How AI can help build smarter systems for every team with Eric Glyman and Karim Atiyeh of Ramp

0:05Hi, listeners, and welcome to KnowPriors. Today, we're here with Eric and Kareem, the co-founders of Ramp, one of the fastest growing fintech companies of all time. We'll talk about the path to self-driving money, how they make AI and systems thinking fundamental to every part of Ramp, and why using AI in financial services is not as risky as people might think. Welcome, Eric and Kareem. Well, today we're super excited to have Eric and Kareem from Ramp joining us, so thank you so much for coming. Thanks for having us. It's great to see you both. Maybe you can tell us all about the origins of Ramp, how you all got started, what you're focused on today, what roles you play at the company.

0:39Ramp, in some ways, is sort of the sequel or taking our last business forward. We started about a decade ago at a business called Paribus back in 2015. I would probably reframe it now as an AI agent company. Essentially what it did is, let's say you bought something from Amazon, Best Buy, Macy's, whatever. Every store would guarantee that the next week the TV went on sale, you could get the firms back if you asked. So we built it. It was an email app. We scanned your inbox for receipts. We detected what you bought. And if you were eligible for a price drop refund, we wrote an email as you, or chatted with the store as you, to ask for the difference.

1:15They would respond, give you the difference, and we charged a cut. We launched this, within a year we had almost a million members, and we had a life-shitting offer from Capital One to buy the company. So we were thinking a lot about first how you turn data into savings. We didn't know the first thing about credit cards, we learned it. We saw it was a great, large industry, very profitable, but deeply misaligned with customers. people were obsessing over the question of how do you get people to spend more money or more points and we got very interested in the question of you know if you actually listen to people they don't want points or cash back they actually want more in their bank account the best way to do it is not get people to spend we get people to not put differently like not spending a hundred dollars in the first place is a hundred times better than getting a dollar or one percent back on it and so we got obsessed of wow what if you put paribus in a card what if you had you know card and software that was designed to get people to spend less in the first place that seems more aligned with customers, a different way to go to market, and fundamentally a better product.

2:09And so back in March of 2019, we incorporated the company. Today, we're just shy of five and a half years old. And I would say a lot of what we're working towards today is this question of can you have self-driving money in an organization? Can you build really the primitives? Can from in one place, issue cards, make payments of all kinds, manage approvals, even automate accounting. But really what we're doing is when you have the right print, it's abstracting away all the tedium. If you don't need to add receipts to a transaction, you tap a card, we pull the receipt from your email, and you're done.

2:43You don't need to tag every transaction and how it should be booked in your books and records. And so today, 25 ,000 businesses use it from small startups to Shopify to Boys and Girls Club of America to farms to anything in between. And, you know, so that's how we got into it, what we do. And I'm the CEO and run a lot of the, you know, business functions, Kareem CTO, of course, technical, but also we can get into it overseas, even, you know, marketing other functions that we see think are becoming more technical. It's been an amazing run over five years. A lot of the value you're talking about are like, as you said, for organizations.

3:22I remember talking to you guys when you were just getting started, you were like kind of discovering that there was all this value in businesses and how businesses spend instead of like your first company was more consumer oriented. Like, how did you decide to like make that shift and like learning about that audience, investing in it? I'd say a lot of the ideation phase of Ramp, we were talking to a lot of our friends, people in our community, and it just so happens that a lot of them were either starting early stage companies or joining early stage companies. And a lot of the problems that we're facing at a larger scale were some of the ones that we're trying to solve for consumers first.

4:01and the funny thing with businesses is the better they got and the larger they got the more wasteful they would get and the less they would know about their spend. Not only is it a more interesting and in some cases bigger problem to solve it just scales with success in some ways so the better companies were even more interesting opportunities for us so we went after that and there was I'd say another realization we had early around like you look at the user experiences of different products out there and the ones we use as consumers on a day-to-day like they obsess over every single interaction every single flow like Instagram's amazing Robin Hood did that very very well for trading stocks and investing and then those same people who use these apps in their daily lives show up at work and are expected to use tools that were built in the 80s and are incredibly slow and very clunky and there just wasn't a good feedback loop between like what the people wanted to use at work and what they were they were using in their daily lives and we saw an opportunity to really bring a lot of consumer thinking around like ui ux obsession to complicated business problems where did the first um savings thesis come from like what you know if if the idea is like spend less money and have less waste.

