The SMB Context Margin Paradox

0:00 Hello and welcome back to the Fintech Takes podcast. today's episode is about one of my all-time favorite topics. One of those areas in financial services that is fun to talk about and fun to research because it's so so freaking hard to solve. Small business lending. if you've read the newsletter, if you've listened to the podcast, you know, I could not be more interested in understanding why small business lending is uniquely challenging. how it has changed over time as technology has been applied to it. We've seen more fintech competition and crucially for a conversation in 2026, how AI might finally unlock some of the challenges that we've been really stuck on in small business lending for decades. the perfect person to have this conversation with is David Snitkov, today's guest. He is the GM of SMB at Ocurus. someone I've known for a long time. fantastic basketball player, by the way, and a really, really great conversation partner and thinking partner on this very tricky, challenging, and fascinating topic of small business lending. David and I cover a lot of ground and as you will hear from our conversation, he brings a lot of firsthand experience to this topic. I learned a lot and was frantically taking notes throughout the whole episode which you may hear. so without further ado, here is my conversation with David Snickoff.

1:43 Okay. Well, making his Fintech Takes podcast debut. David, thanks so much for coming on. >> All right. Thanks, Alex. Great to be here. >> it is delightful to have you here. I feel like you and I have probably talked many times over the years about small business lending, one of my favorite topics. probably like in between games on the basketball court maybe. but this is the first time we're doing it. >> It's a great time to talk about credit.

2:10 >> Exactly. Exactly. Like you know >> basketball credit eventually you do it with a with a mic and a camera. >> Exactly. Exactly. So we've warmed up and we've we've done our obligatory stretches and now we get to get to the main event. David, I have a theory about small business lending that I want to bounce off you. and you you've worked in commercial lending for a while now. So, I'm hoping you can just help me workshop this theory. Does that sound like a good place to start?

2:37 >> Sounds great. >> Okay. So, the basic observation I've had and I I would say I've more worked around small business lending than worked directly in the trenches. But the observation I've had is that commercial lending is just deeply deeply contextual in a way that consumer lending isn't. you know, consumer lending, while we might like to pretend otherwise because we all like to feel like we're sort of, you know, totally unique individuals, from a data science perspective, that's not really true. Consumer lending is very homogeneous. Borrowers act roughly the same. credit risk decisions really come down to an assessment of reliability and capacity. you know, consumers go through this very sort of kind of predictable pattern through their life where they hit different life stages and they have, you know, predictably different needs. Commercial lending is very different and I think for two main reasons. First, companies are not homogeneous.

3:36 they're all extremely different from each other depending on their size, their location, what industry they work in. >> And the second reason is again different from consumers. Companies aren't static, right? Again, consumers move through life in this predictable, slowmoving sequence with a few big moments interspersed along the way. We really know what that looks like. The lifespan of a company, I think, is much less linear and predictable and certainly much faster moving, right? It's fast, it's chaotic, it's unpredictable. And as such, I think the only way to really safely underwrite a commercial loan is by understanding as much of the context surrounding the company as possible. Now the problem I think fundamentally is that understanding context is really really expensive right you have to have a lot of information about the business that you take in >> and you know quite frankly and this is what you can really speak to it requires a lot of like >> judgment or intelligence or experience like whatever we want to call sort of the art side if it's a mixture of art and science like that art side is very important to understand context which historically has meant that you have to have human loan officers and underwriters somewhere in the process and of course as we know human loan officers and underwriters are expensive.

4:56 Now the interesting thing if you look at the market and how it sort of segments itself, everything I just described is not a barrier to success in commercial lending for large companies, right? Because the size of those loans and the margin associated with those loans can absorb a lot of costs, right? Commercial lending in that respect is kind of like mortgage on the consumer side, right? Like there's more than enough margin to pay for a pretty meaty process with lots of humans involved and lots of steps along the way. obviously small and medium-sized businesses don't have that same margin to play with. And I think that's fundamentally why you see this sort of persistent finding that's been true at least as long as I've worked in financial services that small businesses continually report that they don't have access to as much credit as they would like. When they apply for credit, they don't get approved for as much as they would like. They don't get approved at rates or terms that they would prefer.

5:54 So there's this persistent sort of underserving of small and medium-sized businesses. And you know, every time there's a new wave of disruptors who come into the market, they sort of look at that and go, "Oh, well, you know, I'm sure that's just because banks don't know what they're doing and like we can figure this out." And then they kind of run into these sort of structural challenges that exist in commercial lending. This need for context and the fact that it doesn't scale easily down to the smallest level. and then they sort of have to figure out the same first principles problem and what their solution might be to it. So I guess two questions for you David. One, what do you think about my theory? And two, how have commercial lenders, at least over the course of your career, and you've you've worked in this space, for a long time, how have they tried to solve for this fundamental, let's call it context margin paradox that I described?

