Automating small business underwriting when identity isn’t straightforward
Why is underwriting small businesses so much harder than underwriting individuals? In this episode of The Lending Link, host Nathan George sits down with Michael LaSala from Baselayer to unpack one of the most persistent challenges in small business lending: verifying identity when businesses do not fit clean, predictable profiles.
They discuss why matching business names, owners, and entities is much more complicated than verifying consumers. Fragmented data, thin files, and unstructured information make things harder for lenders, leading to more risk and manual reviews. Michael explains how lenders can use AI to organize messy data, identify useful behavioral signals, and better distinguish real businesses from fraud.
They also talk about how automation speeds up decisions, helps more applicants finish the process, and lets underwriting teams focus on more important tasks. The conversation looks ahead to new trends like Know Your Agent and more applications, and to what lenders should do as AI continues to change small-business onboarding and credit decisions.
Episode Transcript
Nathan George: Well, welcome everyone. We're going to wait one more minute to get started. Just getting all the people let into the meeting. Well, welcome everyone. As people are joining, I wanted to thank you all for being willing to spend some of your day with us. I'm Nathan George with GDS Link, and today we're going to be diving into one of the trickier areas of lending, and that's the small business arena. Lending to small businesses, particularly identity and verification. Many of you've probably had the experience, or maybe you haven't, of researching a company, hopping on a call, and then realizing that all the research you did was for another company of the same name. So aside from a bit of embarrassment on a Zoom call when you're lending, that actually can be a much more expensive mistake to make. So anyway, I want to welcome today a good partner and friend, Michael LaSala from Baselayer. Mike's been in the business for a while. He's on the growth and product team at Baselayer. So, Mike, thanks for joining us. Do you mind taking a couple minutes just to introduce yourself?
Michael LaSala: Yeah. Awesome to be here. Thank you for having me, Nathan. Michael LaSala. I'm on the founding growth and product team at Baselayer, as you said. Been in fintech about 10 years now. Started my career at Bank of America and then spent a number of years at Plaid, which is actually how I ultimately got introduced to the team at Baselayer. Baselayer is a business identity, credit and fraud platform. We work with about 2,200 financial institutions. So about 20% of the FIs in the United States are using us already. And they use us to onboard and monitor the small businesses that they work with, which is quite the challenge. So I'm looking forward to the discussion. Hopefully it'll be very useful for anyone listening in.
Nathan George: Well, thanks again. And actually, Mike, one second. Before we get started, I did want to make a note. This is a live session. It's meant to be interactive. So if anyone has any questions, please pop them into the chat. We'll address them as we go along here. And again, appreciate everyone's participation. Mike, one of the first things I wanted to ask you is why is it so difficult to match business names and to get good information to make in particular a fraud and credit decision when you're lending to small businesses? What is it that makes this such a challenge?
Michael LaSala: Yeah, it's a great question. And the way that I like to think about it, what makes identifying a business so difficult is actually to start with verifying individuals. Why is it easier to verify a human than it is a business entity? The way I think about it is that all humans are similar in that they're humans. They are probably 18 years old if they're showing up at a financial institution. 18 at least. They have a Social Security number, they've established a digital footprint, they have a document, a driver's license or a passport that they can verify. And then we all have a face, hopefully that matches the document. So there's something similar pattern that you can draw when you're verifying humans. And then what's more, on the consumer side, you have a lot of event-driven insights. You kind of know how your average consumer behaves in terms of their presence online, their application velocity. You compare that to business entities. Anyone can start a business at any time for any reason. So if you're a financial institution and a business shows up at your door, they could be a Delaware C corp with a lot of foreign registrations, a lot of officers, they could have a very established online presence, or they could be a sole prop who just has a page on Facebook Marketplace, hasn't even been registered with the state. How do you verify the legitimacy of those businesses in a programmatic way? That's a real challenge because you've got 51 different Secretaries of State, different officers, different filings. And ultimately, if you don't verify the identity of the business to begin with, then all the other data that you pull on that business is just noise. The liens, the lawsuits, sort of the downstream, deeper underwriting information that you would want. If you don't know, okay, we're talking to the right business and the person who we're talking to has decision-making power for this business. Well, you're kind of dead on arrival. That's what we've really focused all of our time. And we have some of the best engineers in America working on this. For us, it really comes down to having all of the possible data you could want and then delivering it to our customers in a really structured way that they can use deterministically. Because ultimately that's what you need to automate. And these days there's a lot of pressure on a lot of people to automate something.