5:25For consumers on like, there's a price change, like you go deal with this, that sounds amazing. Like what was the first hook for a rant? First was just a basic business problem of, I think most startups, their biggest battle is people not giving a shit. Like you work on, you're so hard on the software and there's a million other pieces, you know, apps, tools, cards, how are you going to stand out? And we realized there was this gap in the market. Most of the credit card ecosystem was really designed around kind of ego and excess and use my card and get lounge access and points and you can fly to exotic places.

5:55And it just was so different from what business owners and CFOs and people I know who are trying to build enduring profitable businesses. So it felt very at odds. And so some of it was just, I thought there was a large unmet need. That's interesting. So just like the cultural premise of the card was very different. Yeah. It's not like, let's say like Amex fancy for the individual. The luxury that Amex, I believe, was really owning and building was almost like an 80s conception of luxury. You have lounge access, and it's covered in mahogany, and it's super cool. And luxury in the 2020s is, I can go to a yoga class at 2 p.m.

6:36because I have the time. I'm going to green juice because I'm healthy. And it was just very different. And I think the thing that was so, I think people were missing is time. People had very little time. And so we tried to focus on not just saving money, but saving time and to kind of zoom into where some of our first products was you would see these great early stage startups, really promising, starting to build traction. And then you say, great, show me what you're spending on. And you discover they're spending on like four sets of project management software or like five BI tools that do the same thing.

7:10Because someone tried it, someone passed the card to someone else, the finance person is like, get me the receipts, doesn't ask the question of what are these things. And so some of the first insights was, we'd say, well, show us your credit card statement. We'll show you all the duplicative software, all the redundant spend. And by the way, we helped you cut your spend by 2 % per year. Maybe you should use your card. We'll help you do that all the time. And so it was very simple, was the mechanisms. It was cutting out redundant spend and it was automating processes. And I think it's expanded, but sort of these same principles of, If the one thing people don't have is an hour on Fridays to spend with their family because they're doing expense reports, because they're tagging transactions, they're doing tedious reports, how can we give that back to them?

7:55How can we design products that will automate the tedious to, I would argue, gives people the new luxury of time? A lot of people say that with this wave of generative AI, there's new things you can automate or change. And you all have been very big adopters of the technology internally in all sorts of different ways. It does a little bit more about how you're starting to use it, but for internal purposes. And you mentioned marketing is now a technical question, which is amazing. How you're starting to implement it for customers or where does that matter? How you think about that as a regulated entity?

8:21So I would just love to hear how you started thinking about using AI and then what that's led into for you all. 100%. I mean, one of the early really thesis of Ramp is if we really want to help you save time and money, we need context. right so one of the things we obsessed uh a lot over in the early days is we have some amount of information from the card statement we have more information from the things that we see in your inbox we can get even more information if we're connected we're connected to your erp and you get to a point where like okay there's a lot of it it's very unstructured how do you structure it and really help companies build and automate the workflows so and that's kind of how a lot of the internal ways that AI shows up in our product really work.

9:07We really focus on the job to be done, so you want to close your books. And there are different workflows that are part of this. And we're able to work on them a lot faster and really customize them without having to think about every company individually because we're able to just apply a high-level generic AI algo with some constraints and just make sure that those repetitive tasks become a lot faster. So there's a lot of that that we do also on helping you figure out what bills to pay and when. There's often a right answer, right? You want to pay the bill at the most optimal time. Generally, that's the last day that it's due.

9:45In some cases, it might be earlier because you get a discount. And those are the things that have a right answer. They take a little bit of thinking, but there's generally a way to evaluate and test what the right answer is. a lot of great applications in in helping people just like get get peace of mind on those decisions and not have to like many look at every single invoice uh there are many cases where uh financial operators stress over fraud right and the way that they check for uh fraud is they'll have to uh essentially look at different data sources and make sure that they match that's also something that we could do like so much easier with ai like let's make sure that we have three-way matching and we can match your purchase order to your invoice to the goods that were actually delivered.

