Risk & Underwriting For Business Lending Deep Dive

‍

Watch our panel with Srinath Srinivasan from Ramp where we dive into trends in risk and underwriting for business lending, how to think about your digital credit strategy, best practices in today’s market environment, how digital lending for businesses has evolved and more.

‍

Transcript:

Awesome. Well welcome everybody to today's panel with Head of Risk from Ramp Srinath and really excited to have him here today to dig into all things risk and underwriting for business lending, different trends that he's seeing, the modern fintech stack and a lot more. So to kick things off, Sri, welcome, we're excited to have you today. Thanks David, I'm super excited to be here. Happy to be jamming with you on on all things risk today. Awesome. Well, let's just get started with getting a quick background about yourself.

How did you end up at Ramp, what do you currently focus on and what has that journey been like? Yeah, I'm the head of risk at Ramp and I think most of the people on the call would know at this point in time, which is like think about the risk appetite for Ramp. Think about risk strategies when it comes to underwriting businesses that want to be on Ramp's platform as well as thinking about once they are on the platform, how do you actually manage the risk? That's what me and my team actually focuses on. And for those who don't know about Ramp, Ramp is building the next generation of financial tools from corporate cards, to invoice management, to expense management, to integration of accounting software. We started off as a corporate credit card product about two and a half years ago, but then we have expanded the suite of products and the latest one is Flex, which allows our customers to push out payments to their vendors by 30, 60, 90 days. So we pay their vendors and they can pay us back over 30, 60, 90 days for a fee. So helping them manage their cash flows. I think that's what Ramp is focused on.

Right before Ramp, I was at Marcus Bay, Goldman Sachs for about five years.

We called ourselves a startup instead of a big bank, instead of Goldman Sachs. So we built out a bunch of products, start off the personal loans product, moved on to Apple Card and then a couple of more products. So most of the focus before Ramp at Goldman Sachs was building new products, thinking about it from a risk standpoint and then scaling them over time. And before that I was Capital Loan for about ten years, focused on the consumer credit card space, been a data analyst, been on the business analyst side, I've been on a second line, so mostly in consumer lending. I've always wanted to be in a real startup and I think I call Ramp a real startup, it's real fintech. All of these learnings over the past 15 years has kind of led me to this role and I'm super excited to be at Ramp and it's been a fascinating two years for me. The amount of learnings has been incredible and I keep learning something new every day.

Yeah, and we're very excited to have you given all your decades of experience and risk and underwriting and we're happy customers at Ramp and we're also happy to be working with Ramp at Rutter and so excited to dig into that aspect of things down the road as well. So, yeah, curious, what does your day to day look like as head of risk at Ramp today? What do you think about, what are you focused on? Yeah. I think the interesting thing about being in a startup is that in a matter of minutes at ours, you go from being in the weeds of making a credit decision on a high dollar exposure. Whose financials are looking a little bit iffy. And do you actually want to onboard them to Ramp? To actually thinking about where is Ramp going to be in the next five years and what is the study of products we need to be building? What is the platform that we need to be building, and how do we think about risk appetite? Where is the macroeconomic environment going? And how do you navigate through that macroeconomic environment? And how do you put in strategies in place, defenses in place, so that actually we come out of it successfully. So the fun part of being in a startup is you're in the weeds for a minute and then you're thinking about strategy the very next minute. And so my day is a mix of working with people on the team, thinking about data and analysis and insights and how do we use that to create more value both for Ramp and for the customers of Ramp, as well as like unitary decisions that we make on a, on a day to day basis.

For sure. And I'm curious, like, what are some of the big lessons around risk that you learned from your time at Goldman and Capital One that both you're bringing over to Ramp and how is that thinking also changed over the years? Yeah, so Capital One was largely established organizations with like tens of thousands of employees. Goldman Sachs was large as well, but Marcus was a very small unit. When I actually started, we were about like 30 to 50 of us scale to more than 1000 people by the time actually left Marcus and again at Ramp, again start off as a very small unit. So many lessons learned across the board, from building organizations to manage risk in these different environments, to building risk platforms, to building data and using data science to make these decisions. But the few that I like to highlight, I'll start with building organizations and the thinking that has changed over time, right. For a place like Marcus at Goldman Sachs, I think we needed people who were, like, deep experts in specific areas, because the bar for a company like Goldman Sachs is really high for getting things right from a customer experience standpoint.

From getting things right from a regulatory standpoint, from getting things right from the risk, appetite and the losses that you want to manage to so we needed people who are deep experts who can come in and build their own organizations and their own teams and help us build and scale products in a very robust way. But when you take a company like Ramp, Ramp allows you to take a few more risks. So what you're looking for are people who are knowledgeable enough, but they're not really entrenched in the way that a particular business operated in the past. Right, and they're looking for new ways to solve problems. They're looking to challenge the status quo, if I may. So you want people with agency, you want people with some amount of knowledge, but they can go in and try to change the way we make decisions and change the way that actually you run a product and we build a product. So different skill sets of the kind of people that you hire.

