Lending Better, Faster and More: Harnessing Al to Win the Future of Small Business Lending

Al isn't coming to lending-it's already here, and it's rewriting the rules. In this high-impact session, we will pull back the curtain on how banks can embrace Al to sharpen decision-making, drive deeper insights, and streamline operations-without losing the human touch that sets them apart.

Transcription:
Transcripts are generated using a combination of speech recognition software and human transcribers, and may contain errors. Please check the corresponding audio for the authoritative record.

Mike Horrocks (00:08):
Good afternoon everybody. We're so excited to be here with you today and to talk about how you can use automation and AI to accelerate your small business lending, to truly allow you to do better, faster, and more small business lending. I'm Mike Horrocks, the Senior Vice President of Corporate Strategy at Baker Hill. I'm joined today with Abdel Garcia from Montecito Bank.

Abdel Garcia (00:34):
Thank you, Mike. Thank you for having me here.

Mike Horrocks (00:35):
And we're so excited because, first of all, Abdel is an award-winning banker. So not only is he heading up business and consumer lending at Montecito, he also was recognized in the entire state of California as the Pacific Coast Times Latino Leadership of the Year. So he's definitely got it. Let's give it up for that.

Abdel Garcia (01:01):
You surprised me with that.

Mike Horrocks (01:02):
He didn't see that one coming; he did not. I actually had to verify that with his wife.

(01:10):
All right. Jensen Huang has a famous quote where he says, "AI is not going to take your job. It's the person who uses AI that is going to take your job." And so for bankers, I think we can flip that to where AI is not going to replace lenders, but it's lenders who don't use AI that are going to be replaced. I think we can all agree that AI is the equivalent of saying, "Hey, back in the day, this internet thing, we're not sure that's going to be something we're ever going to use." We want to make sure that sets the tone for where we are today. But as we look at AI and think about the business case for that, we want to make it pragmatic. That's why I'm so excited to have Abdel share what Montecito is doing and some of the things that they're looking at from an AI and automation transformation in their small business lending.

(02:06):
And really it comes down to understanding these four major areas when it comes to your customers or your members for credit unions. First of all, understanding their needs—understanding what is it that's driving the decisions behind those small business owners. Especially with a small business owner, it's not only their business, but it's their personal life; that relationship you have is very tight. And then in terms of behaviors, how do they want to interact with your institution? What are the transactions that they like to have in place? Making sure that you're able to know what those behaviors are and track those. How do they want to connect with you? And what's the context of all the interactions? Making sure that you have that full understanding. And then lastly, preference. The small business owner of today is absolutely stretched. How does he or she want to interact with your institution?

(03:08):
So Abdel, when we think about understanding the customer's needs, what are some of the things that you're doing at Montecito?

Abdel Garcia (03:14):
Correct. Well, at Montecito Bank and Trust, our approach is to enhance, not replace, the personal relationship we have with clients. For the last 50 years, that's exactly what we've been doing and it's working very well for us. Enhancing—what does that mean? It's being proactive, making sure that we understand our client needs by analyzing their data and behavioral patterns. And then again, reaching out to them before they come to us and say, "Hey, I need something." So we have a lot of reports, a lot of data, and a lot of partners. We're a community bank, so our budget is a little bit small, and we rely on partners heavily to help us with the AI piece.

Mike Horrocks (03:52):
Fantastic. So again, it's proactively knowing what those needs are going to be. And that's where the power of AI could truly help in terms of looking at their past behaviors, looking at their deposits, looking at their utilization, and truly understanding that. Now, AI is more important than ever because of some economic aspects. As we look at any business case, you have to have the ROI. And when you look at small businesses, 85% of them—and we've already touched on this today—when it comes to the need for financing, you have to have that speed to saying yes. It's making sure that they have that speed. And then when we look at an efficiency rate, we're also very concerned about that. For credit unions, there's been a 3.5% increase year over year in the efficiency rate. We want to make sure that we are actually trending that the other way.

