Skip to content

Podcast

Ramp Founder Eric Glyman on the Many Ways AI Is Changing Corporate Spending

Cheeky Pint

Source ↗ ← All highlights
  • Platform Built Around Time Saved
    • Ramp scaled to over $1B revenue in ~6 years by starting with card interchange and expanding into bill pay, treasury, procurement and software.
    • The company measures impact as time saved and ~5% expense reduction for customers, driving faster revenue growth for them. Transcript: Eric Glyman Yeah, of course. So first, just the pace that this has come together has been pretty remarkable. Seven years, yeah. I think by the time we were six years and change, the company had passed over a billion a year in revenue. The largest portion of that is card. And so that might be a classic kind of interchange-based model. But behind it… John Collison This is people in their business, they have spend cards. Everyone walking around the company has a ramp card, and you earn an exchange on those. That’s right. Next, you can think about bill payments and software. Eric Glyman Software, it’s a two-something business line, just about two years and some months. That’s over $100 million a year business line in and of itself, and that can be advanced functionality to maybe manage lots of entities to automate aspects of accounting, maybe aspects Of procurement, bill payments. So this can be sending checks, wires, ACH, and that’s predominantly a float as well in some cases, foreign exchange transaction business. Treasury, that’s a product that is about a year old, several billion dollars of deposits. Some of that is checking-like products. Some of that is more of an investment and money ladder type product. And then the other ones are procurement and then travel, which is a bit of an in-kind. And what’s been so interesting is that if you break down and look towards maybe let’s say contribution profit, gross profit of the business. A few years ago, it would have been 90 plus percent card. I think by the end of this year, the second, third, fourth, fifth lines of business will comprise in aggregate the majority of Ramp’s business. And so, it’s evolved into this platform by which if you’re trying to operate your company, it’s just a lot more efficient. You spend less. I think people know Ramp for, we help the average company cut their expenses by about 5% per year. The thing that’s been really fascinating is people are starting to use these products in aggregate. Not only are they not paying for five, six sets of point solutions, but they’re also not wasting time, they’re growing faster. (Time 0:00:34)
  • Automate Expense Review With Policy Agents
    • Use permissive policies plus automated verification: trust employees but shine a light with rules applied agentically.
    • Run expense policies through LLM agents to review receipts in real time and produce human-auditable reasoning. Transcript: Eric Glyman I love this question. And the interesting part, if you really kind of dig into ramps businesses, we have ways of backtesting and actually understanding. Based on how strict your policy is, how things are running, does that have an impact on how much time your employees are, let’s say, doing expenses? How quickly you’re growing? And you can start to actually compare based off of the operating hygiene of the company, what are end margins? What is like a pace of growth that occurs in the company? And I bring this back to- So you can do a correlational study between expense policies and company growth. Yeah. And look, it’s pretty interesting. I mean, the first, there’s an aspect of, in most companies, particularly high growth companies, it tends to resemble something closer to the no rules, rules, network approach of we’re Gonna trust, but verify and shine a light on it. Spend tends to be reasonably permissive, but then part of the breakthrough the past year is you could basically take an expense policy. Let’s say it’s written in plain English. If the flight is more than five hours, you can take business. If it’s under things that would have been like horrible to go and have your managers go and verify the stuff. Now you can run that through an LLM. I think that we’re processing over 100,000 expenses a day that are being reviewed agentically. This is a very fast-growing subset of the business. John Collison So sorry, like you give the RAMB agent this company’s expense policy, and then it applies it to the transactions that are coming through. Exactly. And so you can start to do things. Eric Glyman I think about the episode you’d had with Susan Lee was like incredibly instructive where, you know, she sort of says like it would feel so silly. And it’s not a great use of my time in some sense to be a very expensive machine learning algorithm to do to review expenses. But yet somehow a lot of people and companies are reduced to that thanks to Sarbanes-Oxley and the rules around it. Now you can have an agent functionally that does that takes that aspect of the job. And so it has access to the full set of data that you would, all the transaction related metadata, the receipt data, the timing data, the policy data. It then can go in real time review. You know, was this in or out can have the full audit trail explaining its reasoning. And today it’s over 99% accurate, which turns out it’s much more accurate than people are. Most people don’t know the expense policy and expenses are pretty automated. (Time 0:04:58)
  • Why Bill Pay Remains Antiquated
    • Bill payments stayed antiquated due to complex accounting rules and inertia, not lack of technology.