And at Ramp we call it, we look for slope and not for the level and we look for what somebody can be in the next three to five years and what kind of an impact they can have at Ramp. So from an organization structure perspective, we are looking for very different people to actually hire. So that is one big learning over the past six months, I would say, for sure. And I'm curious, how do you balance that need for folks who are like experts given risk? It's like one of those very knowledge based, expertise based type of thing with that slope and how do you think about that trade off? Yeah, to be honest, I think we usually jerk that solving your risk is not rocket science. It actually truly is not. And I think the amount of people that we have hired in the past and Ramp who have absolutely no risk experience but have an analytical structured thinking, problem solving mindset that if you give them the space and if you tell them the problem that you're trying to solve, they apply newer techniques that traditional risk analysts would not even think about applying. They think about solutions that traditional risk analysts would not even think about applying and that smooths the needle in how we make some of these decisions.

So I would say it's important to have enough people who have a knowledge of risk and how to actually manage risk in the past. But I wouldn't say that is the be all and end all of looking for talent. There are really smart people who can come in and completely change the game for you, for sure. And then in terms of big trends that you're seeing today when it comes to risk and underwriting, what would you say are some of the big buckets of things that you're seeing and I'm serving and paying attention to? Yeah, let me focus a little bit on the BDB space and the small business and commercial space. The main thing that has been happening over the last, I want to say like four to five years is the digitization of data. Small business and commercial underwriting and risk data has always been a little bit behind consumer data. In the consumer space, you can get a lot of Jews out of your credit bureau data, which is where most people make decisions.

And with the advent of banking data and plan, I think that has plugged a little bit of the gap in the underserved segment in the consumer space. But in the commercial and small business space, a lot of the underwriting still uses accounting data and financials through PDFs and Excel documents, and people are parsing out information when they're actually making decisions. But in the last year to two years, I think the number of solutions you have in the market of digitizing this data has grown phenomenally. Right. Banking data obviously has been one. You can create an entire cash flow slate miss statement in a PNL or a proxy of a PNL just using banking data. If you have the right banking connections, there are vendors who are actually tagging the data in the right way and creating these for you, and I think those add a lot of value.

All of accounting data with APIs through accounting software like QBO and Sage and zero can now be digitized. Something that will take hours for an underwriter to actually parse out from documents can now, in a matter of seconds, be extracted and used commerce data with the platforms like Shopify and Amazon. I think Ruth started in that space as well, like digitizing that data. And that getting into the mix now allows you to have a lot more of the puzzle pieces of underwriting a business in a digital format, which also now allows you to build models that would predict delinquencies, that will predict the usage of the product in a more efficient manner. So now you're gathering the data in a much larger scale than you would and are able to build these things more efficiently to get to better outcomes. So the digitization has been a major changer, I would say in the last four to five years, especially in the commercial bureau space. I'm curious how much time you end up investing and analyzing that data, like putting together models.

What's your philosophy around that? Yeah, that's a great question.

We live in a world where resources are fairly constrained, especially if you're in a startup, and so you need to be very clear about where do you actually want to put those resources, right? And we have been on this journey where our early phase was getting as much of data as possible made available to us. So we spent the first six to twelve months connecting to newer data sources, extracting those set of data sources, and making sure those are available for us for analysis. The next phase of our journey has been to leverage data science and leverage the modern techniques to kind of extract more signals. And from there develop and scale out models for us. The second part of this journey also has been like for a fintech like Ramp, the volume of data that we have access to is limited. The kind of data that I spoke to you about, for example the banking data or the commerce data or the accounting data, you need businesses to come apply to Ramp and the data is collected as a part of the journey in the consumer space. The way a lot of the newer companies have built their strategies and policies is they would go to the bureau and say give me anonymized data for millions of records and then what happened to those in the next 6, 12, 18 months and what was their performance? And for you to build that model from the vast amount of data becomes very easy.

Right, but when you're talking about a startup where you're collecting data as the applications come in, the volume of data that you have access to is very limited. So that's why I think the techniques of extracting more information from fewer data points becomes critical. And I think that's why data science again is something that is very nascent in the small business and commercial cost base and not a lot of people have really cracked as to how to extract more juice out of limited data. We have some really good data scientists who are actually giving it a good crack at this point in time. That's awesome. I'm curious, with this digitization and proliferation data sources, what does that allow you to do? What's the output and the outcome that you're shooting for and would love to dig further into transitioning there like how you're thinking about it. Yeah, absolutely.

Let's break it down into a few components. The first one is digitization of data allows you to have more real time data connection to your existing customers. So for example, once you connect a commerce platform through Rutter, as long as our customers give us the permission to keep that connection live post onboarding, you have that information almost on a daily basis. Now you actually understand the revenue that is actually flowing in. Same thing with accounting data. So Ramp as a product allows you to close your books much more efficiently when you integrate your accounting software with us. So again, we have a better understanding of our businesses and our portfolio.

So the digitization also creates ability to understand businesses in a more real time manner, which also means that you can manage your risk in a much more efficient and effective manner than like in a large bank. We are waiting for every month for a commercial credit bureau data to be available to you. The 30 day lag is significant and the world is shifting towards more real time risk management. I think that has been a big mix for us for sure. And something we touched upon in the prep session was this idea around Bespoke experiences and like personalization of different types of experiences for different types of businesses. I'm curious, is that something that you think we're going to start to see given? I think we've seen that on the consumer side of things where the last two or three years has been a proliferation of all these different banking products and like fintechs, I'm curious, how do you think that would go for the business side of things? No, I think the policies and the strategies in the past always have had some amount of segmentation. But I think with the ability to do this in a more real time basis, with the ability to do this in a more digital basis, what it allows you to is get more granular in terms of how you actually develop these policies and maybe strategies.

