Grow the Pie: AI, Lending, and the Credit Gap

Adam Blue talks with Bill Gravette, GM of Symphonix at Q2, about what happens to the loan process when AI enters the picture. From document processing and credit memos to alternative data and fair lending, they dig into where AI is already earning its keep in credit decisioning and why the loan officer, far from being replaced, may be one of the biggest beneficiaries.

Watch

Subscribe

Related Links

"Start at the Fuzzy Edges: A Practical AI Strategy for Lenders," Giridhar Thoopurani and Bill Gravette

"The AI Adoption Journey: A Survey of Credit Union Leaders," Filene

AI for Everyone, Q2

Symphonix

"The Minority Report," Philip K. Dick

"Minority Report," Steven Spielberg

Transcript

Adam Blue

Hey everybody, wherever you are, and welcome to Cut to Context. I'm Adam Blue and I'm joined today by Bill Gravette, GM of Symphonix, a Q2 organization that works on lending in the AltFi and banking space. Thanks for joining today, Bill. Really great to have you here.

Bill Gravette

Thanks Adam. Appreciate the opportunity and love the podcast. Thanks for all you're doing.

Adam Blue

Thanks. I want to talk about lending today. Specifically, I want to talk about the decision of whether or not a loan gets made. If you've ever applied for a loan, I would assume you have, Bill, because I don't remember you being a descendant of British royalty or anything.

Bill Gravette

I have. Yes. Car loans, home loans, all kinds of loans.

Adam Blue

It's nerve-wracking, that time between you do all this work and it can feel a little naked to go down and ask somebody if you're worthy of, I don't know, $40,000 to buy a car or a few hundred thousand dollars these days to buy a house. And I think a lot of us believe that AI probably is going to play a role in credit decisioning. I think that's interesting in some ways and I think it's terrifying in some ways.

But most notably, there was a recent survey, and we'll put a link in so you guys can go find it for yourselves, the viewers. And it said that two-thirds of credit unions are planning on using AI for credit decisioning. And so, I have very mixed feelings about that. And so, why don't you take me through, since you're a lot closer to this part of the space, for sure, Bill, why don't you take me through, what do you think that means and how do you feel about it?

Bill Gravette

Yeah. I think, it's definitely a conversation topic. My initial reaction hearing that stat from you is two-thirds actually feels low. But I'll acknowledge that I've got a focus segment of the market. Most people I talk to day in and day out are folks that are focused on lending as a priority in their business, and that's why they're engaging with us. If I think about that denominator of all credit unions, I think that, probably the two-thirds maybe feels a little bit more right. Credit unions, like all financial institutions, they're constantly balancing lots of different competing priorities. They've got deposit pressure. There's M&A activity that's a constant presence in our market. Everybody's thinking about their portfolio strategy and how lending plays into that. You've got some uncertainty in the economic environment and then an ever-evolving regulatory landscape that we're all accountable to keep up with.

And so, it seems reasonable probably that there's a third that are working on other priorities, and maybe aren't as focused. I don't know what the timeframe for investment was, whether that's '26, '27 on AI, specifically in the credit decisioning process. But I think, where we spend our days and probably where this conversation takes us is, regardless of whether it's two-thirds or 80% of that market, for the folks that are thinking about it, which, I would include us in that bucket for sure, the big question is, what's everybody trying to do with it and how do we think it weaves its way into the credit decisioning process?

Adam Blue

Take us through a little bit. What does that look like? Which part does the AI really participate in? How does it fit into the flow? And how can we imagine that that might progress in the near term and then the medium to long term?

Bill Gravette

Yeah. AI is a big topic. It's a very broad topic, depending on how you use it in context. For us with Symphonix, as you alluded to, we provide an end-to-end lending platform that includes loan origination, it also includes loan servicing. And when we lay out our marketecture and have our product conversations, we talk more generically about automated decisioning and insights. And we see AI as an emergent and critical component in that layer of the stack.