(04:47):
That's something that's absolutely critical. And also 44% of all small businesses have expressed that there's a challenge getting credit from you all. We want to make sure that we are able to solve that. If you're looking at these efficiency ratios, again, the top line being the credit union, it's just inching up year over year. And for the banks, we see that there was a little bit of an increase, but as you look at this last part, it's a slow incline back up to less efficient lending, which is absolutely something that I know is top of mind for you guys. So all right, cool. Let's move on. Let's talk about that small business lender in the future. Abdel, you talked about that relationship, and that to us is the absolute superpower of a lender, especially in your community.

(05:44):
It's understanding that relationship, the importance of it, understanding the importance of empathy, the emotional connections that you have, and also coming up with some real creative solutions. I know in your market in the Santa Barbara area, you have some really creative needs sometimes, but tell me a little bit more about that. When you think about your lenders, what truly sets them apart and how do you see AI being augmented intelligence instead of artificial intelligence?

Abdel Garcia (06:14):
Perfect. Again, we rely on our partners. We have some great partnerships with key partners where we get real-time data from them. We're able to analyze industry trends nationally, locally, at the state level, and then across a larger geography like the West Coast. We're able to share those reports with our clients when we see something is happening to the construction industry. We run a quick query of what's available out there in terms of clients within the industry. We give them the report and share with them insights as to things we need to consider. We have tariffs coming up, or cost of funds are going up for them. Here's where we can help you mitigate some of those risks. In addition to analyzing deposit statements, we're able to proactively see if their cash cycle is getting longer.

(07:10):
We're not seeing the deposits coming in as they used to. What's going on? How can we help you? Treasury management products? Or, we saw this recently: fraud is happening a lot. We had a case where the client was sending a lot of wires and we have this tool that can give me a flag to say, "Hey, this client's pattern is off." We called the client and he's like, "Hey, that's not me." We immediately paused it and we were able to recover a lot of the funds for them. They still took a loss, but again, it could have been a lot worse. We're proactively looking at what can we do to protect our clients and help them succeed.

Mike Horrocks (07:48):
Yeah. And later on, we'll go a little bit deeper into proactive monitoring and why that's so important as you look at a homogeneous pool of small business loans, and how you can gain efficiencies through that as well. Again, it's really about augmenting the lender, not replacing the lender, helping them use their superpowers and be that advisor or consultant to that small business. And so we look at it and there's three major use cases that you all are deploying around borrower engagement—how are you interacting with that borrower, thinking about the context and their preference in terms of channels—then risk assessment. It's not just around the credit side, but also around the, "I made this loan, now what?" Because that's the biggest challenge.

(08:45):
I know as a former commercial banker, I never made a bad loan—until I got out of banking, right? So then in terms of automation and data, what else you can use across the board in terms of efficiencies. We're going to dive now into these three different use cases and talk more about how you guys are using AI to your advantage. Okay. So first of all, let's talk about borrower insights. With AI, it's opened up an entire arena of personalization and making sure that you're serving that small business owner. We've seen situations where, thanks to AI and insights from different data sources, the institution is able to see what's happening not only with the borrower, but also some of their suppliers and vendors. What are you able to do to provide additional enhanced services for your small business owners?

(09:48):
It's also been all about streamlining. You mentioned the wires, identifying and locating those transactions before it's too late, and even stopping fraudulent ones. And then ultimately, predictive analytics. From a scoring perspective, I know that's something that's really important for you all in terms of making sure that you are able to quickly approve the deals that you don't want to spend time on and quickly decline the ones you don't want. Correct. Be able to focus on the relationships where there might be a little bit more of a gray area. When you think about understanding the business owner, what have you guys done to better understand that business owner and what their needs are?

Abdel Garcia (10:37):
Perfect. Yes. Again, industry analysis is a must. We need to understand the client's needs and how the industry is performing. Once we have that, you're right: we have an SBSS score that we use for those clients where, if they meet a certain criteria of low probability of default, we approve it. For those clients that are in a gray area, we may reach out to them. We may look at more data on how they manage their accounts—their checking account, are they overdrafting? If yes, then what's causing that? All of this information takes a very long time to gather.