    • Upgrades happen in-place (checks, cards) but the loose network of PDF invoices resists easy modernization. Transcript: Eric Glyman Comes to bill payment in particular? I mean, I think this is some of what Alex was bringing up. It’s the constraints of programming things, you know, in an if this then that kind of world are very heavy. There’s a lot of complexity. And when you think about kind of the nuance and algorithms that govern, you know, why do companies spend money under some circumstances, how hard it can be to record, you know, I wouldn’t Overlook, let’s say that you’re a manufacturer and you’re buying some asset which you’re going to use for five years and it’s going to depreciate. It’s a very different accounting treatment versus I’m buying like a pay-as SaaS app. All the complexities around that that might govern whether or not you decide to spend. And then finally, once you get all this detail, like how do you review this and decide where to allocate your next marginal dollar is very complicated. John Collison But like many systems are in place upgraded, despite the fact that that’s pretty complex. So credit cards started with no real-time authorization system, which is crazy. And you just like hope that they were good for it. Exactly. Yeah, yeah, exactly. Exactly, the machine with the pleasing sound. It would do great in the current ASMR environment. But then they added, like, you could call up and get an authorization. And then they obviously added kind of the current systems we know. The modem. Exactly, yeah, yeah. And then the modem. And so kind of in place, credit cards were upgraded with much better capabilities. And similarly, if you look at kind of checks and how they work, even though we kind of think of checks as antiquated, it used to be the case that physical checks had to be flown all around The country to be like physically settled. And then there was the Check 21 Act in 2000, where they said a scan of a check is good enough. And so they could be digitized by the banks and then shredded. And then there was the you can take a photo of a a check with your phone. And so we’ve managed to take this super old-timey check system. And despite the fact it’s super old-timey in some ways, it has actually been meaningfully upgraded. And if you looked at the bill pay system from the outside, you would say, we should have DNS for companies. So rather than a company sending you their bank account details, you should like look them up in some central clearinghouse. And that way you confirm that you’re not being phished. (Time 0:12:00)
  • AI Is Changing How Software Is Built
    • AI is shrinking the half-life from problem to shipped change and altering how code is written and maintained.
    • Lines of code may become liabilities as models enable re-writing and outcome-driven engineering rather than deterministic paths. Transcript: Eric Glyman All blurry. It’s totally crazy. Like, you know, there’s designer shipping code, marketing at ramp reports into CTO, Kareem, my co-founder, who’s doing some of the best marketing I’ve ever experienced and seen. Shipping code to production too. I just think that the half-life of you see a problem to how long it takes for you to go fix it and do something about it is shrinking immensely. Or if you want to change the color of a button, you can just say, at ramp inspect, can you change the color of this button? And it goes and spins it up and it does it. Within minutes, it verifies and validates it. And so I actually think in some sense, you know, kind of these classic barriers that software businesses had are clearly going to erode, right? John Collison Yes, but, you know, there’s the old joke of it’s much more fun to write code than read code, which explains a lot of software engineers’ behavior. And it becomes easier to change the button. Is that like another instance of it being more fun to write code than read code, where lines of code are a liability and not an asset because they are something you have to reckon with? And so are companies at some level incurring some tech debt now where they are adding a bunch of code, which will maybe be harder to reason with later? Or do you just get bailed out by the models getting better? It’s really interesting. Eric Glyman By the way, one of the other interesting sub-conversations I’m hearing a lot too, is when you think about where tech debt comes from, there’s a set of conditions and trade-offs you make, You write things in a discrete and deterministic way. Under these circumstances, follow this code path. Under this circumstances, follow