So for example, if you think about ecommerce businesses, you can start to get granular as granular as cash conversion cycles and each business having a different one and not just tailoring your policies to it, you can also start tailoring your products to it. There are places where if the cash conversion cycle is a little bit longer but the consistency of the payments are very clean, then you can actually give them a bit more float, which is the flex kind of a product that actually comes in knowing that these businesses need a 30, 60 or 90 days to pay back. But it's a very stable cash flow that is coming in at a later point in time. So creating policies and creating strategies just to meet these kind of segments and needs also becomes available to us. So I think that is one area. I'm sure that there is a lot of research and innovation that is happening for sure. And then switching gears a little bit, I'm curious, what are you seeing within the fraud side of things? How does this additional data impact, what are other things that you're looking at when it comes to building out the fraud programs? When you're starting to lend these massive amounts and then starting to take on more risk as well? Yeah, similar to the credit side on the digitization process, I think on the fraud side, I think most of the admins have been a lot of new vendors and a lot of new players who are thinking about fraud risk in slightly different ways.

And all of these are built in the modern tech stack with the ability to integrate with these solutions in a very efficient manner. And most of these solutions are also not very expensive. Right. And I think fraud is very spiky and usually the way it happens is if there is one fraud event, you can be easily taken for a ride for a few hundred thousand dollars to a few million dollars very quickly. So a vendor who can try and block it for tens of thousands of dollars, the return on investment is significant. So your ability to use multiple vendors to kind of solve for fraud events becomes a lot easier in today's world. I've used this analogy of Swiss cheeses in the past because both for credit and fraud risk management, you don't have to think of just blocking it at the very first step of when that incident could happen.

So there are different players, places and different layers in which you can put protection in place and you can have a little bit of a few things floating through, but if you can catch them early enough so, for example, if a fraudulent business, a synthetic business, or somebody has stolen another business identity slips through the underwriting process. But if you have early transaction monitoring and you're able to stop the transactions as early as possible, the dollar losses are not very high. You can stop them at the time of the payments being made and stop credit bust out risk from actually happening. So you need to have controls in each of the places and then you have different dials that you can turn around to optimize the overall flow across customer experience losses and the growth that you actually want to get to. So that becomes very interesting. And I think fraud is one area where the advent of newer vendors and newer solutions has been prolific and that has been adding a lot of them, for sure. And I'm curious, with all these new vendors and automated solutions like additional data, how do you think about relying on those systems versus continuing to build out your dedicated fraud and risk team and risk analysts, fraud Analyst I'm curious, how do you think about that risk? How do you balance those two things? Yeah, of course, that's a great question.

I think one of the learnings for me, being in an early stage startup and building out an organization, is that your organization structure and your roadmap of what you're building out changes every three to six months. When we started off at Ram, I think we were a team of two to three people with the same person probably doing underwriting and customer management, credit and fraud risk. And so the expertise that one person actually developed across the board was incredible. Right? And we had humans in the loop for every single of these risk processes at the scale at which we are operating and the volume of information we had at that point in time, it was okay for us to have humans in the loop. In fact, we learnt a lot where, when the volume is small and you can't build models to actually predict this, you can't use algorithms to actually get to the better outcomes. Humans and human brains are incredibly good at parsing out higher risk and lower risk. And making these decisions and then eventually taking those ideas and taking those insights and then digitizing them has been one of the ways for us to move forward and actually become more efficient over time.

The way that I've started thinking about this is in an early stage, more humans in the loop, more people and more eyes on making a decision, but over time change it to digitizing of that information, expand and grow and then get into algorithms and get into smarter decisions. So get people who can do multiple things at the same time but they can then actually build this automation systems for you. That has been the gameplay for us for sure, yeah, makes a ton of sense and switching gears into just some of the challenges that you guys are facing today. But as you mentioned also as a startup, we're always constantly adapting. A lot of our audience is probably thinking about how should they shift? Given today's market environment and given you've seen post eight like, what would that look like? We would love to hear more about what your strategy is like, how do you think about things? And as head of risk, what are the things that you're coming in on? To say that the macroeconomic environment is a bit volatile today is an understatement.

The challenge with navigating a macroeconomic environment is that you are predicting certain outcomes and it is very difficult to get all of your predictions right. I'm not saying that you should have your point of view on what is going to work and what is not going to work, but for a larger company, they have cash reserves, they have deposits where they have easy access to cash, right? So when you have these set of predictions and when you actually build on those predictions and if they go wrong, the downside is fairly protected and the cost is not very high. But for a start up like Ramp where the availability of cash reserves for us is much more limited. So when we take some of these bets and if these bets turn out to be wrong, the cost could be very high. So I think of it in two parts. One is an understanding of the macroeconomic environment. What are the key vectors that you actually want to monitor? What of these vectors map out to a particular segment or an industry that could get impacted and then just having a point of view on where and what metrics and what levels could shift could lead to an outcome change in the macroeconomic environment.

That's the external part of the monitoring system that you need to build. Then there is an internal understanding of your own portfolio.