Nathan George: So that's helpful. It's a bit ironic just thinking about it from the driver's license perspective that the DMV is more efficient than the business side of things. But you know, that, that being said, when it comes to making a determination about whether or not to extend credit to a business like that, for the average lender today that's serving that market. What are some of the more common core pieces of information? Maybe could, maybe we start there, we can start talking through why it's difficult to get access to that or find the right bits. What, what is it that. What is the information that is most commonly used that you guys see that helps make that decision or make a good decision about lending?
Michael LaSala: Yeah, definitely. We take a very jobs-to-be-done approach and we try to think about what are the highest leverage bits of information that we can provide our customers about a business for the cheapest price at the right moment. And so when you think about credit, as I said, if you can't confirm that you're dealing with the right business that you should be, it doesn't matter what sort of liens have been filed on that business, what their lawsuit history
Nathan George: is,
Michael LaSala: because you're fundamentally not dealing with the business that you're trying to underwrite. So you got to get the right.
Nathan George: John Smith. Right?
Michael LaSala: Exactly. Yeah. Or John Smith LLC in our case. I mean, yeah, it really does start there. But then once we have and we've confirmed with our customers that we've matched the right business, well, then from an underwriter's perspective, we think about what is everything outside of the financial statements or use an aggregator to connect those bank accounts, what's everything that a lender would really care about? And from that perspective, liens especially, you know, how, how many other lenders have, have their claws in this business? You know, if there was a bankruptcy event, have they ever been bankrupt? What other sort of lawsuits do they have against them? There's real, you know, obviously important underwriting decisions that can be made from that alone. But then going back to sort of the behavioral insights that are commonplace on the consumer side, if a business lands on your doorstep, do you know anything about how they've been behaving in the financial ecosystem for that? Have they ever been reported for fraud by, say, your competitors or in some cases a partner? Or for a large enterprise, it might even be within the same company. Have they been reported for fraud? And how many times are they applying? Are they applying in a cadence that would be typical for a normal business? Have there been variations in the applications which would be indicative of fraud? So you can actually get a lot of insight from that behavioral event-driven data on top of the liens, the lawsuits, making sure you have the right industry. And then we actually go a step further where, okay, you know, this, this business applied and they said they were, say an e-commerce website, but then you go online and you see CBD or maybe something even more, little less than that. That's something that historically you just need your average risk analyst to go Google and hopefully find. But if you can share that messy, unstructured data in obviously a structured way and then also give customers the insight on what they should do with the data, that's really valuable. That's what actually leads to being able to automate. And I think we were talking about the other day, Nathan, part of the work that we're automating is, I mean, tedious, painful work. It's going to the Secretary of State and looking up a business name, maybe you got the right address. Searching through UCCs, reading reports very frequently. It is tasks that software is designed, at least in theory, to automate. But because of how messy and unstructured that data is, it really is a challenge whether you're doing it manually or with some of the other providers that are out there.
Nathan George: I want to unpack that a little bit more before we get there though. There's one other problem I don't think I heard in your description of the issues they run into or the data that they're looking for. And this just comes from experience with some of our lenders that we have that are serving small business space and they're the type of client who really need the loan, maybe to fund some inventory or something like that. They might be a new business that's a year old or maybe less than a year old. They don't, there is no, I mean, talking about thin file or no hit, like with the consumer space. There's nothing like you don't, you don't get a hit in D&B, you don't get a hit anywhere else. Is that, that seems to be another major area that makes this such a challenge is the dearth of information on brand new businesses who are the most likely to need or, or be a potential client for a loan.