6:48 >> Yeah. So, first I really like the theory and the way that you've synthesized this. And to back up a little bit, >> lending is very much about decision-m under uncertainty and asymmetry. And the asymmetry, you know, you make a little bit of money on a good loan, you lose a lot of money on a bad loan. And then on the uncertainty side, it's impossible to know all the relevant information about the context as you said, the about the borrower, the environment in which they do business, the macro economy at the time that you make an underwriting decision or other lending decisions. And then you add on to that there's so many different types of businesses. And these businesses operate in different subindustries. They can be highly complex dynamic entities across all these different dimensions you've laid out. So when we talk about that asymmetry and uncertainty, the uncertainty aspect is really magnified in the commercial space. And for me, a few things come to mind there. There's always been in maybe the history of technology this tension between personalization and scale. So historically, if you want to do something that's quite bespoke, you can do that as you were saying for the larger commercial loans. And if you're doing, you know, multi-million dollar loans or credit facilities to larger businesses, then it makes sense to have a lot of bankers and loan officers work on that and for them to be, you know, relatively highly paid expert kind of people. And then if you're doing mass market $20,000 line credit cards, that's also quite easy because now you depersonalize. You you mass market the thing and you do it at scale. And now there's science to doing all that well, but you hope that it all plays out in the numbers, but you have to standardize. Small businesses get caught in the middle and the fundamental heterogeneity of small businesses makes that really difficult. And then the same thing, you know, lenders are always navigating. So there's that tension.

8:41 They're also navigating the friction >> between asking an applicant for more data, adding more steps to an underwriting process again to better understand the context and providing a good customer experience because any individual lender is not the only lender that a customer could talk to. So yes, if they could get all the data in the world, they would. But any customer who's good and has options will say, "No, I'm not going to give you a hundred months of my bank data or sit through an in-person interview. I'm going to go to another lender." And you might actually negatively select the worst customers who are more desperate for financing. So this context, bringing context from the larger commercial space into the small business space >> is very expensive and timeconuming. And it's also rare in the sense that you're not going to find the same industry specific expertise all in the same person. So the person who really knows how to underwrite, you know, a dentist is not the person who really knows how to underwrite an ice cream shop or a small manufacturer. And so having this many people is not really doable all but for the largest institutions. Now the question is are we entering an age where AI lets you overcome some of these context barriers that previously existed? And you know then to your second question what have I seen over the course of the last you know 15 20 years I think regardless of the the time that's passed in terms of innovation and technology the lending equation the the fundamental lending business model is unchanged. You know, every every lender likes to think they're different and they're different, but but fundamentally you make money on interest and fees, you lose money on credit losses and fraud losses, and you spend money on operational expense, cost of acquisition, and cost of capital.

10:34 >> And so, I think we've seen within that model three different categories of of the approach to the context that you laid out. And we may be in the early stages of a fourth. you know, one is just do what you would call traditional underwriting, but cost engineer it and cost engineer the process to scale it. So, you know, still have manual underwriting, still have very heavy documentation, but you know, cost engineer it either with technology or or outsourcing or whatever it is.

11:05 >> Sure. You've also seen some small business lenders adopt almost a consumer-like underwriting model where they're using, you know, largely consumer data, largely consumer data sources, largely consumerlike machine learning approaches to predictive analysis and then say, you know what, do I either, you know, have high enough rates to make the margin work or just somehow make it up in scale, which is very difficult. There's a third category of lenders and and these are the more recent ones let's say like most of fintech for the past 15 years >> who have said let's use automation and machine learning to cost engineer the process but in a in a good way that actually provides a good customer experience and then use sophisticated analysis cash flow underwriting targeted marketing dynamic customer management that's worked out well but I don't think it's the absolute frontier of personalized small business credit, personalized small business credit for each of these types of businesses. I think we're now just starting to enter this age where we have AI agents that can eat a lot of context and then you can do a degree of personalization and personalized analysis that was previously only possible if you had highly paid humans who had built up a lot of context and expertise.

12:28 Yeah, that that explanation makes a ton of sense and I'm glad you kind of walked through the history and the the way in which the market has sort of changed over time because you do see these waves of disruption, right? small business is such an obvious target. There's a ton of small businesses. It also and I think this bears, you know, stating plainly >> like small business is is a very like noble segment of the market to go after, right? like there there are certain parts of the market where it's like a very tempting target because not only can you go make money and build a great business, but you can also go home at the end of the day and go like I contributed to the you know foundation of economic success in America which is powered by small businesses. Like it's a it's a business that I think people want to try to make as successful as possible.

13:14 >> And you know you're right like there are different ways to sort of get around the scale problem. to use one example from like an earlier era of fintech. Like I remember when Square was first getting started with some of this stuff and like you know Square like they they started doing small business lending but it was almost more like micro merchant sle prop business lending like it was people with Square readers at a farmers market stand on the weekends. Right. And to your point >> there is a size of business where the loan amounts are very small. you can sort of contain it within an existing environment and Square was very smart about sort of putting lending in the context of the payment processing that it was already doing. So it sort of like scoped the problem to a particular size that it could sort of derisk. And then of course like you know for small businesses that are that small you know that's a business where it really is a reflection of the individual and the consumer. And so you can use consumer credit data. You can apply a lot of the things we do on the consumer side to the smallest small business owners and you know roughly speaking it works fairly well.