You guys talked about it in episode five with Corey a little bit. I liked the way he framed it up. We think in terms of business rules, and we use business rules today for very discrete things like our policy conditions. These are the discrete sets of logic that must be true in order for an application to get decisions, whether that's an approval or a rejection or a conditional approval.

We use machine learning, we've used that for some period of time, and a lot of our customers use it. And so, they use that to take their portfolio of applications and evaluate changes to their credit policy and things of that nature. Generally, those AI machine learning models don't always make their way into the actual runtime, but they're oftentimes helpful in informing which variables factor into a credit policy, and those things tend to get reflected today into traditional business rules. AI is the newest kid on the block, and we're super excited about it as are many of our customers and prospects in terms of opening up opportunity in various areas.

One of the problems that's really plagued, particularly business and secured lending for a long time, is just document processing. And I think you're probably as big a fan of OCR technology as I am. Anybody that's in the industry that's tried to parse a bank statement or article of incorporation or a state registration filing knows that OCR has historically been very sensitive to where things are on the page. You scan that thing in a little bit crooked and it fails. And I think some of the things that we're seeing initial success in is using the fuzzy nature of the LLMs and their ability to not just interpret text out of a PDF or a PNG file, but even reason across that. We see some good success taking a bank statement, not just pulling in the transaction history and the balances, but using the LLM to summarize those transactions into a chart of accounts that the lender may use for spreading. Or just tell me what the average recurring deposits are into this account.

Some of that fuzzy logic is where we're seeing the early initial returns. And I think it really has opened up a lot of possibilities to have that fuzzy level reasoning that AI brings to bear. That said, just to broaden out a little bit, I consider vibe coding—whether that's the right name these days or not, I don't know—but vibe coding's definitely an AI-based process that we're getting a ton of value out of.

Just last year, we did some revamping to our user experience layer and we went to a new React architecture. And in doing so, we coded that AI first from the ground up and it significantly cut our dev times. We did a tech talk internally within Q2 last year where we talked about, we used to have to get pixel perfect in terms of translating Figma designs into functioning HTML and JavaScript, and vibe makes that super easy now. And so, we're seeing great success using AI in the vibe coding world, even beyond just traditional coding. 

There's a hybrid world that we're seeing a lot of success in where, a lot of credit decisioning today does require a fair amount of deterministic logic. But writing out those rules, this object, this attribute, this operator, this range of values can be cumbersome. And I think, you and I share the belief that anytime you've got a domain specific problem to solve, it's always better to put that control as close to the expert as possible. And workflow engines and WYSIWYG type stuff has been around for a long time. And that allowed us to get the technology to be a little bit more consumable and put it closer to the edge within the business. 

That said, even in a workflow system, you can't get around writing the fine-grained logic when you're doing things like credit decisioning. And that can still be a barrier because not everybody's comfortable, not everybody has the attention span to go through and build out those rules. And so we're using AI to make it easier to take an English language credit policy and translate that into discrete rules.

Now, the discrete rules are still being evaluated at runtime when the decision on the application is made. But from a usability perspective, it makes the generation of those rules a lot more accessible to people. And it also allows us to find any logical holes in that if-then-else logic, to make sure that we have a better solution at the end of the day.

Adam Blue

That's really interesting because, if you think about from the beginning to the end of the origination, you've got a person and they don't really want a loan, they want the thing, whatever it is. That's what's important to them. And then the financial institution needs to understand the risk or whoever the lender is, and they need to understand the terms and the structure. Then there's a series of policies that wraps around that, and underneath you end up with language. It's human language really, right?

Bill Gravette

Yeah.

Adam Blue

And I remember when I first started working with Legal on contracts, a long time ago when I started at Q2, there are phrases, there are tokens really, like, "commercially reasonable effort." That actually means a very specific thing.

Bill Gravette

It does.

Adam Blue

And it's been litigated out in the courts and there's an understanding. And so when you say Company A will make a commercially reasonable effort to do X by such and such a time, you learn what that means. And so, I think, across an institution, across an FI, there's more than one loan officer and there's more than one person that participates in a decision, and loans all have their various shapes and sizes, just like customers. And do you think there's a possibility that maybe bringing AI to some of the credit decisioning loop actually creates better explainability or more consistency, more transparency, in conjunction with a human? Because, you let me know if I'm wrong here, but I don't think that we have any FIs or lenders that are telling you, "Hey, we just want AI to decide whether or not people get a loan and then we'll move on with our lives." I'm guessing that's probably not the case.

