next -grade product.

16:45In so much time, I think, is stolen by people having to learn an amalgamation of tools stitching together processes and done right, the world should feel more frictionless, which should feel more smooth. And so I would say like, I think if we can't accomplish that, I think it's been a failure as an industry to deliver that for people. Well, one thing Eric that I'm curious about is you think about ramp has a lot of pride in its culture, right? And it's talked about externally. It's something that you guys are known for speed and quality and real customer obsession. How did you build the culture and engineering in particular?

17:21or you and Kareem that sort of embraces wanting to implement AI and wanting to do it well, what did you guys do there? Because I think that there are places that are a bit more resistant to change and less embracing of a change like this. Well, so a couple of things. I mean, I would go back to again, like, what is ramp? Like, we're not some AI -driven finance tool that we're going to market. Do you want to adopt AI and your financing? No, it's, we exist to save you time and money. We lead with the benefit. We talk about the outcome that we're trying to drive and we put these in simple terms, but I would say it's actually not possible to use ramp today without using AI.

18:02When people say that it's so easy to submit expense reports or that Suddenly my books are getting closed faster. It's often because AI is inserted at lots of different parts throughout the process And so what I would say is I think for one of the first things that we did to try to get at that was focus on what is the problem? We are trying to help businesses operate, you know, using less time, fewer hours, and less capital. And AI was a means to an end, but was not the end. So we didn't want to have technology in search of a problem, but really focus on what is the problem. Then once you start to decompose the question of like, what are all the areas that are wasting lots of time?

18:43It turns out a lot of time, it's process automation. And so I think it just became like the right tool for the job in so many different cases. And I think rather than trying to say, would you like to use this tool instead, you say, would you like to, you know, do your expense reports? Yeah, exactly. And I think that the other thing too is in so many businesses, I know there are a lot of founders listening to this. I think abstract makers away from the problem directly, which is to say, you know, you can go to a small store with 10 people and everyone's talking to users. And then somehow you go to a company with 500 to 1 ,000 people and you, you know, try to figure out who's talked to a customer over the last week.

19:35hopefully all the salespeople. But you start talking in marketing and engineering and people haven't done it. And what I would argue, look, I think the most important thing beyond is empathy, if you're trying to make a product, you need to have great taste. You need people who are making to so deeply understand really the experience of building the pain that people are going through. So what they're actually, the customer here and is actually doing their run your business that you understand it not just decently, but in some cases better than the customer has. And I think only then can you actually build products that are so well designed that they can actually automate the task, they can do it more efficiently.

20:22And I think part of why as we're releasing products, engineers are on the call with customers when we ship, they're accountable for metrics and how it ultimately performs. and I guess you really won't find people at RAMP who haven't talked to customers with any level of recency and I think that's just a core part of what makes great product cultures. Eric, since you've been building with AI for so long, do you think that the last year or two with these foundation models has been a kind of discontinuous step change in terms of your ambitions and what you're building with AI and in what way or do you think it's kind of just gradually compounding more and more AI magic in the product.

21:01I think it's been a step function change for sure, at least as practitioner in the market, not a researcher. I think it's become increasingly obvious to us that this is the biggest shift to productivity, certainly in my lifetime. I think it has all sorts of ramifications for for builders, right, and people building software. You know, when I think about like, what are the sources of durability and moats and a lot of software businesses? Sometimes it's, you know, there's more features, there's more integrations. There's lock -in. You've been using my tool for 10 years and it's really hard to take all your data out.

21:44And I think is now, today, functional, you can have human level reasoning. And in some cases, superhuman level reasoning, available through an API. I think it has profound ramifications for people building businesses. And I think it's expressed not just in the ability to understand large sets of data and act on it. But it's a wider variety. I think that there's, you know, I can say, you know, beyond even just the services we're providing to RAM. Part of how we've grown so quickly is we have AI automation in outreach. or if SDRs that are multiple times more productive than at competitors, it's changed how we do customer service.