14:26 >> and we've seen some success in that area. Another one that you described is okay, you know, if we're a like a cabbage or an on deck or a blue vine or sort of that generation of sort of fintech lenders focused on SMBs, you know, it was cost engineering but with a sort of tech first mindset, right? Can we reduce the amount of paper documentation and get more of this data digitally? I remember talking to someone who was like, "You wouldn't believe how much money we saved when we transitioned from you know, going to do in-person site visits when we were loan loaning money to a laundromat or whatever to just having a human go on Google Street View and just make sure it looks like a laundromat and we know it's not fraud and we feel reasonably confident in that." And so there was, I think, a lot of like fat on the bone that you could trim away. Yeah.

15:17 >> But I think the flip side and something that you mentioned and I want to just sort of double click on and get your take on real quick is this all sort of runs up against the psychology of small business owners which I do think is important to also talk about. Right. Something I didn't something I didn't mention before but I think is very true is that >> another difference between consumer lending and small business lending is who you're lending to. And small business owners are are a quirky bunch, right? they don't really care about nor do they want to spend any more time than they have to on the back office side of their business, right?

15:56 Like they got into being a small business owner because they like construction or they like making pizza or they like you know installing septic systems or whatever the the thing is that they do. and all the back office stuff, including all the financial services components of it, is not really their specialty. They're not trained in it. They are not particularly sophisticated consumers of financial services, at least until you get up into the much larger size businesses where there's a little more specialization.

16:24 And you know the other component that they have that I make think makes them a little bit different than consumers is and I've used the analogy of like being a a new parent with a newborn at home like you're just not a very rational decision maker right like it's all about speed and stress and fundamentally like keeping this thing that you have this irrational fondness for alive at all costs right and so >> yeah and look you could even argue that it's rational but with regard to different dimensions like you may not Prioritizing APR, you're prioritizing speed, survival, convenience, personal comfort.

16:59 >> Exactly. Exactly. Like, like objectively, you might not be making the best financial decision, but Yeah. As you say, you're optimizing for a different set of outcomes, right? I remember when my wife and I brought home our first baby and like they just let you leave the hospital with like this is insane. I can't believe we're doing this. And that first night, I think we bought like $800 worth of stuff on Amazon that we realized we needed. And like I was not making rational purchase decisions at 3:00 in the morning, exhausted and scared out of my mind.

17:29 It's somewhat analogous to like what it's like being a small business owner. And you know, to to the point you made earlier, like they really prioritize speed. And one of the things that's so funny about you know small business owners is in the aftermath they will reflect back on like the decision they made to take this loan versus this loan or whatever. And they might be kind of bummed about the price they got or the terms or the service or other things like oh I wish I had like worked with a community bank or a credit union. Maybe I would have gotten better on those dimensions. But that's not what they were optimizing for in the moment they made the decision. They were optimizing for speed. So I want to David maybe deconstruct a little bit like what small business owners over the last let's say 15 to 20 years have sort of maybe had to give up or sacrifice because I think that what fintech companies have done over the last 20 years is really optimized for speed and convenience and this relates to digital and a lot of those sort of innovations that you were describing but there is a trade-off that small business owners have had to make in embracing that speed and convenience. Can you talk a little bit about that?

18:40 >> Yeah. And so there there's a whole lot there. I think one, as you said, there is somewhat of a a nobility in small business lending and that we all want to help small businesses and and I always say this to the team, you know, at at Oculus and other places too. It's like why are we here? Well, >> credit is kind of this magical technology. It's a technology that a capitalist society uses to express its confidence in the future. And it lets us make commitments today based on this optimistic projection of what's to come.

19:08 And consumers, businesses, even governments use credit to grow. And if we can do that in an accessible, transparent, fair, and and efficient way, we're going to have a higher level of economic growth in society. And we, you know, want to empower all these, as you said, very quirky, niche small business owners because that's what, you know, that's what makes a a community work. It's like, you know, you mentioned having kids. Having kids is one of these realizations. Owning a home is one of these realizations where it's like, I didn't realize that, you know, you need to have a bee guy and a well guy and a septic guy. Like there's a lot of guys in each one of these people as a business owner.

19:44 >> Most of those guys now, by the way, David, if if you ever need a guy, I now having done home ownership long enough, I I have a Wasp guy. I have a Well guy. I've got them all now. So, and and each of these people has like a very unique small business with all their tools and and it's it's funny like you you mentioned Square. If you're lend, you know, Square, we work with Square, of course, and they, if you are lending to a business who uses Square as their pay payment processor and you're lending in a in a quite constrained way, you can do that really well almost with a consumer-like underwriting model and then actually, you know, scale it in a different way when you want to go beyond into into different share of wallet types of customers and have larger loan products. And so, you know, I think what if small business owners had to had to adjust to or give up or compromise with, you know, I do think there's this view of, let's say, old style community banking and and to almost straw man it where it's like you watch some some old movies or old TV shows and people have this idea that there was like a community bank in your town and you went in there and they knew you and the banker knew you and it was very personal and they took care of your financial needs.