Bill Gravette

No. And I don't think that will be the case for the foreseeable future. As you mentioned, there's a lot of obligations that sit on a loan officer or on a credit team. And if you think about it at the end of the day, these loan officers, part of their job is relationship based. Whether they're working directly with the applicant, whether they're working with a broker or working with a dealer, I think you said it well, it's like, nobody goes out to get a loan. They go out to go do something to achieve their goals and objectives. And whether that's an individual or a small business, and it is a vulnerable state because you're opening up your whole financial picture, you're opening up your hopes and dreams. And so, a big part of that loan officer is building that relationship, establishing that trust, and consulting with their applicant to figure out, what is it that you're trying to accomplish? Can I assess that you're trying to accomplish that responsibly?

And then the second part of their job is really to build a business, as you mentioned. The loan officer is a steward of the lender's time. In many of our customers, they get more applications in than they can process. And so, they’ve got to make a decision that says, "Can I help this individual applicant or not? Maybe I refer them out, maybe I just tell them I can't help them." 

They're a steward of money. Lenders are either lending out their own capital, they're lending out capital that they borrowed from somebody else, and they have an obligation to do that responsibly in accordance with the terms of that funding and to show a return on that money that they're lending out. And so, you're right. It is a risk-based decisioning that they're making that says, "Hey, is this in line with the thesis that we have for our portfolio? Is this aligned with the products that we're offering in the market today or could foreseeably offer? And can I get that done?"

Because speed to answer, and this is where I think AI becomes super important, is there's a lot of variables to manage here. It's an omnichannel process. You've got multiple roles in the back office that are interacting. You've got the applicant themselves. Sometimes you've got multiparty applications where it's a business and all the directors of that business. You've got a broker or a dealer in the middle. And so, the time to a decision is a factor of the actual work that needs to be done and all the coordination and communication that happens up and down that process. And so, everybody's looking at AI as a mechanism that says, how can we get better answers quicker to our customers?

And then to your point, whether it's to the applicant, to the internal credit board, or to the regulator, explainability is important. Because if I'm going to make a decision on an application, I've got to explain why I made that decision. And then I've also got to show the work underneath it that says, "Hey, I didn't just make this decision because I thought this person was a citizen in legal standing or a business in good legal standing. I actually checked the documents and I got the documents." And so, we're really trying to think about, we think AI plays a critical role in helping orchestrate and execute some of these processes. I think the things that we're experimenting around the edges on is, what's that balance between having the structured process gates and the deterministic things that have to happen to make sure things stay in policy?

How do you allow AI to do more of the work in an automated way and to reason across the art of the possible at any given moment in time? Because a lot of this happens in parallel, but still record that audit trail and make sure we have good accountability in terms of who did what? Was it the applicant? Was it the broker? Was it the borrower? Was it a human being in the back office? Or was that a systemic action that decided to close the task because it validated that a driver's license was a driver's license and the data match?

There's a lot that goes into it and I think everybody's optimistic that the reasoning, the generative capabilities to do things like credit memos and decisioning, has a lot of potential. Even on the omnichannel communication, AI plays an important role, but it's making sure that that's all mapped to those first-class citizen nouns and verbs in terms of stages and gates and tasks and policy conditions that we've built over the last decade in the platform.

Adam Blue

I like this train of thought because I think in a sense, and I think this is still true, I think 60 to 70% of small business lending in the U.S. comes out of customers in the Q2 footprint, right? It's community and regional financial institutions. And those loan officers, and they're portrayed so unfairly in media. Every great movie about a business success starts with some stern, balding, middle-aged white guy with the “Denied” stamp, right? We've all seen that trope. But in a way, they have an extraordinary responsibility. They're the unsung heroes of grassroots bottom level market capitalism, because they're making choices about who can you really trust with capital that's been gathered from the market, from regular people, and we're all expecting that they will make very good choices that are in line with the choices that we would make if we had their knowledge and experience?