this other one. In a world where there’s LLMs and kind of the models themselves are improving, I think it’s entirely possible that the way that code is written is you say, here’s what I’m solving for, Under these kind of conditions, here’s what I want to occur, go write the code that drives this outcome. And maybe with the models such as they are today, can get it done in spaghetti code written fashion, but it works even though it’s some Rube Goldberg machine underneath the hood. But if these models get much smarter, you might write your code base in such a way where you say, every year, rewrite the underlying code, but here’s the outcome I can drive and today you Accomplished my outcome 90% of the time, 95, 98, 99, 100, and you just have kind of self-healing and writing code all the way through and this notion of like underlying code does goes Away because we’re writing things in this different manner. Obviously, there’s real conditions in which like that’s not going to, and that won’t occur. Like if you’re kind of writing code that needs to have, you nines of accuracy and uptime, that’s probably not the methods you’re using. But if you’re a growth engineering team, that’s probably what you’re doing today if you’re on the leading edge of this stuff. And so I find this stuff very fascinating. And when I think about the deeper implications of this stuff, I think that if you are completing a small amount of cognitive work, let’s say you just are an expense app, the spend has occurred, You need to go write it down somewhere and get someone’s approval, and that’s all the knowledge work that you’re doing. That’s very few tokens in order to accomplish that, both to create the infrastructure in order to facilitate it, that probably evaporates. I think anyone can probably custom write that kind of app. Whereas if you’re doing much deeper kind of work, such to underwrite a company, provide financing, automate areas accounting. I think kind of the fitness function for companies becomes, can you actually do things in such a way where even if you could kind of spend tokens on it, it would take more tokens to create The thing or do that work than the system maybe that you’ve built to kind of drive that outcome. So, I just think the rules of what it means to be a software company are changing. (Time 0:17:20)
  • Data Becomes The Enduring Moat
    • Proprietary, historical, and hard-to-recreate datasets become stronger moats as models commoditize surface-level features.
    • Aggregated domain datasets (legal records, WHOIS, ADS-B) retain high value that models cannot instantly replace. Transcript: Alex Rampell So I met this company, Vilex. It’s a 25-year company. This guy, Alex, bought up every legal record in Spain, like probably going back to 1492, like buys the, here’s the Ferdinand and Isabella, like, you know, here’s the document. Like, I’m going to take a picture of that. I’ll sell it to every law firm. And he built like, it was like a $20 million a year SaaS business, you know, bootstrapped for 25 years. And then it went to like a hundred million in one year. Now, why is that? Because he used to sell this document or like, you know, he used to sell a subscription to Kirkland, Nellis and Latham Watkins and all these big law firms. And they would take, a paralegal would take that. They would turn it into like a document. They’d charge the client $10,000. And, you know, ChatGPT can’t do that. Like, Gemini can’t do that. But you know who can do that? Is Vlex. And now, so another good example of this, of like who has proprietary data? A lot of times the data is free. This is the really cool thing. Or like it’s sitting, like I wouldn’t call your data free, but it’s not for sale data sets. So do you know domaintools.com is my favorite business? So domain tools runs a cron job every day on every single internet website. They do a whois lookup. So, Whois lookup every single day. I made a historical record of that which none has. So, if you want to see whoownsstripe.com in 1999, like, well, it was free in 1999. So, if you invent a time machine, go run that Whois query in your terminal, but you can’t do that. So, you have to pay them. But all that data is free or flight aware with ADSB data. Like, so have a lot of these weird businesses. And of course, this is not a new idea like FactSet, Bloomberg. You have aggregators of data, but that becomes so much more valuable. (Time 0:23:32)
  • Ramp Data Shows Stronger Business Health
    • Ramp’s transaction data signals broader economic strength and faster adoption of AI than official surveys show.
