The Exchanges

Q Uh, this is your second company, right? You did Paribus before. Maybe talk about that prior venture, and then, um, how did you get to start Ramp? Why Ramp, and how did it all come about?

A Yeah, yeah. Um, so Paribus was about a decade ago. Um, if it was a 20, 24, we would have branded it as an AI agent. Um, but it was a weird savings app. So basically it was an app that lived in your Gmail or Yahoo. And the premise was, let's say you bought something at Amazon, Best Buy, Macy's, whatever. Um, let's say you bought a TV for a thousand dollars. The next week it goes on sale for 900. Every store would guarantee that you could get the difference back if you asked. We built an app that asked for you. Um, it would detect receipts in your inbox, go and scrape and track the prices, um, ingest the policies, and if you were eligible for money back, it generated an email as you, um, sounded like you wrote to the store, uh, chatted with their chatbots, and you would, as a user, wake up the next day to a hundred dollars or whatever back, and we charged a percentage, and, um, really fun, sort of insane, weird business, but we, we, we launched it in 2015. Within a year, uh, we had about a million customers, and Um, originally we wanted to partner with Capital One, and, ah, they said, you know, we'd actually like to, to buy the company, and so that's how we ended up there. Um, and in many ways, it was kind of the precursor, um, that led to what RAMP is today. Um, we learned a lot about turning data, ah, into savings, um, to go trigger some action based off of it. Um, We ended up …

AI assessment note: “in many ways, it was kind of the precursor, um, that led to what RAMP is today.”

Q So you very much coming from that DNA, uh, and, and, and jumping into a ramp, I heard you say that, um, externally people think of you as a fintech, as a corporate card company, but internally, You guys tend to think of yourselves more as an automation and workflow company. Do you want to just talk to that?

A I mean, so first, some of it just starts with like, what is the enduring mission of the company? We, we believe that we exist to help companies spend less money and spend less time. And we, we measure it, we report on it. Um, and if you start going a layer deeper and you ask, okay, well, where are companies spending money? Where are they spending time? Could they be spending less time? If so, it leads you down this question of productivity and We're, I, I, I think, when I think about like the original and marquee product of Ramp as being a card you can tap and your expense report is done for you, um, it texts you when the receipts are in, you can snap a photo, it pulls the receipts from your email, automates all that. It sort of stemmed from following that through. And so the problems in the fintech world from, from our purview was not, how do I get a credit card? People have credit cards. They have too many of them. Um, they get called too often by credit card companies. The problem is people don't turn their receipts in on time. The problem is finance team members are spending a lot of time manually tagging transactions, and so a lot of what we're venturing to do is really simplify the process where, um, you can functionally inject the expense, um, policy into how a card actually behaves. You can pull the data directly from the merchant. You can, um, append receipts from the …

AI assessment note: “we exist to help companies spend less money and spend less time”

Q Great. Can you talk about the sort of the tools and platforms and sort of the data infrastructure? Are you guys a Snowflake shop? Are you a DPD shop? Like I read something about Metaflow. What, what, what do you all use?

A It's, um, you know, I would say we use, uh, I think in some respects, probably almost all, um, is, uh, um, Is what I say the truth. So we use Snowflake heavily. Um, AWS is, um, you know, like a lot for a lot of our storage and core app. Um, uh, we just open AI and Azure, uh, as well for some of our more AI, uh, oriented. Uh, there's certain aspects, um, you know, of our platform, um, uh, work coming on, you know, vector databases, ClickHouse, um, uh, I think has been, um, fantastic, um, Uh, Materialize and others were, you know, again, you're probably wondering why are we using so many different databases? There's certain use cases. Materialize is great for, um, detecting and preventing, uh, real-time fraud. Um, happy to go a lot deeper, though. Um, and also, you know, overlaid on all this, we have, um, you know, analytics platforms like Looker or others that, you know, enable, um, you know, less engineering-oriented folks to go ask questions about the data, go and dive into this.

AI assessment note: “So we use Snowflake heavily. Um, AWS is, um, you know, like a lot”

Q of that whole wave was a lot of, uh, well, you know, I mean, I guess it's, it's slightly different for companies, but, you know, FICO score is actually a pretty good, uh, it's pretty good, you know. So have you, have you found, um, that there is a juice there to squeeze in terms of, like, just using, uh, machine learning and more data sets to do better underwriting?