10:29So there's many of them, but the one decision that we've made early is financial professionals care about the job being done and they care about the observability and having control and a lot more than they care. Like they don't actually care that much if they're using AI. They just want it to be like fast and accurate. AI function is something that applies a lot of sort of thought and reasoning to different aspects of what build a pay or other aspects of the business. And then my sense is as well, you're also doing some really creative things internally. Yeah. Yes. Can you tell us a little bit more about how that's impacting functions across the company today?

11:05A hundred percent. I mean, so a lot of people know outside in, say the ramp is one of the fastest growing FinTech of all time, one of the fastest growing software companies. Some of this is good positioning and timing in a good market. it's some of this was like very, very early adoption of AI into augmenting the capabilities of our sales team, of our marketing teams, of our underwriting teams, all the way throughout. And I'll zoom in on one. I think that there's a lot of startups now starting to think about AI sales and automations. And internally, years ago, we had built a functionally outbound automation team.

11:41And what we had noticed is, this is back years ago, we had very little resources, but there was one sales rep who was incredible. He could go out and book far more meetings than anyone else. And we were trying to figure out, we're like, wow, if only we had two of him, this would be great, or three. And we want to understand, what is making this person so productive? And so one of the unusual things we did is we had a few engineers sit down with him and just track what he did during the day. And he'd follow kind of his calendar and turn it, wake up in the morning, and there'd be a new set of companies that raise funds or the people that he was in touch with who moved companies or all kinds of stuff.

12:19And he would form his own list. Then he would go and try to guess at people's email. Then he would try to go and send different copy out. And it turns out that actually, so he had the right mechanism, but there was lots of manual steps. And so before jumping straight to, I'm gonna hire a AI salesperson agent. And we said, let's give him an Ironman suit. Let's go. And the things that he's looking at, let's go and pull those signals so the lead list is done for him. Let's pull those emails. He doesn't actually have to guess at it. Let's run some A-B tests. He can send things, but use kind of base templates and figure out what's actually working there or not.

12:56And the net effect, as you kind of flash forward to today, the number of meetings that sales development reps at RAMP are able to book each month on their quota is multiples of any of our next closest competitors. And so it translates into a radically higher sales efficiency and the ability just to invest more heavily in scaling. And so that's part of how we've been able to continue to scale. You see similar aspects, and Karim can go deeper into how we're looking at it in aspects of marketing. But I think a lot of great marketing is thinking about a CDP, customer data platform, and understanding who people are, what's the intent, and how do you generate great creative.

13:37In the world of AI that we live in now, the cost of creating creative art has never been lower in human history. You can create amazing visuals, images, text, you name it. And so a lot of what we try to do is sort of decompose what the function is and how do we use the radical improvements of foundation model capabilities connected to data. Well, one question on this, like how do you what do you think makes RAMP different in that? I mean, maybe there's many things here, but, you know, most organizations, like the idea of resourcing, understanding the SDR function with engineers and then executing against tooling for them is just like it's not going to happen.

14:17Right. You know, division of the organization. We only have some engineers. Engineers are not interested. Like what makes that work for you guys across the org? Two or three things. First, engineers are actually interested in and gold against business problems. that actually drive a P &L, which is very different. Often in many, especially traditional companies, this is one of the things that felt like torture at Capital One of engineering and technology was a cost center. It was only an L. There was no P. There was no profit. And so it was just a question of where are costs and how can we rip things out, not how can we grow revenue and profit.

14:49And that's fundamentally different from the get-go. I think it's important for any business to think about. I think going deeper, a lot of what people think about is how do we minimize costs? One of the inherent questions that we're asking is, time is money. Where are you spending your time? Think back to the salesperson example, really what we were most interested in is not how many dollars were we spending, but where were the hours going and where can we automate? Which I think is a great framework to think about the use of AI, which I think its best use case today is productivity. It shouldn't just be, hey, where is cost and how to use AI to rip out cost.