    • Savings-driven margin improvements can make small efficiency gains equivalent to much larger revenue increases. Transcript: Eric Glyman Tell us about the economy? It is, I think it is stronger than many people understand in lots of ways. I mean, first, I’ll go back to one of the things that perplexed maybe our team for a long time. And our economists are on the team, saw this data even as recently as last year, where the Census Bureau would do this periodic surveys where they’d go out and ask, how much is your business Using AI to produce goods or services is very, you know, refined economic way of wording the questions. And, you know, they would come back with these pronouncements saying a single digit percent of businesses in the US have adopted AI. And we looked at our data and, you know, we support over 55,000 businesses. We lean a little bit towards tech, but not heavily. It kind of resembles the distribution of you would see in the States. And well, the majority of businesses have used AI. You look at businesses, whether they’re paying for- That’s in they subscribe to ChattGPT or Antriotic or something like that. Exactly, or maybe a business that is a true agentic cognition or something like that, where this is kind of vertical application. And so one, there is this disconnect between kind of the use of tools if you look at, of how quickly businesses are adopting and responding to these new tools versus what maybe people Report on. Next, growth. I think it’s been clear over the last few quarters is the US itself is re-accelerating. GDP growth has gone from maybe the 1% to 2% area to 4% to 5%. And you could argue how much of this is a little artificial with subsidies, but it’s been pretty significant. But subsidized by what? Some of the big, beautiful bill, as well as some of the tariffs and where that’s been reallocated. I think some people, I would argue on the whole, maybe unfairly said this. I think that there’s more durable growth inside of the businesses. But I would say overall, business health is much stronger. And I think that the really interesting thing, and again, some of this is specific to the data that we see, but I think that businesses are generally getting more and more savvy about Finding tools that help them be more efficient. And I think what skews our data is maybe the savings we drive for people. Sometimes the way of explaining it is like people know the Ben Franklin would say this phrase, if a penny saved is a penny earned. And this is an awesome aphorism, but average American business has an 8% profit margin. And so mathematically speaking, a penny saved is equivalent to 12 pennies of revenue earned. And so if you save businesses a lot, you end up with these outcomes of the average ramp basis materially outgrowing kind of the regular U.S. Average. And what we think, we can clearly see this and demonstrate this for our own set of customers. (Time 0:33:10)
  • Cut Cost With Controls And Market Signals
    • Add fine-grained card controls, single-use cards and merchant blocking to prevent waste and phantom subscriptions.
    • Use vendor-level market data to flag overpaying and prompt procurement to renegotiate before renewal. Transcript: Eric Glyman To have these fine-tuned kind of controls, this part has been, I think, probably the biggest lion’s share of this. Let’s go back to one of the earlier conversations of why do people use checks, this horrible system and purchase orders. Well, one of the advantages of checks is unless you mail it to someone and you sign it, no money can leave your company. And whatever maybe person knows who’s ever had a credit card and gone to a gym once and had a good New Year’s resolution, but as you go a couple of times, you’re like, God damn it, this gym John Collison Is charging me. The difficulty of sending a check is a feature, not a bug. Eric Glyman Yeah, exactly. But part of what Ramp was the first to build. One-time cards. Single-use cards, but also more than that, merchant blocking. So you can, whether it’s on one card or 10,000 cards at once, build in a kill switch to say, okay, we’ve signed this new deal with this merchant. We’re only on Uber. We are no longer taking lifts. And anytime someone goes and tries to go on merchant A or B, you can do this. Or you sign a new contract and you say, I’m going to spend $10,000 only with Salesforce. And on the 10,000th and once dollar, they try to charge you, it declines. And what gets to happen? You get to have a conversation with their sales team who really wants to make their budget. And if you look at this mathematically, for most companies that come over a low end, maybe a very hygienic company that is more giving out a few cards in the old world, you know, could durably Cut their expense about 2% a year by doing this. Some of these very laissez-faire businesses are saving just 10% just through these types of things. Next, you start ending with these more fine-tuned kind of controls where you can say like under these circumstances, I want to go and have kind of charges occur. Maybe you have an engineer stay till 8 p.m. At night. You can go and buy a meal. But if you take an Uber on a Saturday, that should turn off. And so, you have the cards. If it’s like an Uber on a Saturday, auto decline. But you can text and say, actually, it’s for work. You send back a yes, it turns on. And so it’s all these little tiny paper cuts that actually start to go into run. Next, it’s vendor data. One of the unique assets of RAMP is anytime you spend money, you upload a receipt. Maybe you upload an invoice or an MSA. This kind of takes us back to the the Parabas days. We know across hundreds of millions of purchases every year, not just that you spent money at a software vendor, but what did you spend per seat? And so if you’re a customer on RAMP and you’re about to go and pay this large bill on random SaaS vendor, we can show you in real time before you kind of send the funds, you are paying 20% more Than the rest of the market. You know, here’s kind of the cost curve, what others are paying. (Time 0:37:30)
  • Aggregate Demand Enables Better Deals
    • Ramp can aggregate customer demand to negotiate vendor discounts or surface faster-growing vendors to customers.
