Podcast: AI for credit portfolio management

Whitney McDonald 13:36:18
hello and welcome to The Buzz a bank automation news podcast. My name is Whitney McDonald, and I’m the editor of bank automation News. Today is December 3, 2024 Joining me is Scott Weller CTO and co founder of FinTech en fi. He’s here to discuss how enfi is using AI to eliminate the mundane tasks of credit portfolio management. Thanks for joining us, Scott certainly,

Scott Weller 13:36:38
and Thanks for Thanks for having me on the the show today. My name is Scott Weller. I’m the Chief Technology Officer and co founder of enfi, and we are building an AI assistant to assist with all of the things that are really mundane and challenging when it comes to managing complex credit portfolios. And you know, my background specifically has been as a technologist and a serial entrepreneur. I like to solve interesting and hard problems. The last company I founded, or co founded, its name was called Session M, and we were a data platform that was integrated with payment rails and point of sale systems to really make sense of that data for large enterprises that were running massive loyalty programs. And so we sold that company MasterCard in 2019 and built a division within MasterCard called merchant loyalty, acquired a couple other companies and really created something really useful and helpful for global merchants. And while I was at MasterCard, I got sort of exposed to the world of servicing, financial credit, financial institutions, and got to peer into some of the challenges that they have around people, paper and process, which, which, I think was one of the reasons why I got really intrigued when I discovered the problem and fi solving with my co founder, Joshua.

Whitney McDonald 13:38:07
Well, let’s take that a step further. Let’s talk through the problem that you’re solving here with en fi. It was founded in 2023 talk us through what you’re what you’re trying to address through envies, operations,

Scott Weller 13:38:17
absolutely. I’ll tell a little story about, you know, when I decided to leave MasterCard and take a break, I had been angel investing in, you know, helping other, you know, smaller startups in the in the emerging, technology community, and that’s where I met my co founder, Joshua, and we were helping a couple companies that had been really affected by the Silicon Valley Bank disaster, and they were really challenging getting commercial credit. The process was taking really long. It wasn’t clear why. They had to continually submit updates on, you know, information about the company. It was unclear whether or not they were really going to be qualified for the credit that they needed to grow the business. And Joshua and I got really weird data, you know, our hammer is just being data geeks and solving data problems. And so really started asking a question of, like, you know, what’s really going on with all this data behind the scenes? And so we were able to get connected with a bunch of folks who were subject matter experts in the credit space around just, just how much people paper and process there is to underwrite a complex credit deal. And we’re talking about inventory loans, mezzanine debt. We’re talking about venture loans. We’re talking about, you know, capital call line of credit. We’re talking about things that generally require a fairly sophisticated amount of underwriting, and then over time, it also requires a lot of data gathering, a lot of updates from the borrower, and a lot of you know, additional analysis to determine and rewrite the risk. So we immediately saw an awesome data problem solving we had built, Joshua had previously, previously built a data platform. I had previously built a data platform. It felt like this is a really great place to take all of our data related chops, our AI experience, our automation experience, and put it to work to make create a simple, useful and lovable experience for portfolio managers and underwriters. So we felt like really blessed that we undercut. We discovered this like opportunity space. And as we started working, we ended up talking to more than 50 banks and private credit institutions in the space. You just talk about the idea, the concept, show them a prototype, and got a lot of excitement around just creating and simplifying this process to unlock more revenue on the front end. And so that’s what we’re doing. And Fi is an AI based assistant that helps automate aspects of the information gathering and information processing workflows related to credit risk analysis. We reduce the total amount of time it takes to sort of assess an inbound loan, and we also, over time, really simplify the portfolio monitoring process, so within minutes, you can detect whether or not a covenant has been tripped, whether or not we’re trending towards a potential issue with a risk of repayment or variety of other analysis that we might be running on the on on the instrument. And that really helps us sort of make our entire operation more efficient, and then thus. Can accept more loans, more companies can get the lending that they need, and the the institution or the fund can grow. And so we see this as a really big unlock for something that’s traditionally a cost center.

Whitney McDonald 13:41:24
Now, a couple of things to unpack there, and I think that a lot of data reflects just this, that financial institutions right now, they’re, they’re ready and willing to invest in AI. But the question is, you know, where do I start? Or where should AI really fit into this piece of the puzzle? Maybe talk a little bit about how and fi gives an opportunity to, you know, implement some AI within your strategy, you know, while maintaining compliance, while seeing, you know, quantifiable revenue gains, maybe talk about what some of those conversations look like. We felt