What are the segments that are vulnerable to you in your book? How much of an exposure does actually sits in these kind of vulnerable pockets? What levers do you have in case things turn south? It depends on the products that you're building. If you're building a personal loan, if you're building a loan scan of a product, a six month, twelve month, 18 month loan kind of a product most of your levers are at the time of underwriting because after that the money is out the door. And if you think that the environment is shaky, your investment has to be more in creating payment plans to ensure that the customers can still survive if something actually goes south and your focus is there and your focus is on pulling the underwriting if you're building a credit card kind of a product. So the question is how much exposure is sitting out there? And can you reduce your exposure in vulnerable segments so that the dollar losses is not very high? Credit limit decrease strategies historically have never actually worked because to the businesses that are not doing well, they would have already used the limit. And for the ones that you actually pulled back, you may end up like pulling back for a lot of the good customers and they're actually not happy about it at all. So there are levers that exist, but how efficiently can you actually pull the levers? So those are tied to your product, those are tied to your segment, and how good are you in terms of understanding that becomes key. So having a playbook of here are my vulnerable pockets, here are the levers that I can actually pull.

For me to limit my exposure, my risk in there becomes important, then it becomes a question of risk appetite. And the last piece of the puzzle in my mind is for an early stage growth startup, you have to ask yourself two questions what if I was wrong in predicting the timing of a recession? Right? And what if I was wrong in saying that I called it too early? How did it actually affect the growth of the startup? And what is the opportunity cost that is lost? What if I called it too late? How much incremental losses would I have taken by continuing to operate in the space with my existing strategies? And what is the trade off of those two outcomes and which one can I live in at this point in time? And that is an evaluation that you need to do every month or every three months, especially in this kind of an environment. And at one point in time that sequence will flip. And I think that's when you start pulling some of the levers, for sure, yeah. And I'm curious, what would you say is different about this time or is it going to be the same or I'm just curious in your point of view, and that obviously we're not Oracle, but I'd be curious, what are your thoughts on where the market is headed and how it will impact all these startups? That a lot of folks may have raised a ton of capital in the last two years, but we'll be looking to go out and either try to extend that runway or raise more capital in the next year. This is probably one of the most overused lines in the industry, I would say, about recessions. Every recession is differently, different segment gets affected, the depth of every recession is different, the duration of every recession is different.

So from that standpoint, I think it's going to be like we were saying, I think you need to have a point of view and you need to have a prediction so that you can think about different options if a particular scenario plays out and what is your playbook in that particular scenario. But outside of that, I think if you are a start up trying to think about managing your business through a recession, the few things I would recommend, I think, I'm sure this has already been debated by other people as well, is like having an understanding of your expenses and your costs is key. You need to know where every single dollar that you're spending at this point in time, and if you don't know that, get some resources, have them do some research and find out where the money is going. Try to have an understanding of the return on investment of every single dollar that you have, so that you have a rank ordering of what you can start cutting out in case push comes to show. And it's you need to save some money. I think that's the second part of it, then. The third part of it is in terms of your own underwriting strategies and policies.

If you're in the world of lending money out, having a tiered approach of what is the first ten businesses that you may want to offboard, what is the next 100? What is the next 250? Already having a set of things that you can actually do again becomes key. So planning is a major part of it. The timing of the execution is more art than science. The playbook itself is a lot of science, for sure. Before we dig into talking about the tech stack that you're using and how you think about that, we've got a couple of questions from our audience that I think are actually pretty good questions that we should dig into before we move on. So curious, like what learnings from your time at Goldman Sachs and Marcus have you taken and applied with risk management and underwriting at Ramp? If you would love to hear about the specific ones that come to mind. I know we touched on the organization building side of things, but would love to learn more about the specifics for Risk Management.

Yeah, I think Ramp and at least most of what I did at Marcus was on the consumer side and Ramp is in the B to B side. So slightly different worlds in terms of taking the learnings from it. I'll detail a few differences outright. I think on the consumer side of things, there was a way for us to experiment and learn from it. There was a way for us to experiment from pricing and price elasticity. There was a way for us to experiment in terms. Of running foundational tests or finding pockets of opportunities for us.

Because in the consumer space there is scale that you can actually achieve and you can get insights from it and make decisions efficiently. But at Ramp, with the small business side, if we were to run experiments, it'll take us years to collect the data and actually make decisions. So when you're in that world where you cannot collect the data so efficiently, you have to take strategic bets. And that's where let's take one or two steps in the right direction, expand the policy, collect some data and then course correct if needed, becomes a more efficient way of actually learning about risk and growing the business than actually running broader experiments. Eventually when you get scale, you start experimenting as well. But it's a bit of a staggered approach, I would say. Absolutely.

Awesome. We have a couple more questions coming in but we'll save those for the end. So let's move to the next section where let's dig into the underwriting tech stack. And so curious, when you first joined, whether it was like at Goldman or at Ramp, how did you think about building out the tech stack for risk and underwriting? Obviously you've seen like a proliferation of startups and vendors the last few months, few years and so would love to learn more about how you thought about that. Sure. So almost all of risk management is a collection of data, having a set of rules that applies to the data leading up to a bunch of outcomes. So most risk platforms start with a question of a decision engine because a decision does that for you.