    • Showing alternatives and negotiated offers before renewals helps buyers direct marginal spend to better deals. Transcript: Eric Glyman There are this large swath of businesses that I’ve been fascinated by called these group purchasing organizations. A lot of these, you know this well, I came out of that in the healthcare world where it’s a very small set of end, exactly supplier. But if you can go and aggregate demand, you could say, okay, for these type of purchases, you’re going to offer this procedure at this account for this type of equipment, you’re going To give this level bulk kind of discount, and you’ve gone and you’ve done that. And I think these are very popular now in the private equity world. When I look at ramp data today, there are dozens of, you know, more and more every year of merchants where, you know, there’s, we are sending billions of dollars. You know, it is truly, it is the client’s money. They are going where they’re going, but we have a sense of actually, okay, where is this actually going? And can you go and say, across the ramp buyer base over the next 12 months, this is how many dollars will go towards you. Can we negotiate a discount? The other way, which maybe starts to, sometimes these businesses veer closer to advertising, but where it gets really interesting is we see the fastest growing businesses. What are the businesses that people are getting really excited and are adopting really quickly? And we might have a signal of these are actually good businesses and good tools. And actually most businesses should move towards that. Could you go and actually start to say, hey, your renewal is coming up in 90 days. Have you considered this other business who has provided a 20% complimentary welcome lower price to you? And so there’s multiple forms that can take place with a short version of it is I think, one, it’s yes. I think you can offer a bulk discount. (Time 0:42:44)
  • Sell Time, Not Just Financial Products
    • Sell time and operational work, not just financial products, to create higher customer value.
    • Help customers avoid wasteful spend because saving dollars yields more leverage than small rewards programs. Transcript: Eric Glyman Explicit, I think what’s happening is the marginal cost of time, of knowledge work. That is a much better abstraction. Is what’s going down so rapidly. And I think about our customer base. Most of our customers don’t have a single software engineer, let alone a software engineer for their finance team. And if what we are very good at doing is selling functionally sets of work, maybe it’s embedded in a financial operating platform, but expenses done, accounting done, some type of knowledge Work done, and you can deliver that, that is immense high leverage value for these customers. If it, let’s say $5 before they didn’t do it, now they can get $5 of value, but maybe the cost of, it’s a software type cost, maybe it’s pennies of tokens to actually go and do that. That is a great business to be in because it’s very high customer value. We can capture just a small amount of that and build this business. (Time 0:55:04)
  • Paribus Exit And Capital One Lessons
    • Eric recounts Paribus origins and how Capital One bought his prior company.
    • He highlights Capital One’s experimentation, data-driven lending and eventual spinout as lessons for fintech. Transcript: John Collison One is one of the biggest founder-run financial firms that people in Silicon Valley don’t talk about. What should we all be, and has been extraordinarily successful, what should we all be learning from Capital One and their success? And actually just start, how did they? Like what product really broke out for them? Eric Glyman Just what is Capital One’s success? I think there’s a lot that makes them amazing as a company. Both the founders are excellent. Rich Fairbank, Nigel Morris. I think it’s worth people reading up on their stories. It’s founded in the 90s? Is that right? So, not quite. Okay. From a legal standpoint… I’m very woolly on my Capital One history. I think that legally the incorporation in some sense for Capital One, I think it was in 1994, but the actual start traces back to the 80s. Okay. At the time Rich and Nigel had met, they were consultants. And they had this insight that the business of credit cards was, well, it was lucrative, was not serving most of the country. The way you could kind of think about credit cards back then, you know, Diners Club, maybe others have heard of it. It was sort of, it was a way for like rich business people to get together and have lunch. And like they put the card down and, you know, the restaurant would pick up the tab and they could go and pay the restaurant back later. And they had kind of navigated that. And kind of the simplification is if you had a very high credit score, you could get one of these cards with the benefits that it had. And if you didn’t have a credit score that met that, you could have a debit card. And that was basically it. It was like a very kind of linear cutoff. And what Rich and Nigel, the insight that they had had is there must be some curve. We should be able to test this. Maybe we give people cards that have above an 800 credit score or something