20:53 that maybe worked in a very ideal case. But if you, you know, weren't the right kind of business, you didn't know them, you weren't in the in crowd, and you weren't of the right, you know, group or country club or or whatever else that looked to that banker like the person who deserved credit or you didn't have history. You basically had no choices. So, I would very much of course take today's system over that one where borrowers have choice. they can shop around and they can say do I want to go to a a bank, a credit union, a large financial institution, a non-bank lender. Of course, many non-bank lenders now are either, you know, building banks or buying banks and integrating that into their strategy, including many of the companies that that you were mentioning there.

21:33 >> So, but I think that the speed is a really big thing and we actually do with our with our partner on deck, we do a survey of this every quarter where we survey a bunch of small business owners. Then we actually look at a lot of the quantitative data that's running through our systems and we ask them, you know, why, you know, did you go to a bank first for credit or did you go to a non-bank lender? Why did you do that?

21:56 What were your top priorities in applying for credit? What's your use of funds? Why are you applying? What did you try first before you applied for credit? You know, did you max out your credit cards? Did you delay payments to friends and family? And over time, we've seen, you know, the number of people going directly to non-bank lenders or you'd call innovative lenders, not traditional banks, going up. And many of the reasons that they state have to do with speed, convenience, and as you said, just a desire to get it done.

22:28 >> Yeah. >> You know, a small business owner is a busy person. They're probably, you know, a passionate person about whatever they do. And >> you know, you talked a little bit about documentation and and data and making data accessible. Like most people would not believe the volume of documents that are still involved in small business lending. And and just to put some numbers on this, >> we right now Acer we process, you know, we will process today one and a quarter million pages of PDFs.

22:59 >> Wow. >> And that's in addition to all the digital data we process and everything like that. But in small business lending, there's a lot more PDFs than most people outside like the guts of the industry would think. And a lot of that is due to the acquisition channel dynamics. So in business lending very often applications are coming through brokers. And so businesses will go to brokers, brokers will collect bank statements and then they have some routing logic and they'll send them those around to the lenders or funders who will, you know, usually wrap them through us. will do cash flow analysis and then they'll make an underwriting decision. It's also because while there's, you know, very good digital aggregators, you know, and we work with Plaid, we work with others in the space, the coverage of commercial bank accounts among many of these digital aggregators lags the coverage of consumer bank accounts.

23:52 And then the other psychology thing for business owners is that sometimes I think part of this is because some of the small business lending brands are not as well recognized household brands. They don't necessarily want to connect digitally >> until they know that they're getting approved. So, you know, we see this all the time. Someone will submit documents, get initially underwritten with documents, and then be totally good with connecting digitally right before they get funded. and then the lender can maintain an ongoing connection and manage their customer and everything like that. So it's a somewhat more nuanced process.

24:28 >> H that's fascinating. Yeah, I mean I I did want to double click on data because I think that that fundamentally that is kind of the foundation of the problem, right? And you know as you said coverage on the consumer credit bureau side is pretty robust, right? I mean, like I, you know, >> I probably complain and criticize the credit bureaus on the consumer side more than anybody. Like, it's kind of one of my one of my one of my hobbies. But, >> for all of that said, they work pretty well and they have really strong coverage and good data quality and, you know, like plenty of things to nitpick, but it works pretty well. And again, it's this homogeneous population that has great coverage.

25:09 >> You know, that just does not exist on the small business side, right? And you know, you can find bureaus that have some coverage of some data, but like the reality from a small business perspective, and this goes back to kind of the inherent differences between consumers and small businesses, is small businesses like it's it's hard to really like overstate this like consumers are born and then they hope hopefully live a nice long life and nothing really changes that much. Small businesses are born. Small businesses pivot 19 times. They change their name, they move into different industries, different locations, they die all the time. Like small business is just a really dynamic area compared to consumer. And so it's not even really a critique so much of the commercial credit bureaus because they do the best that they can, but it's just a really hard problem to wrangle from a data perspective. And then of course, as you mentioned, there's a ton of paper documentation, bank statements, PDFs, financial statements. There's just, you know, tax records. There's all kinds of stuff that's still transmitted via like, you know, PDFs or like faxes. Like we're still faxing things in some corners of commercial lending, which is wild.

26:24 >> More in mortgage than commercial lending, but it it happens. Goodness. It may happen. >> Yeah. I mean, it occasionally still happens. Like, it's it's it's it's very out of step with how kind of modern consumer lending that we might think of works. I want to talk a little bit about where things are starting to change on the data front. and obviously I I guess a couple things I would flag and then you correct me where I'm getting this wrong. as you said like systemtosystem integrations are now I think more common than they were. And so obviously the same sort of open banking consumer permission data revolution that happened on the consumer side has also hit commercial. you can connect your bank account, but there's also other sources of data too that I think is kind of interesting, right?