Scott Weller 13:42:01
it was really important to focus at first. You know, I think our vision was really big in terms of covering each aspect of the lending supply chain related to complex credit and we, you know, I think to be successful, you have to find a part of that workflow where there’s a fairly big impact. You have access to enough data to train the AI for this, for that particular funder institution, and then you also have the ability to kind of unlock, you know, some efficiencies or some speed or some revenue associated with that use case. And so we chose what we found and what I would even if, even if an institution was like, looking for a place to go, do some, you know, testing, some, you know, so AI approaches, I would choose something that’s fairly repetitive, but then involves enough data to sort of determine accuracy of the implementation. All AI models generally have a certain amount of accuracy when you’re applying domain specific data to to that model. And so I choose something that was like fairly focused. We chose portfolio monitoring as a place to start because we felt that if we could map what is defined in a lending security agreement with with compliance packages like and do that without having to have individuals gather the data, spread the data, and apply the rules around the covenants that we would create something that that would really be an unlock in a process that we’re that activity is generally a cost center. It’s something that is performed for compliance reasons. It’s before, you know, perform for risk, risk rating reasons. And the focus around just the monitoring piece is big enough to sort of demonstrate and prove that you can have an impact on everything else. If you get that one thing right, there’s learning to then apply to deal screening and learning to apply to underwriting and learning to apply to future opportunities. So so we chose something I felt that was like, that we felt was really focused, but also could have a big impact, and that we could learn from and in addition, we also made sure that our early customers were also design partners. They’re willing to sort of lean in and learn with us. So we didn’t, so we could, you know, kind of learn where the AI had the biggest impact. We could learn where it introduced the biggest risks, and iterate, and iterate from there.

Whitney McDonald 13:44:37
I like that, having a focus, I think that we see across the board, a lot of, yeah, we’re gonna be investing in AI and, okay, where, what does that exactly look like? Having a specific focus, that you can, you know, invest in a specific area. See how it’s working. You know, measure those results. Tweak exactly just that area is really important. So maybe we can get a little bit into the how behind this. Talk us through the tech that drives en fi. How does it work? How does an institution start leveraging this tech? We

Scott Weller 13:45:12
wanted to create a simple, useful, lovable experience for something that’s fairly mundane and repetitive. And we, like I said, we started focusing first on the compliance cycle and focusing first on the risk re rating process of portfolio monitoring and management. And we decided that we wanted to also leverage all the documentation that’s currently available within the environment, and have a very easy onboarding experience. So these ended up being like really sophisticated engineering challenges. So we can, today, with our technology, we can read essentially use AI to sort of use natural language processing and modern AI models that we’ve trained to read a loan a security agreement. We can then extract all of the covenants that exist in that in that agreement between the bar and the lender, and extract them as obligations. We then convert those obligations into testable rule sets. Yes, we also learn from those obligations what metrics we’re tracking. You know, maybe we’re tracking a debt ratio, maybe we’re tracking a certain threshold, maybe we’re tracking a certain repayment cycle. We then can read a compliance package in a few minutes, and this might contain all sorts of information depending on the loan type, balance sheets, cash flows, inventory, inventory reports, depending on the type of loan, and we can apply that data to the rules. So within a few minutes of reading an LSA and reading a compliance package, we can actually let you know whether or not any of your covenants have been violated, and that’s like the first place we start. Generally, that process, depending on the loan type, could take hours or days, depending on like, the sophistication of gathering the data, re running certain downside analysis, having to go back and forth with the borrower on on negotiating what format their cap table should be in. There’s just so much, you know, I would say, busy work that happens that we that we convert into kind of an automated process. Now the analysts can actually think about what, what? What additional analysis should we be doing on top of this process to actually truly assess risk? And so your goal is to free up that time. But under the hood, we’re using a variety of different models to assess, assess the extraction of the financial information extraction of that LSA. We’re deep in natural language processing. We’re deep in using large language models and small models. We train our own embeddings models so that we can do similarity searching. We use knowledge graphs so that we can understand the ontology of this documentation and the relationships associated with different entities in the documentation. From a security perspective, we like tokenize all the private information and put it in encrypted an encrypted form, so that you’re not flowing different private and sensitive info through your AI pipelines and your automation pipelines and a variety of other other techniques. But really excited only takes minutes to get, like, instant compliance, which is like, one of the like, simple, useful level of things we wanted to

Whitney McDonald 13:48:32
achieve, yeah, speeding up any type of mundane process, right? That’s the key here, in a compliant way. I know that you mentioned making sure that the right data is going in and, you know, having that those safeguards in place, that’s also key right? Now, we talked specifics, we talked about the focus, we talked about the technology. I’d love to kind of take a step back here, talk a little bit pick bigger picture about AI and how it’s changing financial services, other automated processes that you might see freeing up additional human resources. Maybe just talk a little bit about AI in action, where you’re seeing those you know, tangible, quantifiable, qualitative results using AI.