15:27it should be how do you augment and have every hour go a lot further i think is just some of the different framing but there you should add a couple more a lot of it also like goes to like the types of people we we hire as well in the first place i think a lot of organizations will have maybe a list of 10 competencies and skill sets and post the interview you'll get in the room it's like well this person checks eight boxes but doesn't really check these two and there are two very different things that we do is one uh we really care about spikes uh a lot and we also care about uh people who wanna i don't know are like very entrepreneurial and want to do things their own way and are kind of contrarian to some extent and when you hire people like that which again some organizations won't hire because well i don't know if that person wants to stay here forever or i don't know if that person fits the capital one multi's capital one as an example or really really any other company i think at ramp i really i mean from the beginning i've always I've seen it as my job to just get really raw talent that is incredibly ambitious.

16:25And it is my job to keep them interested and to keep the company interesting for them to continuously see these challenges and be attracted to them. So there are definitely different types of engineers that we hire at most companies. It's interesting because when I look at the companies that I feel that have done some of the most interesting things over time or have been capital efficient or whatever it is, they often end up focusing on certain forms of automation early that other people don't consider. So early Google was that way, actually. The online sales and operation teams, if you extrapolated out how big it would be, by the time the company was 10 ,000 people, they would have had to add, I don't know, 50 or 100 ,000 people to deal with that volume.

17:02So they started building internal tooling early for it. So I think it's kind of a common theme for companies that are very thoughtful about this. How did you all think about what to build versus buy? Have you ever thought about actually spinning this out or offering it as a product to your customers? I'm sort of curious how you think about those dimensions of this. So I think that question gets asked a lot and like you generally get with a nuanced, like it depends answer, which is generally right. But the one thing that doesn't get talked about is we can kind of assess generally whether if you decide to build, if you've done a good job building or bad job building, it's very hard to assess when you do, when you decide to buy, whether you did a good job buying or not buying.

17:42and it's not something that really comes up in uh say i don't know a performance review or in the way like people get evaluated internally which is kind of crazy when you think about it like we talk about people being great in organizations because they're great at hiring and recruiting and but you never hear anyone talk about how this person picks the right vendors is really good i never measured up procurement that's super interesting yeah and that is one like skill set that we we like we did focus on very early the way we pick uh the right vendors is almost like essentially interviewing them.

18:12We primarily like to talk to the engineering teams. We care a lot about the slope and the rate at which they're progressing as opposed to whether they have gaps today or not. And that has served us incredibly well. Out of curiosity, why not and I know it's a very different type of business so that may be the answer, but why don't you just start offering some of these services to your customers to use the SaaS products? In other words, there's this whole wave of AI happening right now. There's a lot of companies doing what you mentioned on sdr automation the marketing stuff is a little bit behind but before i talk about it like you're basically um dog fooding products they could actually have real scale potentially in the same customer segment that you have i'm just curious like why not going off of this for the world especially when like there are a lot of engineering teams out there who um they experience these pumps pretty abstractly yeah right like they've never you know been attached to an sdr for days on end like you described so there are there are real advantages here i think it's a fair question.

19:07I would say like the... Hire us. I know, it's like, hold on. We need a strategy function. Help us do it. No, I want to come back in a week and be like, you were right. No, I mean, I would say the simplistic answer is just simplicity in the sale. Like everything we do is centered around saving you money and time for finance organizations. And when you sort of think about like the order in which we've done things, it's expense and cards instead of needing two apps to buy one thing and Amex and a Concur, you just tap a ramp card and your expense report does itself. Then you add in bill payments and you add in procurement and travel.