27:10 you know, as we talked about before with Square, Square knows a lot about you as a business because they process your payments. You might even manage like marketing or inventory or other things through Square's kind of commerce enablement platforms. And those platforms also exist at Shopify and any number of other places where you might operate your small business. A lot of small businesses actually use multiple platforms for sales and marketing and inventory management. So it might be spread out across a bunch of those. You can in some cases permission access to that data. same thing applies from like an accounting perspective, right? And what's interesting to me is that I think there's like a a trade-off it seems between the ease of accessing that data or is the consumer permissioning that data or the customer I should say permissioning that data or as you said the small business owner being willing to give access to that data at different steps in the process versus what I would think of as like the fidelity of that data. And and one of the things that someone told me a while ago that I I've always remembered is they're like look picking on accounting data for an example like accounting data is great.

28:20 It's relatively easy to connect to your accounting system and like pull that in but this is the way the small business keeps track of its own information. It's not necessarily a perfectly accurate reflection of the actual financial status of the business. And in fact, because small business owners are notoriously not accountants and not necessarily good at the back office bookkeeping parts of their business, you don't necessarily want to totally always believe the accounting data, even if they're very happy to permission it and give you access to it. So, can you talk us through like kind of the frontier of >> yeah, >> digital data within the small business landscape and kind of what that looks like?

29:01 Yeah. So, accounting data is a really interesting place to start because I think people do have this assumption that accounting data is a reliable source of data and if people are doing their books in QuickBooks or or Zero or whatever that you can permission that data and you know if it's if it's high quality, it is valuable. It is less likely to be high quality the smaller the business is because you probably started a business because you were you you were good at baking or you loved selling books in a bookstore or you you know were really good at at killing bees at your house or my house but not because you were so great at bookkeeping. And so some of that >> the only small businesses that are great at that are bookkeepers, right? Like if if the small businesses accounting shop then you're good. Then you're good.

29:42 Yeah. >> Fantastic. Give us your give us your accounting statements. you know for for larger businesses, larger line sizes, maybe a business has a greater degree of financial and accounting rigor and so you can count on that. And of course we see our customers who get in to larger loans and lines requesting financial statements, accounting data connections, all that kind of stuff and they'll do a broader underwrite. What we generally find it across small businesses is that is where cash flow underwriting really shines and that bank transaction data is just the best way to truly understand the nuanced operating dynamics and financial health of a small business. Now it's difficult of course because there's 8,000 something banks and credit unions in the US and so you get this bank data in many different formats and qualities. Sometimes it's through documents, sometimes it's through digital connections. The accuracy there is really, really important because the analytics really matter. Now, the interesting thing is you still need a ton of, you know, contextual knowledge to properly understand what this bank data represents. You know, does a bank transaction represent revenue, expense, proceeds from a lender, payments to a lender? You would define those things differently for different industries.

31:00 But I think you know the frontier now and it you know it's partly this is partly standard practice for the most innovative lenders today that we see but it's also part of the frontier for how people are pushing it. >> Can you get as much nuance as possible out of this banking data? understand the transactions, their categories, their counterparties, the patterns, the periodicity and use that to basically map >> to all the things that you would want to understand from traditional financial analysis. And and then where I think the context helps, imagine you were a manual underwriter looking at that, you would spend a long time then researching everything else about the business, everything you could find out about them, the economy in which they do business, how they compare with their peers. Now, AI agents make that faster or at least more more reliably scalable.

31:52 >> Yeah, that makes sense. I mean, the maybe double clicking on that, one of the things that I'm curious about is more on the analytics front, right? So, stepping one layer up from the data, I mean, you've talked about the fact that like specialization is really important to understanding context, right? And the way I would sort of describe community banking doing small business lending and kind of that like old-fashioned it's a wonderful life sort of vision of how it used to work is that you know George Bailey didn't necessarily have a great deal of expertise in any one specific industry or vertical. Like he didn't know that much about baking. He didn't know that much about car repair. like there's a whole bunch of things that like he only had a cursory knowledge of, but that was sort of overpowered or rendered somewhat unnecessary by more of like the character community side of it. So, it's like, you know, as a community bank, I might only ever underwrite like two dentists office for loans, but I know enough about those individuals. I know enough about how they interact with the community, how many dentist office this community can support. Like I sort of substitute my local knowledge in for my lack of knowledge about any one particular vertical and I roughly make up the difference on being able to make a goodish loan at relatively small sizes. as we move into the future where you know everything's kind of gotten flattened out and if I'm a small business owner as you said I have a million different places I can look and you know it might not be the community bank that I'm walking into especially if I have to take a day off work to go sit down and do an interview with a loan officer. It would seem to me like one of the things that's likely to happen is that small business owners will start to sort themselves a little more by industry or vertical and try to seek out lenders that have a little bit more of that specialization because and I think this is probably more and more true as you get like bigger like we have we've focused more on like the small small business side but like I think there's a huge gap in the market with like mediumsiz businesses that just don't fit neatly into any of these buckets. Like, we're fairly big. We need to make a big capital investment in >> whatever, new trucks or new ovens or hiring these people or doing whatever.