And I think if you're building something from scratch, the question you need to ask yourself is do you build a decision engine in house or do you actually buy it from somewhere else? Right? This question will go into the engineering culture that you're building, the product culture that you're building, and the risk culture that you're building. And this is a collective decision between these three teams. I would say in making this decision, if you decide to build it in house, the thing that I would caution you on is switching from an in house solution to an external solution when you're moving really quickly is almost like you're changing the engines of a car when you're actually driving and it becomes incredibly difficult for you to do. So this might be a decision that you may have to carry for a long period of time. So think deep and hard. Not long, but deep and hard as you actually make this decision. So I think that's the first decision that I would say once you've made that decision, if you go with external third party vendors, it becomes a slightly easier world because they come in with a lot of connectivity to the data sources that you will want to use.

So a lot of the focus becomes on how do you store the data in house to make sure that you can actually use it for analysis, model, development, and execution as efficiently as possible. If you go with the approach of building it in house now, it becomes a question of how are you building it? And there is a roadmap for you to actually build it. Because if you want to connect ten data sources at the same time, it's the same engineering team that is going to be building this for you. And you may not be able to do this overnight. So you have to have a clear prioritization order on what is the first data source that you after, and then the next one, and then the next one, and the next one. Having said that, I will also throw this in there if you're starting that early, every data source is big enough for you and the delta that you're going to get is massive enough for you. So if you miss a sequence, if you get it wrong, it's okay.

The price that you pay is not very high as long as you have a roadmap and you keep making progress and you don't delay the next onboarding of that information. Once you have this data, there has to be a piece of a point in time where you have to make the decision on is the engineering team still building the rules for you? They're coding the rules for you. Or have you reached a place where the pace at which you're generating policy, the speed at which you're changing your policy is so high that the engineering team needs to build this rules engine box for you to go in and deploy the rules for you and that becomes your second major timing of a decision point. And once you arrive there, I think that is going to be a major conversation between product and engineering and risk to build it out the right way. But yeah, I think those are the two approaches. I would say there are a lot of solutions out there that gives you access to data very quickly, but if your engineering stack wants to own it, that is definitely a viable option. And I'm curious, like, how do you think about your relationship with the product and engineering team at Ramp and your previous companies and how has that evolved? What have you learned about best practices and how risk should be working with product engineering and the rest of the organization? Yeah, in the fintech space, I think you'll find two types of fintech.

You'll find finance companies that are doing tech and you will find tech companies that are doing finance. If you're in the second one, I think I see Ramp more as a tech company. Doing finance is your engineering is at the heart and core of it. One of the reasons why I joined Ramp at the beginning was Nick, who is an amazing partner for me on the engineering side, had head of Risk and Finance engineering has his title as I was actually speaking to him. So there is an entire team that is dedicated to just filling out our risk stack and that kind of a culture gives you immense comfort that you will actually build it the right way. We have a product manager, in fact we have two product managers who are purely dedicated to risk and this also means that when we build products, risk is a part of the thinking. It is embedded in the way that we actually build the product itself but it also makes it more efficient to build controls in place.

So ideal approach is for product risk and engineering to be a single team with a single focus, ideally with a common set of KPIs and OKRs that actually they want to go solve and then they create the roadmap together and then go and play it out. Oftentimes it becomes risk has its own roadmap, then you need the help of product and engineering to build it out and product just serves the point of translating risk of minds into engineering requirements and that is a very painful world to be in and I don't think any of the three teams really feels comfortable with it. So I think the more and more you can build a cohesive unit of product engineering risk, the more efficient you're going to be for sure. And I'm curious for some of the new products you guys just launched, whether it's like Ramp, Flex or even the sales based commerce underwriting product. Like how have you guys approach thinking about what new products and innovations can be built on top of all these new data tools and also just macro trends happening? I'm curious, how have you approached that and prioritizing all the thousands of things that you guys could do as a startup? Yeah, I think that's always a difficult question to actually answer and that's why startup also give you the benefit of taking some of these bets and actually learning from it and then building on top of it. Right, but in this particular example, if I take Flex as an example, for us to develop a Flex like product we need to have certain data sources available to us and risk and underwriting is in the middle of this crisis because so far Ramp has lent somebody for up to 30 days. The Chargecard product is a 30 day product and you need to pay back.

We are now going into 60 days and we're going into 90 days at this point in time. So if you don't have a good underwriting and risk management stack, it is going to be impossible for us to develop this product and scale this product. So the product manager who came up with the idea and who developed this is the same product manager for risk. So Sam Packard is an incredible PM on the Ramp team and I think he drove this product development. He also worked on developing the policy with us and helping us execute the policy in the way that we actually wanted. In fact, the ideas for some of this was embedded in the previous generation of underwriting models where we wanted to use the same models to see if we can evaluate businesses with different products with different levers that we could be pulling. So that way it's a consistent experience even for our customers that they are actually not saying that you're approved for this, but you're not approved for this.

And all of that starts becoming a little bit messy from a customer standpoint because they don't care where they are putting their money and they should not I think they should have the flexibility of doing it. So embedding the risk and the product teams together also helps us innovate this way and build products and controls in the way that we want to. For sure. That's awesome. And so before we open it up to the Q and A, last question on my side is we have a lot of product managers, founders, engineers in the audience today. And I'm curious from all the challenges you're seeing and being so deep into space, where do you think, are there opportunities for infrastructure companies to build or fintech data companies to be building it? And what's on your wish list of things that you wish you could just snap your finger and get access to it? I'll still focus on the BDB space.