A It, uh, there is, I think, um, and this was, I think this was like a big part of the New York tech story probably a decade ago, um, when there's a lot of, um, you know, fintech lenders, um, but I say is it's real. I think that a lot of those companies overstated, uh, how useful it is. Um, and to give you a sense, um, you know, for us, um, you know, if a sales team is talking to, um, you know, a prospect, it's almost a hundred percent of companies. Um, that we're able to, to approve. Um, and so you might say, okay, this can go from, you know, um, Maybe if you're, you're just getting started and you're not using great heuristics, maybe you've, that number is 60, 70%, then it goes to 80. I think great ML based models can allow you to go, um, maybe from low nineties to the mid, something like that. Um, but it's not going to totally change the, the nature of the business. I, I think for us too, um, um, um, part of why we were so interested in the corporate space as opposed to consumers was, um, the level of losses were so much lower. Restricted the complexity to us, and so in consumer underwriting, um, if you have a subprime type card, you might lose five to 10% per year. Um, you know, if you're really going aggressive, um, more general prime underwriting might be, you know, half a percent to two percent losses. Um, the corporate, if you look at Amex's disclosures, um, you know, his…

AI assessment note: “there is, I think... I think that a lot of those companies overstated”

Q How did you sell initially? Maybe, uh, you know, a second or like a minute or two on, uh, on go to market. And how does that work?

A Yeah, so initially, um, ah, there's a very funny saying, kind of in our world, or in the venture world, which goes something like, you know, if you go and you ask someone for money, you're going to get advice. And if you ask someone for advice, you know, maybe, maybe, maybe you'll get money. And, um, that was kind of similar to how we approached the initial sale. We'd go into, you know, we'd try to get introduced, or we knew somebody, we'd go to different companies, hey, we have this idea for credit cards designed to help you spend less. Um, we know some things about savings, we spent a few years in that. Um, I have some ideas about ways you could spend less, but are, are there places in your business you feel like you're spending too much? How do you run this? And we tried, metaphorically, to sit on the same side of the table as people, try to solve a problem together. Um, in some portion, people say, actually, this is really interesting. If you build this, These are real problems I have. If you solve it for me, like, I'd be interested, and I'll use it, and some, you know, maybe I'd invest or, or, or whatever, but kind of went from there, and that helped us in the early phase that people kind of agreed on what the problem space was, were interested, and in some way invested in the mission, um, and allowed us to start, you know, doing work, and, and it's evolved significantly. …

AI assessment note: “sit on the same side of the table as people, try to solve a problem together”

Q there's like a number of use cases that sort of Predates the, the big, um, you know, generative AI intelligence product that we'll talk about in a second. Um, do you want to talk about like the, the, the sort of the spectrum of use cases, uh, that are, that were addressed, I guess, before generative AI through regular kind of like what is not now known as traditional AI?

A Yeah, for sure. Um, back then just called good old machine learning, you know, prior to all this. And so I, I think, um, I'll, I'll go into a couple of things to just have, like, loosely we, we think about just AI at, at RAMP, because we, we do have, uh, an applied AI team, which, you know, is very horizontal. There's almost no part of, you know, how we operate that's, that's off limits, and I'll go into the depth of it, but there's a certain category of just how RAMP operates, um, how we sell, um, how we create ads, um, uh, how we understand the gong calls that we listen to, all of it, but there's a lot around just, like, the pure play operations, um, There's, um, uh, there's, uh, certain parts that power, um, whether it's underwriting the nature of the product itself, so it has less to do with the productivity of a salesperson or marketer, but it has more to do with, can we underwrite you? Um, can we prevent fraud? Can we match your receipts? That's another class. And then there's two sets of productizable, uh, AI. Um, some are zero touch, uh, and others are, which is probably the most emergent case of this agentic, um, AI, um, where it's not just, Um, you know, powering behind the scenes experiences, but the things that you can call in order to drive an outcome. Um, so I'll go into all of them. I would say some of the, the original use cases, um, in terms of powering our pro…

AI assessment note: “some of the, the original use cases... underwriting, um, was very heavy.”