34:20 And in a lot of cases, like that is like it's almost more like VC investing where it's like we know that we can take our business to a whole new level if we do this, but we need capital to do that. Well, okay. When I go explain that to someone, they might need to have a pretty good knowledge of like my industry and what the trajectory of success looks like for a business in my industry versus another. So, I'm curious as it relates to like analytics and processing the data and even things like I know you're a student of like the growth of vertical SAS platforms and kind of more specialization.

34:56 >> How do you see that playing out? because it does feel like all things being equal, there is a benefit from a small business or medium-sized business's perspective in working with a lender who understands your business and your vertical and can underwrite to that knowledge more specifically. >> Yeah, totally. And and to to just illustrate that, I remember back in the the early growth days of peer-to-p peer lending, which then became marketplace lending. And you know, I' I'd of course in in my business back then and we would meet a ton of new lenders and sometimes when people said, "Hey, we're starting a new small business lender." And I would say, "What what kind of small business lender?" They'd say, "Well, well, all of them, you know, every every business."

35:35 I'd say, "I don't know." That that makes me a little bit worried. so so first I'll talk about the analytics and how you specialize that and maybe get into vertical SAS. You know, even when we're talking about cash flow data that we were just talking about, you know, fundamentally, what do you want to understand with the business? The the revenue, the expenses, the balances, counterparties, debt capacity, payroll, negative events like like overdrafts. A ton of context is required to evaluate that. And the context is very often industry dependent. So to give an example, imagine, you know, you're a small business, you have transactions in your bank data and you have a credit from an insurance company. For most businesses, that's not revenue. It's just a claim payment. However, if you're a dentist or an auto body shop, that's how you make your revenue. So now, if you're a company like us who does these analytics or if you're a lender using this, you need to treat all these industries a little bit differently. So even the analytical layer has to become specialized. Then the question is once you have those data points, how do you put it in context with a business's operating model? Also, of course, the operating model of a retailer or a manufacturer or a service business is very different. And I remember thinking about this a lot when I was at Cabbage where it's like if you have an ice cream shop in Connecticut that makes $50,000 a month in revenue. I I have unfortunately never run an ice cream shop. And I don't know whether that's good or bad. Maybe that's a lot of money. Maybe that's a little bit of money. Not all revenue is created equal and different types of businesses have different gross margins and cost structures. But, you know, we can say, "Hey, if we've seen 30,000 ice cream shops over the past few years, we can contextualize that number and draw a distribution and say, you know what, this one is in the 63rd percentile of monthly revenue and the 41st percentile of monthly payroll relative to its peers in the same industry and geography." And then you know then the next level is okay can you put that data in context with a lender's particular customer segment and product and portfolio. So I think the the credit and fraud risk of an individual small business doesn't exist in a vacuum like it's not an isolated property of that business. A lot of is dependent on the product the environment the lender how the lender runs its business. And so >> yeah, >> lenders really gain from fine-tuning a lot of these cash flow analytics and other data points to make them most effective for their own businesses. And we actually end up doing a lot of what you might call forward deployed data science in that area. you know, so vertical SAS you asked about and I think that's a really interesting interesting trend where it's like >> the shape of a business's context comes so much from the industry and and operating model in which a business operates. So it makes a lot of sense to develop vertical software that can form like the data accounting CRM workflow engine for that type of business. And then if you're doing that and you have all the data, you have the information running through you, you are theoretically in a very good position to provide financial products for that type of business. And they can be, you know, powered by this shared vertical specific context. And and I still think, you know, there's a lot of of juice behind this trend and I think it's a really interesting one, a really good one. I think we're still very early in that cycle. And so a bull case for this is well because of AI now number one it makes sense to develop vertical operating systems for even more niche industry categories >> where there may not have been an ROI before but now you know if it takes less to get started less to build that and bring it to market it makes more sense to develop these niche platforms and that really lends itself to offering highly tailored and useful financial products. The bare case, I think, is that maybe AI actually leads small businesses to want to change their vendor stack more frequently and then maybe they actually don't want to tie their financial services provider too closely to their operating software. I think it's it's not going to be one or the other. It's going to be a little bit of both, but we're quite early in this trip.