I spoke about digitization of the data upfront right. The challenge with being in the BDB space is that you will never be able to get to 100% of digitization, right? There will always be this bank which is so small that somebody has not figured out an API connection for them to get you the data. There is always this bespoke accounting software that you can't actually digitize and get an API and get this information. So there is always this part where it's going to be driven by documents. And I think one of the other obvious developments has been the OC ring of documents and reading that information out.

Inscribe is one of the partners that we work with, an amazing partner. But I think the space and the scope of what we could be doing in that space is significant. There is so much complexity in there and I think there are some shortcuts and there are some innovations there that can actually get us to better outcomes. The fact that the businesses and these products are not being developed together and they're being developed independently means that it's a little bit more tricky to actually get to more cohesive solutions. So I think partnering with these vendors in a much earlier stage of product development, the same way that risk and product inside a company develops a product, if these companies are developing products for solutions directly with some of the fintechs and the lenders that they're working with that can move the needle much more faster. And I don't think there is a lot of companies that do it at this point in time and I think that would be an interesting experiment, I would say absolutely awesome. Well, we have a lot of questions in the Q and A.

It's definitely one of the more engaged audiences we've had and so let's roll through them. And so the first question is can you walk us through the thinking of partnering with Capchase for Ramp Flex versus building it fully in house? Yeah, I think it's a complex problem to actually break out and I think the way we've been thinking about it is there are short term needs and there are longer term needs. And I don't think Ramp in the space of thinking about longer term needs from where we are with the charge card products is a 30 day product. Something like a Flex is a 60 90 day cater for product. It's a small extension of the world that we are actually occupying at this point in time. That is one and the second one is the bill pay product that we built is an incredibly intuitive product and adding something like Flexing an invoice for the next 30, 60, 90 days also is a very natural way of us like scaling a product and giving more options to our customers who are actually using it. So it's a bit of a strategic thinking around the areas of our world and what a few degrees of separation versus where some areas are like far away from other areas of separation from us and taking those baby steps? I think that's been the thinking for sure.

Yeah. Makes a ton of sense. The next question is, from your experience, do you see any hesitation or pushback from end users connecting their ecommerce or accounting systems in order to receive some of the new products or benefits of it? Yeah, again, a great question. I think over the last five years I would say, I think businesses and consumers have started understanding how some of these work and have become more comfortable actually providing this information. Right. The first part of it is it is not like underwriters do not collect this information. When they want to make this decision, they are going to be asking for this information.

How is your revenue growing? What portion of the revenue is coming from ecommerce versus brick and mortar stores? These are very traditional questions that underwriters actually write ask for. And for a lot of the businesses it becomes painful for them to actually provide this information in a manual way. Right. So the digital availability makes it easier, less friction through the process to actually get that. And for an ecosystem that like Ramp is looking to build, where we are looking to save time and money for the businesses making things more efficient. The connective ecosystem that we built is helpful for both the businesses who scale on Ramp because that allows us to give them access to more credit as they grow and as they scale. Right.

So the benefits of us knowing that information and being aware and being a step ahead also helps the businesses also stay ahead in the game and get access in the right timing. So it's a bit of a win win in my mind, but of course there will be business, we will not be comfortable, we will always have fallback options when that is not available to us. But this definitely opens it up for the people who are comfortable with it. For sure, yeah, in classic Ramp fashion. I'm sure this is something that you've tested and at least I know on the other side it's something that hasn't had any impact in conversion rates, which I think is very exciting to know. The next question was what type of risk management is involved in allowing companies to send money internationally or receive money internationally? I'm not sure if that's a piece where Ramp or your experience at the moment and you draw from. So if I separate out the risk management principles and thinking about approaching it, I think the few vectors for us to obviously think about is there is currency risk and how do you actually manage it between the timing of when a payment is made and when things are collected.

So there is that risk that you need to manage. Fraud is another risk that actually opens up once you go international, the exposure to fraud risk and the vulnerabilities and the visibility that you get becomes higher. So you need to think about that a bit more differently and have the right amount of protections in place, I would say. And if you do go into lending in different countries, obviously the whole regulatory flow of what you can do, how you can do that, each of it becomes expansive if you're purely thinking about payments and the movement of payments of employees who are international. Because I think with COVID the world has actually opened up. We have customers who are based out of the US and it's a US driven company, but they have employees who live abroad, so they need to be paid abroad or like they have to have expenses abroad. So that part of it is still you don't have the underwriting risk components to it, you still have to think through the fraud components of it from that standpoint.

So depending on the product that you're building, depending on how deep you're going in international, the vectors start changing. I think that's the high level view but like happy to jam offline if you want to reach out for more details and think you through the regulations and strategy. That's a completely different world. Yeah, that's opening Pandora's box for sure. Next question is what have you found most useful in predicting credit quality, accounting, banking, or ecommerce data? Oh, this one's a hot one.