Michael LaSala: Yeah, yeah, absolutely. And maybe the highest risk of fraud too in some sophisticated fraudster scenarios. And I think if, if you're a lender and you know, very commonly they'll, they'll have limits or minimums on say, the amount of time that you've been in business. But if you're at that cusp, right, like what are all the pieces of information that you would want? Usually you're going to underwrite the owner of the business, the individual himself. And so there's a bevy of tools out there and there's credit bureau information. All that is of course important. But I think too, to your point, if it's a new business, how can you know if they are legitimate not just in the eyes of the government, but again in the way that they have been operating? Do they have a ton of liens? If they're a young business, they probably shouldn't have. They been applying for a ton of loans in that case, the best case scenario is that they may be desperate for cash and so that could present an underwriting risk. And what's more is if you're a lender, you know, if you decide to work with a business, especially a young business, very frequently, you're going to have some sort of covenant against credit stacking and you might have the bank account connected through say a Plaid, but you might not have all the bank accounts connected or they might have found some way to skirt the system. So if you can know, hey, this business who you have a loan out for has applied to five other small business lenders in the past month, that might be very valuable to them just to know that. So I think that again is why we have so prioritized the network architecture and that event-driven insight.
Nathan George: Yeah, well, let's pull apart the challenge of that. You know, disparate data, different formats, very unstructured, things like that. It sounds like what you're saying, correct me if I'm wrong, is that the way most lenders are dealing with this today is manual review, which creates a significant bottleneck. Right. For every X number of applications you want to take every month, you've got to have one full-time employee to manually review. So it's a significant bottleneck. How, how are clients, how are you guys helping clients today get around that, that limitation? Because I mean I've looked at Secretary of State filing websites and some of the sites where you can go do lien searches and things like that. It is really complicated for a human to read those. I mean I'm not, I don't do it every day, but so it's harder for me. But that's a really big challenge. How are you guys tackling the challenge today?
Michael LaSala: It's a very painful challenge. I know because I've done it thousands of times at this point and it's mind-numbing work for us. We like to make anyone who we're going to work with feel very comfortable about what they've been doing historically and how it would work, say if they were to work with Baselayer, that we are going to be extremely precise in many cases even more so because it's such mind-numbing monotonous work than the manual review. And obviously it's much faster if you're getting this information by API. So we like to do, you know, we, we refer to it as a confusion matrix. What are the true positives and the true negatives of the data that is presented to us? Because the customer knows what the outcome should be. And then when we present to them, hey, the outcome that you landed on, say with manual reviews, is at least identical to what, maybe even improved to what our output offers. Well, now you know that you can get that same job done in an automated fashion. You don't need to spend 5, 10, 20 minutes on each filing just to know that they're legitimate. Now what's more, if you're doing this manually, it's very easy to miss something because it's so many different data sets. There's so much context included in that final decision. So you really need the power of, of AI, of software to pull all of that together and again, just present it in a really easy-to-use way. But it needs to be precise, it needs to be accurate. Otherwise you get the output and then you still need someone to verify the output and then you haven't really automated anything.
Nathan George: So it sounds like there's opportunities beyond. And by the way, interrupting myself, I want to hear more about how that, you know, what is it the AI is looking for? Like what is. If you give us some examples, I'd like to do that. But it sounds like if you could free up some of those resources from the tedious manual review work, there's cross-sell and upsell opportunities or maybe optimizing the offer to be more fit for the lender. There's ways to make more money, more margin, sell more things or help the client that you're serving in a better way because you've got that resource looking for those opportunities rather than just looking for identity match and fraud alone.
Michael LaSala: Oh yeah, absolutely. If you can come to a decision faster, whether it's yes or no, that is better for the small business owner because they just want to get the job done of opening an account, setting up payments, getting a loan. If they're left in this waiting game where they're just not sure, that is a ton of pain. Especially in the case of business lending. It could be their business on the line where they need to know what the outcome is.
Nathan George: Or they could go to the next website down and fill out the application there and they're all automated and then you're out of luck, right?
Michael LaSala: Absolutely. There's big conversion risk that's right,
Nathan George: so let's unpack a little bit more. And I know might be skipping ahead on some of our topics we wanted to cover today, but the use of AI to understand or glean more data out of this unstructured mess from all these different government sites and things like that that we're doing. Can you maybe give us some examples of what you pull or what you were able to extract and how you put that together to a more digestible format?
Michael LaSala: Yeah, yeah, absolutely. Our view of what AI is really good for and where business onboarding and monitoring is headed, and you kind of alluded to it, it's taking a ton of unstructured data and then making it structured. That's sort of step one is just making it. You've got so many different officers, so many different states, you've got the website that they applied with, but there might be another website online that AI can go and find.
Nathan George: Like their store might be on a different site, the retail store, if they're doing e-commerce, might be on a different site.