19:40And so it's more products that continue to help the same economic buyer. And I think there's a lot of friction from a go-to-market perspective when you're selling to different buyer groups with inside of organization. And so I would say the overly simplistic answer has been probably we just haven't thought deeply enough about it. Maybe we should. I do think what you were saying, though, is actually, I think, worth emphasizing, though like at a lot of companies i think everyone's heard the saying of you know no one gets fired for hiring ibm like i kind of should be yeah i think i kind of think at ramp you might get fired for picking ibm right like you actually want to pick vendors that um not based off of just they have every checkbox but they are properly sloped and they're going to advance and if you play this out in a year or two they are going to give your organization a fundamental advantage um and kind of you think about just what AI can do of augmenting systems like actually building great pipelines rails by which to operate is is I think never been more important.

20:40If you play it forward a little bit more right like if you have these like AI salespeople, AI finance people, AI and how do you determine whether you pick the right one or not if you're not if you're like replacing a lot of functions that you would have hired like some people for with essentially an AI professional. I don't think we have the frameworks or the tools to do that today. And also, how do you refresh that over time, right? Because often once you buy something, there's this incumbency bias. You just keep going with it. So you don't actually know what's in the market usually as well. One question on this, because valuing time and productivity feels like such a core thing for RANF.

21:19I've, you know, a lot has too, but I've been on the board of companies that sell productivity in some way. And increasingly, AI companies are selling productivity as part of the value, it's kind of hard, right? I would make an argument that a lot of organizations don't necessarily value productivity of their people versus direct top or bottom line. And so I'm curious how you guys think about it internally, how you sell productivity to your customers. We get classified a lot as a fintech company and we're fine to be in that box. I actually think we're a productivity company. I agree. Like primarily what we're selling people is time.

21:58It's expense reports that do themselves. It's books that do themselves, keep it clean, are more accurate. It's a procurement process that actually is just upload the contracts. We get the approvals. We show if there's better prices available. And you just have to worry about the thing you want to buy versus just jamming it through your organization. And I think it is productivity. And what I would say is some of what helps us is in the finance organization. Think of it from their perspective. unlike R &D and the fancy parts of it, they're G &A. And CEOs say G &A has to go down every year. You cannot hire more people.

22:33And so they're asked to do the job of multiple finance teams as the companies get bigger without new resources. And we allow them to do that. We sort of create this scale leverage for them to do that is some of the way that we talk about it. It definitely helps to be a product that's free to try, that pays you to use it, and shows you ways to cut out costs. And I think what Kareem was saying is, it's funny, I think in 2024, we're almost like in this, I think back to the 90s, probably in 97, 98, your stock went way up if you said we're whatever.com, it was really good. Then pets.com came out and it was a liability and it was really tough.

23:10And now you've got saying we're AI this, AI whatever product. Like we show, not tell is a big part of what we do. Like we talked about automation, about increased accuracy, because there's fewer processes that you need to go through. AI is there. It's some of how we do it, but it's not the way that you lead with it. And so I guess it's all to say is like we really focus on what is the outcome we're driving. How much will you save in terms of dollars? How many hours are you getting measured against this way? And can we really connect it, at least from a positioning perspective? I mean, on your point of like, how does that show up in the sale?

23:50It's like a third order effect. But like we think that the companies that pick ramp will become more successful, will grow to in some cases become bigger. And as a result, like we will make more revenue. So I love it. Natural belief in your own customer. But I mean, there is. I mean, like we're starting to have like statistical significance around things like like underwriting risk and like risk of going bankrupt. up then it's a lot lower for for ramp than like anyone that we've seen out there and we think i'm sure some of it is like the types of of customers that we attract uh we think we have a brand that probably attracts people that do care about running a good business uh so that helps but also like the the hope is that like if you're on ramp you're more likely to run a successful company okay so you said you're a productivity company not just a fintech company yes but you are still a fintech company yes uh and you know like i talked to a bunch of um like large customers including like financial services like traditional financial services players and uh unfortunately i would say like a level of adoption and use cases that matter is still like marginal in most of these businesses um and the the reason i you know i'll hear from let's say large bank capital one type company without naming them is uh you know it's too risky right like uh ai doesn't work at the level of reliability on the workflows and use cases that we care about and like something something regulators compliance is important i know you guys believe that but you know i like i call kareem and he's like oh yeah we're trying this and like it's you know it's useful in this way or we're building something internally um how do you guys figure out how to take that risk when you know your processes your product have to be like financial services robust quality i honestly don't think there's that much risk if you constrain the problem well right because at the end of the day so like if you're trying to figure out like how to use ai to help you categorize transactions it's not an open-ended question it's like pick what like make up categories it's like hey i know what the right categories are like tell me which one of these could it be and the way this could show up in the product is every single form is pre-filled and you can edit it if you need to every single list is pre-ordered and in many cases we can be better at asking you just a question that matters as opposed to like like the same form that we ask you every time right so like you might be traveling and you went and got a launch we don't have information to know if this was a launch with a candidate or a launch with a customer we can ask you just that question as opposed to, I don't know, how many people were at the event.