I'll go back to the bespoke nature of creating policies and strategies for different segments. I think there isn't a single size that fits all, right? I think if you were to take an early stage, startup accounting data is not going to be useful at all. They're probably not even keeping good books at this point in time. So banking data, have they raised funds? Who has actually invested in them? Those kind of questions become a lot more relevant in that space. If you're thinking about ecommerce data where they're not going to keep a lot of money in the bank, the cash is going to be incredibly thin. So it's more driven by revenue and cash conversion cycles and expenses and marketing and written on investments on their marketing spend. That is where the accounting data and the commerce data can actually give you a little bit more information on how do you actually do this.

If you think about very stable businesses that have been operating for the last 20 years, commercial credit bureaus could be a good fit because there is no way that they have not borrowed money from different lenders in that space and they have a track record of doing it. So there isn't a one size fits all. You need to go bespoke with the industry that you're working with, the sector that you're working with for sure. And this next one is a fun one. Which predictions went wrong as you navigated COVID and post COVID as well? To add to that, I'm curious, what were some of the lessons for you as you're navigating that period of time as well? I think the one thing that I think everybody would say was in the risk management space was that they did not anticipate the amount of regulatory influence that will come in. I think COVID was one where people expected to have a lot of losses and most businesses actually had fewer losses. In fact, in all of lending, we are still operating at delinquency rates that are pre code levels and the numbers are creeping up to kind of like pre code levels, but they're still way below pre code levels at this point in time.

The regulatory impact that reduced a lot of the losses during the time, I don't think anybody anticipated the extent of that impact. I think that has been a lesson learned. I don't think it will apply in the next position that actually happens. Highly unlikely that that is going to be a vector at this point in time, for sure. The next one we have here is while a Fintech team ramps up to having a corpus of digitized business data and credit outcomes associated with it that can be used to build ML models. What are some of the most effective heuristics you've seen for making credit decisions on top of this digitized data? I think this is similar to the previous question I was saying there isn't a single single answer that is in there. The challenge I think I spoke about briefly, especially in the BDB space, I think in the consumer space it's a different world because the volume of data that you have, you can actually easily apply some of these ML techniques.

It's a little bit harder in the BDB space, but there are pockets, for example, like transaction fraud, there is millions of transactions that actually goes through and that is an area where your data gets more mature very very quickly. And that's an area where machine learning techniques actually help you actually make these decisions. But the challenge with deploying that is your SLAs of making a transaction decision on approving or declining a transaction is very very small. So you need to build an ML platform that can actually execute on this model that you may actually be building. And that is the challenge. And I think as closer to real time high SLA decisions that you actually have to deploy these models in your data stack, your engineering stack, has to be incredibly strong for you to be able to support it. And so depending on what type of a business it is, your investment in that becomes key for sure.

Yeah, makes a ton of sense. And then this next one is something that we touched upon in our prep session actually. What do you think will happen to all the neo lenders and neo banks that have popped up in the last few years? So again, the favorite that I've thought about this problem is what kind of product suite have you been building and how much of what you have built is embedded in the life and functioning of the consumers or the customers that you're actually building it right? Products that have been developed in Silos, if you're building unitary products, I think those I would say would be at the highest risk of kind of slowly dying away when push comes to show and their survival becomes more difficult. Businesses that have been built on products that are more well thought out, that are more connected, that are embedded in the ecosystem of either the businesses that you're working with or the consumers that you're working with have the chance of being surviving for a longer period of time. And I think that if and when some of the clawbacks happen in some of the businesses go out, there is an opportunity for consolidation and those businesses I think will thrive and I think they will go for sure makes a ton of sense. And then what is Ramp's biggest current pain point around risk? I think you touched upon a little bit earlier with the charge card product and I'm curious what's keeping you up at night and what are the big things that you're thinking about? I think a couple of things. Like one is like hiding has been a bit of a challenge and I think for people that are sling, we have seven open roles and decreased risk operations.

I'd love to talk to people if they're interested in talking to about roles. Hiding is one. I think we need more people, we need a better understanding of what is happening and I think that is key. The second part of it, I would say is Ramp is still a very small company. And I think if one of the larger businesses that we are working with is drifting towards a bankruptcy and we are not smart enough to pick it up, we don't have the data to pick it up and that turns out to be a loss, that unitary loss will look bad on a books. And that is always the trick of small business lending. In the consumer space you have volume and you can do low dollar lending.

In the small business and commercial space you're making larger dollar bets and a few of them will go wrong. And so having a good understanding of the risk appetite, having a good understanding of what that can look like and being able to forecast it and maybe prevent it if you can, I think that is something that always keeps me yeah, for sure.

We had a couple different questions that came in and basically how do you think about the risk of losing access to some of these real time sources and relying on third party data and which data sources do you wish for more consistently digitized to help with credit decisions? I think there's like a couple of similar threads there where they're asking both how do you think about relying on third parties for this data and using access to it? And then two, are there any other data sources that you wish existed so we could tackle one first and then the other? Yeah, of course. I don't think there is a need for you to have access to all of the data all the time. Right. I think striving for that is an impossible goal. So the question is if you have multiple data sources, which are the ones can you check the box off that? You have enough visibility, you have enough pieces of the jigsaw puzzle so that if something is happening, you want to be aware you are able to pick up on it. I think that becomes key. I think the second part of it is I think you also need to think about it in terms of the level of exposure that you're actually taking.