Michael LaSala: I've seen it many. I think Microsoft Support was sort of the infamous story from a few years ago. Like microsoftsupport.com, people thought they were dealing with Microsoft. That's obviously a huge example. But we've seen very sophisticated fraudsters who will make up a fake website that looks exactly like the one that is the legitimate website. But they change like one word. You know, it's like performance.com or performancelab.com.
Nathan George: Yep. It's an old school method for check fraud. Right. They would change one letter, start their own company with a name that looked exactly like the names that. Yeah, that was an old check.
Michael LaSala: Yeah, exactly right. Catch Me If You Can. One of my favorite movies. Great Christmas movie. Yeah. For anyone out there looking for one. But. But that's exactly right. And that's an example where if you're doing manual review, you know, and I can show you these offline, it would stump me. So it stands to reason, hopefully it would stump the other people. Maybe I'm about the median. But so finding all of the websites, the social media, the review pages online, AI is really good at that. We have a great team who's been focusing on that for the better part of three years. And then reading all of that information, synthesizing it, and then matching it up with everything else that we've found, whether it's from the Secretary of State, the UCCs, the lawsuits, all of that unstructured data, making it structured, matching it up with, with one another. So say the address that's on a Secretary of State filing, matching that with the address that's on the website, same thing for officers and executives. And then I think the real North Star with AI is to not just structure the unstructured, but then to tell you what to do with it. All of this information matches up, therefore you should approve them and move to the next phase of underwriting or onboarding essentially. And that is very much why we've built both a score to tell folks, hey, this is an A, good to go, move it on to the next phase while also providing them the structured data underneath it.
Nathan George: So I'm curious how and maybe AI is not the only way you can address this problem. But with business lending and FCRA rules, you've got this weird juxtaposition of sole proprietorships, right? Like you've got some FCRA and other regulatory commitments you have when you're underwriting a loan based off of the owner of the business and their tax ID. But at the same time, the purpose of the loan is for a business, which you've got to somehow validate or verify, right? So you got to be careful what you're using to make the credit decision. The fraud decision is a little easier. Like how, how do you help lenders when they're looking at sole proprietorships in particular, how do you help them distinguish those categories, those areas of risk, and make better decisions based off what your AI and other tool can reveal?
Michael LaSala: Yeah, yeah, absolutely. Putting the, the sole person behind the proprietorship put aside, you know that that will be the bulk of the underwriting that you're doing because there's no difference between the business entity and, and the individual. You know, you could just set up Nathan's French Bread online, start selling your French bread or your gingerbread cookies, which I'm told are fabulous, but you're not actually a business in the eyes of the government. So, you know, maybe I'm a lender, maybe I'm just a bulk. You know, I sell flour and you're trying to buy $10,000 worth of flour for me. This is a real use case. How would I know that you're legitimate, that you know, you're actually have started Nathan's French Bread and that you're doing some business for yourself.
Nathan George: And this is where it's not, it's a, it's a fantastic vector if you're for first-party fraud, right? Like, you know, your loan sizes are going to be bigger, your payout's going to be quick. I mean, it's a right target, it seems.
Michael LaSala: Buy a bunch of flour and then never pay up. Absolutely. And so where we can help with that, if you'll forgive the phrase, is almost the vibe check of the sole proprietorship. And what I mean by that is basically what your, again, average risk analyst would do on, you know, Nathan's Bread Shop. They would go online and they would try to see, does Nathan's Bread Shop have an address on, on Google? Like, are there any reviews? Do they have, do they have a website? Can you pay for things on that website? What are people saying about it? If they have social media, do they have a high number of followers? What are they talking about on social media about Nathan's Bread Shop?
Nathan George: Three five-star reviews.
Michael LaSala: Yeah. Or, you know, when, when was the last review? Right. Because it looks like Nathan's Bread Shop, two years ago was getting all sorts of reviews, but no one's, no one's reviewed since what happened? All of those are sort of those fuzzy signals that they take time to find. If you're just doing that search yourself and it raises questions that you just, you can't really answer. You need to go deeper and sometimes that might mean a step-up check with the sole prop. But if you can create that flag and you can say either, hey, Nathan's Bread has 5,000 followers, everyone loves it, you can check out on the website, nothing to worry about here, that's extremely valuable. And then similarly, if you can say, hey, this doesn't seem to match up with a business that's buying $10,000 worth of flour would need. You would think if they're going to buy that much merchandise, there would be something that says, yes, Nathan's Bread is a great shop and he's going to use that flour to sell bread and you'll get paid.