And so, it's an interesting unanswerable question, a bit of a Zen koan, which is like, how much cowbell is too much cowbell? How much distortion is too much distortion?

I love the Flaming Lips, but sometimes I'm like, "That is a lot of fuzz on that track." And so, the idea of bringing AI in to help them, and you said something really crucial there that I think is interesting that I want to push on, which is getting to a decision quickly. Because I think if you're going to tell someone, "We can't lend you this money for this thing that's part of your dream," the sooner you can tell them that, the kinder that feedback is, right?

Bill Gravette

Yeah.

Adam Blue

And so, I think one of the things about AI that some of us, a lot of us are maybe not great about is, it's hard not to imagine the zero sum outcomes. AI makes these people more wealthy, these people less wealthy. AI takes away these jobs, but it gives these jobs. I'm coming around to the view that it doesn't have to be zero sum. It could be, but it doesn't have to be. And so, when you think about it from your position and where you work, if it doesn't have to be a zero sum outcome around what AI can do, especially around lending, credit decisioning, helping loan officers, how do the wins manifest? Is there a way that we go down this path, we apply this technology, and things get no worse for anybody and better for everybody? Almost like a violation of the Pareto Principle in a sense, but is that an achievable outcome, do you think? Or am I just entirely too naive about this?

Bill Gravette

Yeah, I think it is. And that's the way we get up and work toward every day. On the win side, Q2 is a very mission-driven company. And our top-level mission is to build strong and diverse communities by strengthening their financial institutions. And that mission is founded on the belief that communities are stronger when individuals and businesses that make up those communities can realize their full potential. And we believe that that is the most probable outcome when people have options in terms of how they consume financial services.

And so, within the Symphonix business, where we're focused specifically on lending, when we first launched the brand, we created a bunch of T-shirts for ourselves and for our customers. And on the back of it said: Stronger lenders make or create stronger communities.

Adam Blue

Pretty good.

Bill Gravette

And that derived belief that says, the communities are most vibrant when the individuals and businesses there can achieve their maximum potential in a responsible way. And today, there are people and businesses that are probably stifled in their ability to reach their potential because they can't get access to the credit that they need. And a lot of that really comes down to how economically viable is it for lenders to service various segments of the market. 

And so, a lot of lenders today use a heuristic, which is a FICO score, or we deal with customers all around the world, and there's an equivalent of that, a credit score in each geography. And that's a cheat code. It basically says, "Hey, if you've had credit in the past and you've paid it off, then you're probably a pretty good bet that you're going to pay it for me."

But that creates a barrier to entry for the credit market. In a lot of markets, it takes two years to build a credit history. And you've got this chicken and egg problem in terms of, I can't get credit unless I have a credit score, and I can't get a credit score unless I get credit. 

There's other problems that that creates is, we're becoming a more global economy and people are moving around from country to country. And we've got some folks on our team that immigrated from other countries, came into the U.S. I asked one of the guys, I was like, "How'd you get a car when you got here?" He was like, "Well, I was lucky because my brother was here and he co-signed on the loan for me."

Adam Blue

Wow.

Bill Gravette

"But if that hadn't happened, I would've had to walk to work or take Uber all the time." And these are people that had credit scores in their home countries.
We have a customer in Canada. I was talking to him over lunch one day and he moved from the Bay Area up to Toronto. And he was saying like, "Yeah, man, I have a family. I've been working for 20 years. I've got established credit." He's like, "But I have $1,000 cashback limit on my credit card because my credit score didn't transfer with me to Canada." 

And so, you think about the pressure that that puts on people. And we take for granted in many cases that, "Hey, if I blow a tire and I’ve got to go get new tires, I can just put it on my credit card." Not everybody has a credit card and there's a lot of demographic shifts that are saying Gen Zers don't even want a credit card. And so, if you look around in any tire shop, you'll see a bunch of “Finance your tire purchase now, finance your car maintenance.” And that's the type of entrepreneurial spirit I think we need to foster in order to fully satisfy everybody's ability to afford to do what they need to do.