In some places when the exposure is low, low visibility is okay. When the exposure is higher, you want to be careful about it. So you cut that data in a few different ways so you don't have to have it all the time and you don't have to have it for all of the businesses that are on your platform. In terms of what could be digitized more, I think the commerce one has been a big win. I think that actually helps. I would love to see more accounting data being digitized and more standardized. The standardization of accounting data is a big challenge and I think Rutter is playing a part there.

I think there are some other winners in that space who are playing a part there. And then on the banking data, the ability to create risk signals out of banking data is key. I think that part of it is a little bit of an underinvested part, I would say. And I think each bank, each fintech, each lender is doing that. Standardization there will also be interesting, but banking is one where the connectivity is still not as good as it can be. Would love to get the US to an open banking platform where you don't have to like you can directly work with banks and collect this data that will be more effective. But I think that looks like we are too far away from that right now.

Awesome. How would you go about predicting or forecasting revenues for a small business from the data that the Rutter API provides? So it's very Rutter specific.

Yeah, it's difficult to be prescriptive and how you would actually do this. Right. I think if you have a banking data source, if you have a router like data source, and if you have a little bit of accounting data source, you could start playing around with the ideas of understanding marketing spend. You could also use digital data sources where you get like Google and Facebook ad spends if you actually want to. Right. So you're starting to understand where is the business spending money and is that money converting into revenue and at what efficiencies? So that part of it is a key part of understanding revenue and predicting that. Right.

Then there is a part of it which is in the cases where there is an inventory build up, what does that look like? What is the liquidation rates of those inventories? So how quickly are they actually selling them? How does that actually get refilled? That is another part of the equation that you actually have to think about. And then the last part of it is like how does this company actually run? Is it like a venture capital backed fund company which eventually will raise another round depending on how the revenues are actually growing? Or is it purely a business that needs to sustain itself on becoming operationally positive? Because many early stage businesses are negative from an operational expenses standpoint. But I think there is a point where they actually turn around. But if you're not in that bracket, you have to get to that as quickly as possible and you need to use pieces of the data to actually predict if you're as you get there. So it's a few different moving pieces. It's not like a formula that I can actually give you. And I think depending on.

The type of the businesses also, that actually changes, for sure. And always happy to talk more about redder whoever asked that question. The next question we have here is who do you recommend for Open Banking data categorization? Thank you. You had mentioned Open banking and data categorization earlier. Yeah, I'll give a plug for Headon, the vendors that we have actually been working with, and I think they've been incredibly good at helping us try and create a PNL out of banking data, which is very interesting. So we're starting to play around with that. It also helps you with data consistency.

But the interesting thing about using these third party vendors would be you need to have your own bespoke logic as well, because sometimes the coverage will not be there. Sometimes you might have connectivity issues and things like that. So Heron is a great partner. I would definitely recommend exploring them, but also build a Bespoke and use both. I would say, to begin with, yes. Following up on that, somebody asked, how do you go about using a third party OCR tool versus building something in house where it seems to be a core asset of your data source? As a company alongside banking connections, I would say if you're thinking about it purely from an underwriting standpoint, if you're thinking about bank statements, this problem, I would say, is largely solved by third party vendors at this point in time. So I would say depending on the scale of the manual bank statement problem that you have, inscribe is a great vendor.

Oculus is another vendor. Actually, there are vendors who would run it through an OCR and then they have teams at the back end who will pass this data for you and you don't have to do it yourself. So there is manual human in the loop processes along with an OCR step of it, and you'll get the final output that you actually need. So unless you want to build out an engineering stack for this and you feel like this is an integral part of your product suite itself, OCR bank statements, I think, should be outsourced for sure. Sounds like you'll have to write a blog post in the underwriting tech stack with all the questions we're getting on each part of the stack. The next question is what are some rabbit holes that you've gone down in underwriting that turn out not to be a great use of time? That's an interesting one given your years in the industry.

So I'll talk about one again recently from ramps standpoint. I think experimentation is one thing that we have spent a lot of time actually thinking about. From a risk standpoint, I think we have bottomed out opportunities for us to experiment, but when we bubble it back up, I think it goes back to the point I was making on scale and how soon can we actually get the reads? And there are different areas where we have deployed different techniques in certain areas. We know that the amount of time it will take for us to take scale, we step away and we don't even try. But there are areas where we know that we can actually read these reads much more efficiently and much more faster. And that becomes a better area for you to experiment. Experimentation.

For the sake of experimentation, I would highly recommend not doing it. You need to have a very clear vision on what are you trying to learn. Is that learning important enough for you at this point in time? And design an experiment that actually gets you exactly that, because oftentimes the intent is right, but we don't design the experiment the right way. You get learnings, which just don't make sense. Awesome. Well, we're at time and this has been fantastic session. Sri has been super fun and it's been awesome working with the Ramp Team.

And for everyone in the audience, definitely check out the Ramp product. If you haven't, we're happy to use a Rutter as well. Great. If folks have all questions like yeah, guess, feel free to reach out to Sri. Thanks David. It was a joy talking to you. Hopefully people found this useful.

And again, plug back to you. Rutter is an awesome product. I think we are having a lot of fun using it and the insights that we're getting out of it. So highly recommend it as well. Thanks, David. All right, we'll see you guys later. Thanks for joining us.

Take care. Bye.