Nathan George: So your tools are able to extract and quantify some of those signals so that you can start to make decisions and then monitor the performance of those decisions and over time continue to improve your discrimination model that you're using. Yes, that makes sense. And I know we were trying to keep this very short today, although there's a lot more questions I've got on some of those data sources you used. But I, I did want to get to the kind of some couple of the big questions you almost have to cover these days. You know, we kind of talked about the AI win already, but would love to hear from, from your perspective, from Baselayer's perspective, like, what are the top two or three trends in business lending in particular that you guys see coming in 2026, which is only, you know, 20 days away now, where did 2025 go? But give us some predictions. We can hold you to next year.
Michael LaSala: Absolutely. Next year when we do it in person, we'll see how I did. I think that I gotta make some call.
Nathan George: Kalshi bets.
Michael LaSala: Yeah, I'm cheating with my answer here because I was at Money20/20 and so everyone was talking about KYA or know your agent. And if it's not happening at scale already, I think in 2026 a lot of FIs are going to be thinking, what do we do when an agent shows up at our application page to apply on behalf of a business? And what's more, what do we do if it's a good application? How do we really take advantage of the tailwinds from AI, whether it's through marketing efforts or how you kind of work with the different providers out there. What do we do about it? And I think that's a big question that a lot of people will have a lot of different answers to. But what is for sure is that if you don't have really excellent data, whether it is from the government, from the behavior of the business and their, their history, their, their prior loans, their, their lawsuit history, then things can get really messy really quickly. So I think there's a massive opportunity in that. But with that there's a ton of risks and that's what we're very excited about because we, we think we're in a really good spot to, to help people assuage that concern.
Nathan George: Well, if anything, it definitely opens the door to a volume increase. It seems in inbound applications like you, much more automation, a lot more which, you know, I don't know what you guys have seen, but small business lending is, has definitely been, in a percentage basis, the largest growth percentage for us this year in terms of application volume has come from the small business space which was, was pretty encouraging. But we're down to our last minute so I had a couple more questions but we'll have to hold those. If there's any remaining audience questions, please drop them in the chat. We will do the 2x or 3x answer version in the last minute. We have. I do want to thank you, Mike. It's been great having you on and chat with you for a few minutes and looking forward to what's coming next year and working with you guys to make that information available to our clients for sure. So thanks everyone who joined us. If you did put a question in the chat, we'll answer that offline and email it back to you. The recording will be available online. You'll be getting an email about that. And again, thank you for spending time with us today. And I want to wish everybody a very merry Christmas.
Michael LaSala: Merry Christmas, everybody. Thank you, Nathan. Thanks everyone at GDS. Always a pleasure. We'll talk again soon.
Nathan George: Bye, all.
About Baselayer
Baselayer is a fintech company transforming how financial institutions and lenders understand and manage business identity, risk, and fraud. Its AI-powered platform unifies data from public records, the web, proprietary sources, and a shared intelligence network so teams can verify business legitimacy, detect fraud, and make confident decisions in real time. Baselayer’s tools go beyond traditional credit or Know Your Business (KYB) checks by tapping cross-institution patterns and behavioral signals to spot risky applications and identity inconsistencies early in the onboarding process — helping banks, fintechs, and government agencies accelerate approvals, reduce manual reviews, and protect portfolios from emerging threats. Trusted by over 2,200 institutions, Baselayer’s suite includes identity verification, fraud prevention, risk scoring, lien filing, portfolio monitoring, and AI-driven risk signals that keep decisioning both fast and secure. Learn more here.
About GDS Link
GDS Link makes modern lending simple. Our real-time decisioning platform integrates over 200 data sources with advanced analytics, enabling lenders to make fast, data-driven credit decisions while mitigating risk. From loan origination to collections, GDS Link provides seamless automation, policy monitoring, and AI-powered insights to help financial institutions optimize lending strategies, stay agile in a competitive market, and drive better outcomes. GDS Link was recognized as the Leading SaaS Decisioning Platform 2025 and Real-Time Policy Monitoring Innovators of the Year 2025. Follow GDS Link on LinkedIn here.