22:29It's changed how we do copywriting. We can listen to 100 ,000 sales calls at once and ask what did 100 ,000 people think? It's just things that weren't possible, even just a few years ago. And so I think it's changed really rapidly. And I don't think most people are really, I think people are experimenting in some cases with Chatchy PT, which is great. But I think far too few people have actually started to incorporate into the crevices of how they're actually working day to day and felt it. But I think it should accelerate. When you think about that Eric, just so you're talking about the rate of change and how people are kind of scratching the surface on this, what do you think the job of a forward thinking excellent finance leader looks like five years from now, picked the time frame, versus today.

23:20How do they spend their time today? How do you think they will in the case that they incorporate ramp, they lean into AI and they maximize what this can do? Well, I mean, first, I think it's incumbent on anyone, whether it's a finance leader, or someone building tools in an engineer or designer, like I think people should be thinking about, you know, really automating all the parts of the job. Maybe you don't like or maybe that are low value, because I think there's a whole class of problems in acts in doing work that actually can be automated now. And I would, I think this is something like, I've always internalized and appreciated from our conversations, of like I think that great leaders are able to not just like create focus and urgency, but identify like where there's outsized returns to our times.

24:17And I think what's really true in a lot of finance organizations is yes, maybe the CFO has had the time and focus on where is their value. But you look at actually the calendars and what's happening in a lot of the rest of the team. It's a lot of repetition. Okay, it's the month is closed. We're going to spend the first six days, you know, it's eight in bad cases, 15, right? Tagging transactions, downloading spreadsheets, matching things into it, and that closes is done. And finally, just in the last five days of the month, you're able to do real work. The reason you got into that. Now, I think people should be thinking now about how do I really take those tasks, which are wrote intensive, and turn that process into more automated process, or I'd be thinking about, I think if you're doing this right, I really do believe, in five years for now, these work streams will tend to be more strategic, more insightful, more around where's their value created in a business and having people really obsess over that even more entirely.

25:22And I think to get there, it's about how do you automate the process is how do you design a more efficient system in the interim? Eric, what do you think happens when we have a more efficient system? Like, is everyone just out on the golf course or are we going to find new ways to work hard? It's, um, look, I, for me, I happen to think a lot of purpose in life is, is creation. I think people build tools. I think, you know, want to move things forward. And so, uh, like, don't get me wrong. I'm sure people find more, more time for, for a leisure. But, you know, I, I think of it, um, I think with it, there was a really interesting, um, There was something that was making the rounds a few months ago.

25:59I need to find it. I think it was a set of statistics that the number of book keep, there was a crisis in the US. How the number of bookkeepers had dropped by, I think over the last 10 to 20 years, like a million less bookkeepers were employed. And she was saying, what was happening all the bookkeepers? And it turned out, if you looked at job descriptions for financial analysts, strategists see if it was that it grown by almost a million. And, functionally, there really wasn't much of a change, but people were doing different things. Rather than tagging, tabulating, doing low -value tasks, people, I think, had moved to a higher level of abstraction in doing more valuable work for a business.

26:39And I think, I actually think a lot of that will happen. I think there are certain levels of work that are uniquely human, that are uniquely high value. And frankly, too, I think in many cases is much more fulfilling. And I think that for those who are for leaning, I think there's going to be much more of that over time. And so that's how I think about things for sure. I mean, I think there's also going to be strange things too, where there's certain creative work that computers, in some cases, will do better. But I do think that part of what gives people purpose, or at least for me excitement, like has to do with creation.

27:25And I think that's always going to be a very human thing. I mean, that's as good of a segue as you can imagine for my next question, which is there. You are a founder through and through. And you are someone that a lot of other founders probably look up to. What advice do you have for folks, for creators, for makers, for builders? given the moment we're in, you know, how do you meet it? What should you go and build? How do you approach this if you're someone who wants to be a founder or a builder? Well, first, I mean, like part of being a founder and a builder, I think is just about like running this very long standing and continuous race.

28:02I think that great companies are built over many years and decades. And I hope Bram is the last company that I ever work on. I want to be working on it for a long time. And I think there's always these questions of what's changing in the world, and how is that going to reset certain industry? I think there's a lot of opportunities, and we can talk about the place to be spending time. But I actually think when you look at ramp and part of what's made it work, is really been starting of what are the timeless truths that are not going to change, whether it's now or 10 years from now or 100 years from now.