like that then, but what about 790? There might be people there that can go and pay for this. If, and the way we could go and take on the cost of this, maybe we charge them a somewhat higher interest rate until they prove kind of efficacy. And if 790 works, let’s go to 780. So it’s kind of the BNPL of its day. John Collison There is a population who are, for whatever reason, just below the threshold where banks are giving them good access to credit, where you can actually very profitably lend to them. This is right. Eric Glyman Exactly. And so they, in the 80s, were functionally pitching all these different banks to say, we should go on this exploration. We’ll even run this for you. Let us go and run this. And they went door to door to door to door to door. And the banks didn’t buy it. They didn’t buy it. And then finally, there was a bank in Virginia, Signet Bank, that said, fine, we’ll let you run it. You can come join. We’ll give you this group and some resources to go build this. So, they built this as this division inside of it. And it started to work. As this was going on, the computer revolution was taking off. And it was sort of an unfortunate acronym, but they called it IBS. It’s not- Invisible something something. Yeah, it’s information based strategy. Yeah. And they said, we could use different pockets of data and run these. Before there was big data, there was this kind of data, we could go and test. Maybe this person has a balance of $8,000 at this bank, we could send them an offer at this interest rate or this kind of, you know, you don’t need to pay back interest for six months or 12 Months, and they could start to go and see the mathematical return against this. And this division started really working, and it got so profitable, it became a problem for the business, and they convinced them to spin it out. And so, in 1994, that is the founding, as people know, Capital One today. (Time 0:57:10)
  • Treasury Will Shift Toward Smarter Yields
    • Treasury yields for business cash are low today due to incumbents; product-driven competition and smarter cash routing will raise yields.
    • Smarter capital allocation and real-time cash intelligence will put more dollars to work and lower financing costs. Transcript: Eric Glyman I think there is an incredible amount of money made by institutions who are enjoying the profit pool and sharing very little with their end customers. If you think about the implications of the federal overnight funds rate, this is basically saying, hey, if you want to go and… It’s insane that there’s so little yield sharing in the Current market. It’s crazy. I think that the national average for businesses, these are sophisticated entities with personnel. They’re supposed to be able to manage and put their funds in some place that’s more high yield. I think the national average on checking accounts is 0.07% in the US. John Collison And so I’m guessing you don’t believe that Bank of America and JP Morgan and all these folks will wake up more altruistic one day. And so what is the competitive process by which you think we’ll get there? Eric Glyman I think that new businesses, whether they have their own charter or they work with banks too, that this is not a monopoly. Everyone has competitors and it is a market that evolves. I think that that rate will go up, that the easier it is for people to, whether it’s to create depository institutions, create accounts at these institutions, or create stores of value, Maybe even outside of these systems, maybe in a stable currency. And so, I think that rate goes up. I think part of what’s driven the extremely rapid growth of ramped treasury is it’s just a vastly better deal for customers. Maybe you keep three months of just walking around money in your checking account, but if you know the funds you’re going to receive and what you’re paying out, well, we can move funds On the day payroll is coming due into that account. And then every other day, make sure the funds are earning the highest rate possible. And so I think from a macro perspective, these things tend to go up in terms of the relative yield. I think there’s another question of slack in the system. If you have more and more money can think, right? If the dollars in your company have some level of intelligence, right? Like it’s able to kind of determine when can it be spent under what circumstances. It’s increasingly recorded in real time. And there’s some reasoning around this, some ability to kind of opine of where should the next marginal dollar go. And you have systems that are able to think infinitely about these things, even at 3 a.m. In the morning when most of your team is asleep. Well, maybe you’ll determine I should keep the funds in some level in like a 2% or a 4% yielding account. But I think more of those dollars will go in flight actually to go spend to if you have a business that makes an 8% profit margin a year, that’s a lot higher than what you can earn in the overnight Rate. And so, in some sense, I think you have the dual effect- Smarter capital allocation. I think more dollars will be put to work. I agree with Alex’s macro point of if you kind of understand counterparties, you understand more information. (Time 1:07:10)