39:46 >> Yeah, it's a great point. I mean, I I think a lot about like the transition from like Square, which does lots of different verticals, including restaurants, >> toast, which just specializes in restaurants. And how much better can we serve restaurants than a Square or other providers that sort of work across multiple industries? And how much do restaurant owners respond to that like specialization of like, wow, this software feels like it was built just for me. But to your point of like going even further down, then you have someone like Slice that just specializes in pizza restaurants and they will go all the way to the point of like, you know, helping you figure out how to source boxes with your logo on it for like sending out the pizzas. Like it can get down to a pretty like niche level. And it is fascinating because I another part of the sort of bare bullc case for vertical SAS to me is how like scalable are these different businesses from a funding perspective, right? Like if VCs have to back these businesses, you know, there's only certain so so far down that kind of niche path that you're going to want to go before you're like this isn't really worth our time to invest in. But to your point, if we live in a world of AI and the cost of developing software continues to drop down and get lower and lower, well, you might see like a small business owner who spends 20 years doing pipe fitting or whatever it is that they do and just go, you know what, I'm kind of tired of this. I think I'm gonna use AI and my knowledge of being a pipe fitter to create the perfect niche vertical software platform for pipe fitters, you know, and like just amazing.

41:36 >> Yeah. And it's like and someday private equity will buy me and will roll me up into some large thing, but like I can make this is like the second half of my career is like building this thing with the help of AI even though I don't know how to code. And so I do think that there is a lot of interesting room to run there. And that of course brings us to sort of the AI conversation which we've kind of nibbled around the edges of but I want to I want to have in a more substantial way. So you've talked about the fact that obviously like I think you and I both agree small business lending commercial lending highly highly context dependent for all the reasons we've described. Even something like to use your ice cream example, seasonality matters a ton in some businesses, right? And so it's like, "Oh my god, this, you know, this this business only made, you know, $10,000 in revenue last month." Well, okay, it's an ice cream store in Maine.

42:28 So, the fact that they made $10,000 in January is actually really, really good and a sign that the community loves them and they're doing extraordinarily well. Like, you have to be able to put these things in context. as you said, AI loves to just eat context. Like that's what AI does. Like it it loves to get as much context. Anyone who's messed around with any AI tool for any amount of time knows. It just like asks you questions and it wants more information. Like it just wants to like learn more about the context around the task or the question that you're asking it. And I think maybe to add to that a little bit because we haven't talked a ton about this, but revisiting the data conversation, >> a lot of what we sort of have to make judgment calls around in commercial lending is difficult to structure, right? It's unstructured data to a to varying degrees, right? And you know, I mean, I know you guys have dealt with this at Oculus for a long time cuz even going back to like the sort of early like OCR type days, it's hard to fit things into a a strict machine learning model that breaks as soon as one thing is different by like 10 degrees, right?

43:35 >> You know, AI doesn't have that problem. Like it's delighted to evaluate loosely structured data, differently structured data, completely unstructured data. And so I'm curious maybe starting on that data level, how you see AI and LLMs and AENT AI playing a role and then I'd like to talk a little more about like how it may change the relationship between lenders and small business owners. But let's let's start on the on more of the data analytics level.

44:03 >> So and and some of this question is why it's such an exciting time in the analytics world for small business. And if you think about like you mentioned the the standardized data and the square rectangle if you will where it's like >> what types of data problems are responsive to the past few decades of traditional statistics and machine learning even at large scale still you know deterministic machine learning. >> You need large data sets of very standardized data with high reliability high fill rate consistent distributions good stability over time all that kind of stuff. If you have that, you can build like highly predictive models and a lot of personalization and do that for you know for credit risk for fraud for marketing for customer management cross-ell all the decisions that you make in a lending business you know and that there are companies like you know I I worked at American Express we did a lot of that there Capital One is famous for doing a lot of this and they've been really successful that data exists very well in the you know consumer credit CRA kind of environment in small business it's just more of a mess it's more of the real world. It's more fragmented.

45:08 And so some, you know, you the the type of data, the structure of data, the fragmentation of the data is not always responsive to the same traditional machine learning models. And so this is where AI gets really interesting because if you tune the models the right way, they are particularly good at understanding the nuance, having unstructured data, identifying the gaps, filling in the gaps with context that make sense. And I think like in in a way that the technologies of let's say 1015 years ago unlocked the power of big data. remember when everyone was talking about big data in a sense AI that has been pre-trained on massive data now unlocks the power of small data and we've actually seen a lot of success with this in our own operation even bringing it out of the credit decision for just a second like we use a variety of different AI models in in our work at Oculus and we will use models from the foundation labs from you know we'll we'll use open AI open AI and anthropic and Gemini. we will use smaller off-the-shelf models. We'll use models that we built in house, but we'll also sometimes use open- source open weight models and then prune all the other capabilities from them. So, they don't need to be good at writing poetry or producing a podcast or doing video editing or whatever. They just need to like read a payub really accurately and calculate, you know, the the Freddy Mack income calculation on a mortgage. And that's all they need to do. you take all their intelligence, fine-tune it on our data, make it work for that, then it's a lot more efficient and speedy and cost effective. And so, I think AI unlocks a lot of the power of of this small data.