Q How experimental is all of this versus, um, you're a hundred percent happy with it?

A So what we have out in the product, like, Very good. Um, that stuff is very standard. The, the things that are very experimental, and we try to have a heavy portion of that, generally falls under, um, what we'd call more agentic use cases. And in some sense, sort of circling back to Paribus, you can think about it as this highly limited agent. It would sort of assess vast amounts of data, over a hundred million emails a day, um, and send Price adjustment requests. Um, and we restricted the surface area that I could send emails to directly, um, to a very small amount, but even 10 years ago, agents worked. Um, you know, for over a million people, um, these were in productions. And so, I don't, I don't think the idea that a digital agent can do things for you, um, is a twenty-twenty-four idea. I think that's, that's been around for a while. What we're working on now and really interested in is particularly around two sets of things. One, some of the multimodal models that are coming out, and so, um, when GPT-IV-O model was released, um, we were interested in the question of, you know, look, about half of support requests that, you know, aren't fully automatable now fall under the question of user confusion. I want to do this thing. I don't know how to do it. Can you show me how? Um, and so we said, let's, um, Allow this model to see what the user sees. Um, and so this is in our, u…

AI assessment note: “So what we have out in the product... that stuff is very standard.”

Q So taking one, so organizationally, what does that, how does that translate? Does that mean you have, um, A decentralized data and AI team, which is embedded in different functions. So is that a central organization? And then, uh, if I want to build my automated, as you are going to borrow a person, how does that work?

A Yeah. Um, so it's varied at different points in the company's history today. We're about 850 people, um, uh, in terms of the size. And so the data, so the data team itself, um, reports into, um, uh, our CTO, um, And so on that side of the house, I'll, I'll give you the, kind of the RND org, um, uh, has, um, engineering, um, data, uh, design, uh, product, um, as well as, you know, risk and underwriting. There are some interesting things, too, in there, by the way. We have customer support, um, ultimately reports into product. Um, we don't need to go too deep into that, but I think it's because if people have problems, we don't want to dissolve the ticket. We want to fix the product. Um, um, and, um, Uh, there are certain projects where actually it can be a cross-functional set of, like, it can be data, it can be, you know, growth, it could be sales aspects actually working together on pods, and so on, you know, let's say automation in a sales capacity, you typically see teams where it's a few subset of folks working against it. What I would say about data, I do think that it's important to have a centralized function, otherwise you start to have disparate models. I think one of the most important investments that we make and continue to invest heavily in is our customer data platform. Um, in, in CDP, which is, um, you know, there's engineering, but it's, it's primarily owned, um…

AI assessment note: “I do think that it's important to have a centralized function”

Q What have you learned in terms of people's ability to work with those tools? Is that very natural? I guess maybe you just hire people that Can work with those tools in the first place? Like, the, you know, that whole relationship between, like, human and machine, and, uh, I find a fascinating topic. I'm curious if there's any, any sort of lessons learned?

A Yeah, I mean, a couple sets of things. I mean, so one, often these are, like, disparate systems, um, and disparate teams that don't work together, and so, you know, let's take the, um, sales development rep example. It was effectively, you know, an outbound rep, um, and it was the growth team, um, with growth engineering, all working together saying, we have one goal, we need to book more meetings, um, uh, each month, um, and these goals are going to go way up, and so there was joint accountability, um, to it, and there was different ways each of them worked. Some were focusing more on the data platform itself, and we focused on the automation, um, whereas the, um, sales representative Um, sort of had the intuition. Um, you know, he knew how to get meetings, but, like, didn't specialize, and so each tried to focus on, um, what they were great at in their respective practice, but I, I think the core of it was actually having, um, single-threaded teams with different subject matter expertise ultimately working together, um, um, aligned to it. In, in more of a, a minor note, I think sometimes, um, funny things happen as, as businesses get, get larger, teams get more siloed, Um, engineers, data scientists, folks who are incredibly smart, and I think incredibly, you know, the deep subject matter expertise for, like, oh, no, like, you don't need to get involved in sales or the busine…

AI assessment note: “core of it was actually having, um, single-threaded teams with different subject matter expertise”