And I think technology plays a huge role in that. Because part of why we use the cheat code of a credit score is it's just not economically viable to go sit in everybody's living room and have a conversation with them about their financial picture. And so technology is a key enabler for us to be able to pull in alternative sources of data. Did you pay your rent? Did you pay your Netflix subscription? How's your bank account balance trending month over month? And so by making it more economical to go get some data that's beyond a traditional credit score, synthesize that together and build a more complete picture of people's financials lives or a business's financial life, you don't have to go have that one-on-one conversation with them. You get to a much more scalable position in the business. It then becomes profitable and socially meaningful to lend into that segment. And that creates more options, which then creates more successful communities underneath.

The real opportunity is, let's grow the pie. Let's increase the availability of credit in a responsible way so that we don't have individuals and businesses being constrained when they have good ideas or ambitions. And let's try to keep as much of that running within the community. That's why I love working with the customers we work with because, generally, they're taking money from their communities and lending it back out into those communities. And so you get this vicious cycle of improvement.

Back to your point on losses, I think loss lends itself to a scarcity mindset. And not to quote Elon in any direct way, but I think he does a great job of growing the pie. It's like, we can fight over land scarcity or water scarcity or resource scarcity on Earth, or we can just go figure out if there's rare Earth minerals on Mars and create some plenty by figuring out as a society how to tap into some of that. And so, I think overall the premise of technology is it drives productivity, which drives down scarcity and cost to deliver. And I think, that same fundamental principle applies in credit. If we can make it easier and less risky to lend responsibly into segments of the market that are underserved today, there's a tremendous opportunity out there. And the microproductivity improvements that we see on the ground within the markets that we define today are going to be a drop in the bucket in terms of the opportunity that opens up.

And so, that's how we look at it. We look at it as, as a society, we're not winning today. And so, we're just comfortable to some degree with the status quo, but there are definitely opportunities out there that are untapped. Technology's critical to opening up some of the bottlenecks around the economic viability of serving these markets and unleashing their potential. And that's where we see the net growth in the market and keeps it from feeling like a zero sum game.

Adam Blue

I think that's an interesting perspective for sure. Without letting all the worms out of the can in 30 seconds or less, so really think about it. In 30 seconds or less, what do you think the introduction of AI does around fair lending, the explainability of the lending process, the participation of regulators in the process? How do you see that intersection coming together?

Bill Gravette

I think AI has a lot of impact there. I think AI makes it easier to access broader data sets. I think AI makes it easier to conduct summarization and pattern recognition across those broader data sets. I think AI allows us to reason more completely. Human beings, we're pretty good at reasoning, but we're not perfect, and computers can do reasoning at a much faster pace. And I think communication is super important. Communication between the applicant, the lender, the lender, their credit board, and then the lender and the regulator. 

And so, today there's a lot of data that gets captured across the life of the lending cycle. And I think one of the key opportunities we have is, how do we use AI and these emerging technologies to help summarize what's actually happening and where the opportunities lie, and present those in ways that are contextualized for the regulator, for the applicant to explain a decision.

Because, I think at the end of the day, misinformation is probably the biggest risk in the process. As long as that information is flowing, even as the dataset and the pattern recognition gets more complicated, that explainability and the communication up and down the chain becomes super...

Adam Blue

I like the way you teased that out. As a failed economist, one of the things I remember from my many years in grad school is, markets work when information is perfect. And when information is not perfect, and this is why we have the SEC and insider trading rules and all that kind of stuff, markets cannot function efficiently, and then they can really be problematic in terms of their outcomes. And I think that's an interesting way to think of it is using especially generative AI to consolidate and to rationalize the reasoning that goes underneath these decisions could have a real impact on, even to some extent, the fundamental fairness of how lending happens. I think that's very cool.

What does good look like for an alternative FI lender or a financial institution in the U.S. or a financial institution outside the U.S.? What does good look like approaching this technology and taking the first steps into really using it?