46:55 And AI can really help resolve this conflict we were talking about earlier of the tension between personalization and and scale. And >> where this may go is like >> for a long time and this is still true to a to a certain extent where it's like >> yes a lot of the data and analytical techniques from the consumer credit world will go up market and have like more sophisticated statistical analysis and automation and good customer experience go up market and I think that's also true but at the same time a lot of the traditional underwriting techniques that we're talking about may actually come down market except instead of having you know a carbon-based underwriter you have a siliconbased underwriter >> which is in theory more you know observable and and reliable and and cost-effectively scalable and gets you out of that ultimate blackbox of the human mind.

47:56 >> Yeah. I mean that to me seems like the the ultimate sort of endgame for a lot of this. And to your point about specialization, like you know, we tend to think of AI monolithically, but I think that the value of it really is, you know, how can you as a business, as a lender, put yourself in a position where you can get enough experience to train like the best dentist office underwriter, right? and not even like underwriter for all dentist office lending needs for for you know capital improvement loans for commercial real estate like I mean you can get very specialized and you know because of as you you described very well the distinction between like pre-training on very large data sets so you have this like kind of generalizable flexible intelligence and then layering on top of that a lot of specialization one thing that I've observed that I think is really fascinating is you know these generalizable models they don't need that much reinforcement learning they don't need that many at bats before they get pretty good at doing this because they have such strong general intelligence capabilities underlying them >> the last thing I want to ask you about David is a little bit more science fiction because we're kind of skating into the future here a bit but give me one sort of thought or prediction maybe on how you might see AI agents and the sort of agentic capab capabilities that we're starting to get change the way that like small business lending works on a more fundamental basis because you mentioned something before about like small business owners like brokers are still a huge thing in in lots of different areas in lending certainly in small business where it's like hey I need a loan I have like small business owner I have a guy >> and I I call the guy and the guy routes me to a lender and helps me find something. brokers famously are a place where a lot of bad incentives can creep in. and it would seem like like an AI agent may be the broker of the future. Like give me give me a thought on how the relationship between small business lenders and small business owners may change thanks to Agent AI.

50:09 >> Yeah. And and the science fiction piece of this is fun. And even just talking about brokers for a second. There's such a range of brokers, you know, that there's there's very sophisticated customer focused brokers who are really matching people with direct financial product. There's also a very long tale of people out there who were who are kind of smaller shops and and they're doing all kinds of different things. I think that the acquisition channels for small business lenders will probably diversify over time. you it's just like you know in so many markets you you never get away from >> let's say brokers but they you do they become a smaller proportion of many lenders acquisition over time maybe that's true I do think AI agents have the ability as you said to change some of the relationship between business owners and their lenders because again if you think about what a business owner really cares about you know almost every single business needs capital to thrive you know to grow to manage their working capital to run their business. But every business is different. They have unique context. They operate in a different environment. The business owner has their own priorities. They have different financial background. They have different financial sophistication.

51:15 And then also, so there's this great mix and heterogeneity of like timing, suitability, what's the right product, what's the right size, who are your counterparties, what's the term, how flexible does it need to be? And today, we still with all the sophistication we have in what you would call like plain old-fashioned automation technology and customer experience, we still live in a pretty linear world. Someone realizes they need to apply for credit or maybe they get a marketing campaign, then they go, they decide to apply, they get approved or declined, then they, you know, resurfaced and they go forward, maybe they get renewed over time, maybe they expand. but it's still a pretty linear world. You could imagine a world in the future where if you have AI agents, you know, I think the business owners will have AI agents that are operating on their behalf to say if you were a larger, more sophisticated business and you had someone who worked for you, who had all this financial expertise, but also the context in your business, could they help you optimize your financial life as a small business and seek the right form of credit at the right time and the right terms and really really time and size that appropriately? And I think you'll also, and this is already starting to happen, you'll have AI agents working for the small business lenders at all these different phases of the customer life cycle, helping you figure out of everyone you could prospect, who should you prospect? Of all the customers in your funnel, how should you triage this funnel and and treat these customers in the right way and and personalize the approach? of everyone in your underwriting process, how can you underwrite them with the right context so you can truly make the right decision for that business? And by the way, do so in a compliant way. We haven't talked about compliance here, but again, when you're talking about personalization, now you get away from some standardization, which is usually the opposite of compliance. So, you need a whole different framework there. and you imagine also having AI agents for the lenders working on you know fraud risk, customer management, servicing, renewals, collections, all those kind of things. So I think you will have a a paniply of agents working with each other and again we are super super early in this cycle. and it'll be very interesting to see how this plays out over the next few years.

53:36 >> It really will. I'm deeply curious about it. I think you're right in the sense that we are just so so early on a lot of this and it seems obvious even being early that small business it's just going to be such a great place for these new technologies to play in because as we've been talking about it's complex it's context dependent like like this is the most persistently thorny problem in I think almost all of financial services and it has been over the course of my entire career. I know you feel the same.

54:08 So, I'm very excited to see what changes here. David, we'll have to have you come back on as AI and everything progresses, but for the moment, we'll leave it here. Thank you so much for coming on the podcast. >> Yeah, thank you, Alex.