Scott Weller 13:49:18
I think there are, there’s a lot of momentum around applying AI to the origination process for loans music, like a new company that pops up every single day, and there’s going to be some exciting, exciting companies created in that space. And I think, like as a practitioner, you know, I’ve often, especially working for MasterCard, working within financial services, for a period of time, I sort of made this decision that we want to be more of an intelligence platform than a decision engine. There are a lot of like, I would say, third rails around deciding, you know, making a decision on the behalf of a lender using automated technology today, like, there’s just, there’s quite a lot you could get wrong, and there’s regulations in place that could be violated if you’re not, if you’re not, sort of applying explainability to the process in terms of how you’re Applying. So where I’m seeing a lot of like opportunity for other companies and innovations in space is around this concept of being an agent or assistant that’s assisting with tasks. There’s a few companies have launched recently that help generate, you know, cre lending is, you know, I say the process for kind of coming to a decision, and building your memorandums and building your position on on the on the reasons why, or the risk levels associated with the CRE loan, requires a tremendous amount of documentation. So I’ve seen some cool use cases around just, you know, making that documentation process associated with a cre loan really fast and efficient. I’ve seen a lot of use cases around fraud and anti fraud, like being able to speed up the process of doing background checks, speed of the process of, like, of fact detection. Um. And then doing that over time, so that, you know, the institution doesn’t really have to think about re running those things. They’re sort of happening in the background, and alerts are sort of flowing when there’s observability on sort of a background check the fuel fact issue, I think one of the things that I think we think about quite a bit in terms of adopting AI within any organization is our ability to test its effectiveness. I think, as as financial services institutions like jump into the fray, of like using AI for certain processes, or even if they are applying it in the in the nature of being an intelligence platform or a decision engine, all these things can be testable. And so, like we’ve, we’ve put a lot of work into benchmarking the effectiveness and the accuracy of our AI models, which I think is, you know, you know, because we’re building a product, it’s like core to the product, you know, we really need to know any adjustment or change we make, we need to know its improvement one way or the other, on on accuracy. But I think it’s also hard for like, small teams within certain institutions to build that like that, like level of sophistication. And so I think if I was to start a project within within medium size like fund or a medium size financial institution around AI, you know, I’d start sort of focus on the question of, like, how are we going to measure our effectiveness with this project or with this effort? How are we going to benchmark the effectiveness of one AI model over another, because at the end of the day, like, that’s the level of sort of visibility you need in order to understand whether or not you’re you’re doing something interesting. So there are some interesting platforms emerging that help you do this, which I think is like, where seeing some really like, factual like value for those who are building versus buying.

Whitney McDonald 13:53:14
Yeah, and I think it kind of goes back to the top of the call about having a focus, having an idea of what are you really trying to automate right now, like having a clear set, you know, goal in place when it comes to implementing AI, not just, I’m sure that you’ve heard this over and over again, but not just implementing AI for the sake of it, but having those set milestones that you’re trying to accomplish and set areas of business that it makes sense for,

Scott Weller 13:53:40
I do have to say, and maybe people will disagree with me, but like the average, I would say, frontline credit portfolio manager probably wants to move up within the organization as soon as they realize the like level of like mundane tasks that they have to perform on a month over month basis for the portfolio, and they, and they, and I hope they see it as as a step to move up within the bank or the institution. So that means they’re just not in these jobs very long. And so I think with with what we’re doing, we can help make that experience a much more useful experience, and obviously lead to them doing more strategic things with the organization faster. I think that that’ll also be the same story. I hope for a lot of these other places where AI can be applied.

Whitney McDonald 13:54:30
Now I’m excited that you guys will be participating at Bank automation summit in March in Nashville. You’ll be part of our demo challenge. Maybe we can kind of lead leave off here. What are you most looking forward to at the event? Maybe give a little teaser on what you’re planning to demo. Of course, without giving away too much,

Scott Weller 13:54:50
sure we’re very excited to be participating. We, you know, we look forward to really connecting with the subject matter experts in this space, one on one, and learning about how they’re applying AI to their daily, daily jobs and tasks. And we’re also learning like, where are the friction points, like, where, where teams run into issues and challenges and in adoption, or maybe even in proving the value of AI so very excited to, kind of like, sit down one on one or in groups and really dive deeper from a demo perspective, and when I kind of project out to where we’ll be at that point in time, we’re very early Stage Company. We’re developing a lot of product fairly quickly. I think we’ll, you know, we’ve been building this multi agent system that’s highly trained on, you know, your documents associated with the credit space, and also trained on, I would say that not just the quantitative aspect of doing risk assessment, but also the qualitative so I think we’re going to show off a pretty interesting demo on how our system can solve fairly complex tasks in a few minutes, like I think, and we’re going to choose, like, some that are fairly esoteric. Um. And we’re going to have a portfolio manager actually demonstrate how it’s done from their perspective, to also show, like, how their job has changed by using using a tool. So it’s a little more than a teaser, but, like, we’re really excited to participate, and hopefully, you know, we can do it within, you know, the seven minutes or or however long we have to to sort of demonstrate the value

Whitney McDonald 13:56:41
you’ve been listening to the buzz a bank automation news podcast, please follow us on LinkedIn, and as a reminder, you can rate this podcast on your platform of choice. Thank you for your time, and be sure to visit us at Bank automation news.com. For more automation news, you.

Transcribed by https://otter.ai