26:26That's something that we can answer because we may have access to your calendar and things of that nature. So it really constrains the problem a lot more. And you're not just having AI do completely unconstrained work. Yeah, it kind of categorizes risks in businesses or at least ones that take risks from these sorts of directions as known risks and known risks. And you all have known risks. You know what you need to do. you know where things could go wrong you can fix that i think usually it's the unknown risk that cause real problems for companies and so it's you know you have such a sweet spot in terms of what you build that i think that minimizes that that sort of fear of something really bad happening which is great well i think this is like this um ability to define tasks that make sense and understand like performance against the task internally um is actually i think like a like a very strong competitive advantage in application of ai today and on the productivity point it's like we all i think can agree that like ai's got an incredibly good at translation right like you could there's some areas where like okay you're translating english to code and those could improve but like broad translation it's pretty good the one type of translation problem that we see ourselves solving all the time is like accounting finance speak to like a business uh it's generically bad at math too right yes certain basic math just breaks which often gets into financial related items.

27:44Fair enough. But there's not a lot of that that we have to do, frankly. So can you effectively, are you thinking of fine-tuning a model against certain accounting terms? I'm sort of curious how you think about problem solving or is it just wait for future generations of models to come out? We did spend some time fine-tuning in many places and then we very quickly found out that our time was worth way more and that we should just wait for other generations of models. What we've gotten really good at, though, is hardening our infrastructure so that we can easily switch when we need to and we can quickly evaluate the different models on the on the sub tasks that we care about like i was asked this question by by one of our investors recently was like with like the gpt4 mini how how has that changed things for us he has it brought costs down and how are we thinking about it and my answer is like oh yeah it's it's already in production and for like 90 of tasks that we're running it's good enough so that was a quick switch and within a day we can know very quickly like that yes this is that good enough and we have the right evals.

28:43What is lacking right now from the AI stack? Like if you were waiting for one thing to happen, is there anything specific or is there specific functionality or capability or just someplace where it tends to fail? I'd say like we have been exploring using AI to help you in essentially like end-to-end navigation of like the website or the app. And that's really hard. It's a cool demo like that I've seen. We'll link it in the podcast notes. Our app is changing so quickly and you need to figure out like where do you have like guidelines and constraints and where do you let it be a little bit more open-ended um there is a fun one that i i really like although it's not like quite there in production um and as eric was mentioning earlier like i'm really focused right now on like what lessons learned from really engineering and using ai in our product can we apply to other job families and processes that we run and one of those is we we write a lot of copy we send emails to our customers and at most companies there's some kind of process where like someone needs to review the email to make sure that it follows your brand guidelines that it has a clear cta and those are things that like in engineering you could it's more deterministic you could do that as part like a test suite that you run during your like ci cd flow what if we could build something like that for the copy that we're sending out and we are starting to to iterate on this is like every single email that is going out for the first time can you run it through like an an AI review that is a lot more close to instant and as deterministic as possible, although that's hard.

30:10There's definitely room for improvement there. But I would say the improvement is more around the interface and the knobs that you can use to tweak your model because today it feels like the most common way is just like write different prompts and longer prompts and like that's the main way that you can guide the models. I think Claude with artifacts, it's something really, really different that I love. and it's maybe a new, I mean, it's a different interface. I mean, it's really outputting a mini web app for you that you can tweak and change. And like, we are obsessing over like what the right interfaces are for us.