Bill Gravette

Yeah, I think, good is still evolving, right? Giri, who's our VP of Engineering on the Symphonix side, he and I collaborated on a blog post or a LinkedIn post the other day. And I think we can post that out in the pod as well. 

Adam Blue

For sure.

Bill Gravette

It was called Start at the Fuzzy Edges. And so, back to my talk track before, there's just a lot of operational work that happens within the lender, that's taking unstructured information, whether that's out of an email, out of a PDF, out of a PNG, out of interaction notes that are on a loan application or on a loan contract. And it's super important to be able to take that unstructured information and structure it in a way that fits the data model and credit policy and drives that downstream decisioning. And so I would say today, what we talk about in that Start at the Fuzzy Edges LinkedIn post is, start there because that's where we're seeing the most direct applicable application of some of the newer AI techniques.

And then over time, we're continuing to experiment about how that can play out more deeply or more broadly in some of the other processes within lending. But I talk to lenders all day, every day and they're like, "Man, if you could get me out of the business of having to copy, paste stuff out of a PDF into a form so that I can then make a decision on the application, that'd be fantastic." And then, I talked to a CEO at a bank the other day. He's like, "Man, I've walked my credit floor and I see people copying data out of the form on the application screen and putting it into a Word document." And he's like, "Why are you doing that?" He's like, "Well, we have to write up our credit memo and this is what we've done." And he's like, "Can't we just generate that?" And I said, "Yes, we can."

And we've had capabilities for that for a long time, where you can build document templates and you can pull the structured data out. But one of the reasons why we haven't been able to fully automate that process is there's a lot of color commentary that goes into the notes on a loan application. And so I think we now have a tool set, thanks to these LLM models, that allows us to summarize that, put it into context, and at least draft a more complete credit memo that includes some of the color, as well as the objective data. So that data process can be more of a review, edit, and send, versus having to manually synthesize everything together.

And the analogy I use with our team is, for a long time, lending was a pet business. Every loan had a collar and a bowl and a bed to sleep in. Now it's getting into more of a livestock business where you're just trying to make sure all the cattle are running through the gates, and then you're really dealing with the exception. Occasionally you’ve got to send a wrangler out to get one of the cattle that didn't go through the gates. But as long as the majority of the herd is going through the gates, we can get a lot more scale out of these businesses, which then opens up their ability to go serve more markets and further satisfy the needs for credit within our communities.

Adam Blue

That's fantastic. I love this idea of, we're 60 years into the computing revolution, and there's still a bunch of people where a pretty good chunk of their day is CTRL C, CTRL V.

Bill Gravette

Yeah. It's the killer app.

Adam Blue

And man, nobody sat in elementary school and dreamed of being a cutter and paster. Whether you're a developer or a banker-

Bill Gravette

I always tell people that I think Excel is the killer app. It's probably clipboard. It's probably the clipboard.

Adam Blue

It's probably is. It's probably the clipboard. All right. Well, thanks very much for being on today, Bill. As we close out here, we always drop a little bit of culture at the end. And so, I have a two-parter for us today. You've probably seen the film "Minority Report."

Bill Gravette

I have.

Adam Blue

Steven Spielberg vehicle starring Tom Cruise running very fast and looking very handsome. But underneath, there's a set of ideas in that movie that are pretty fantastic.
The underlying story is from a novella called "The Minority Report" by Philip K. Dick, who wrote a tremendous amount of science fiction in the '50s, '60s, and '70s. A lot of the big tentpole movies that we've watched over the past 20, 30 years actually come out of his work. He's an extraordinarily prescient sort of writer. I recommend both the film and the original novella for "Minority Report."

And in a sense, don't we ask a loan officer to look into the future and make a decision that impacts people's lives?

Bill Gravette

It's the real ...

Adam Blue

It's not, are they a murderer? But we ask a lot of them and so, a tip of the hat to all the people that approve or don't approve loans for all the right reasons out there. And thanks for joining us today on Cut to Context.

Bill Gravette

Thanks, Adam.