(11:12):
Again, we have reports where we're able to push a button and I'm able to see the entire transaction history of a client and get all the reports and data points for that. That's how we have been able to utilize AI with those tools. We also look at trends. That's the biggest thing. What are they doing in the area? How is that performing? We are in a little bubble on the central coast of California. We do not react like the rest of the country does. So those are key indicators that are very important to track and know our market as well, which we're able to get very fast with AI.

Mike Horrocks (11:53):
Yeah. And I think you hit on two great points that I'd like to elaborate on: you are in a little bit of a bubble, right? Santa Barbara is a great place. But I think everyone here could say that they are in a little bit of a bubble for the most part because there are different markets that you need to try to penetrate deeper and expand in. So there is that uniqueness to understanding the market. I think the other thing is that at no point are you allowing AI to run that relationship.

Abdel Garcia (12:30):
Absolutely not. Even with declines where we know their FICO score or SBSS score is really low and they don't meet our guidelines, we still do a second look because we still have to explain that to the client, our regulators, and ultimately our board. With AI, there are still a lot of risks involved, so we still want a second review to ensure that the data we got is correct.

Mike Horrocks (12:56):
Yep, absolutely. And so, again, the benefits of having an AI-driven experience—thinking about how it can augment—in general, what we've seen across the board is that when you're using AI to help you focus on the right small businesses and identifying those markets, there has been a 10% increase in revenue, 20% increase in satisfaction with that borrower, 60% reduction in the selection of those customers, and 13% reduction in costs, all pointing back to helping your efficiency ratios and making it more effective. Is there anything else when you think about the marketing and the outreach as you reach out to small businesses that you'd like to add before we move on?

Abdel Garcia (13:45):
No, again, it's reports that we're able to generate on industry, then we know what new legislation was passed. We're able to see what those risks and opportunities are through there and target our marketing towards those clients in that industry. It's a lot of data that we gather through a push of a button and in five seconds we have a substantial report rather than, as you all know, the RMA thick benchmark reports that are now holding up our monitors. That used to take a lot of time to go through and determine what's happening in industries. Now, push a button, we have data, we're able to target clients. Also, with competition, there are a lot of mergers happening in our footprint and some clients are not happy. We're able to pull reports on maturities coming up for a specific lender.

(14:36):
We're able to call them and gather and grow our bank.

Mike Horrocks (14:41):
And I'm glad to know that I'm not the only person that ever took their RMA peer handbooks and just used them to raise the monitor up a little bit. They have a good market there on that. All right, cool. Let's talk a little bit about AI in terms of risk assessment. That's really your main focus as you head up the small business and the consumer side of lending at Montecito. Our friends at Microsoft shared these with us: how folks are using AI to really free up their time and move away from paperwork to more strategic work. They saw that everyone's using AI on average to do 31% more reading of emails, spend 16% less time in meetings, and 58% file creation. Now, I'm not sure about that last stat, whether or not that's a good thing, unless we're letting AI do the reading for us.

(15:44):
But I know for myself, especially at times when we're at conferences, it's always great to catch up on all my meetings in minutes versus having to be in them. When you think about your lenders out in the field and just how they're using AI to free up time, what are some of the things that you're seeing? What's some of the coaching that you're trying to give them?

Abdel Garcia (16:10):
Well, again, number one is understand your client: know the data before you go out and see them. We make a presentation and in the first five seconds they know whether they're going to like you or not. The more prepared we are, the better we are at making that connection.

Mike Horrocks (16:26):
And you guys are using AI again to have those pre-call reports ready. Correct. Can you elaborate when you think about those pre-call reports, what are some insights or aha moments that some of your lenders have had?

Abdel Garcia (16:42):
Well, one of them: we had a prospect in the flower distribution business and their cashflow was completely out of whack with the market benchmarks. When we understood the client and talked to them about that, they were paying the flower growers way ahead in advance before they planted the crops and taking all the risk themselves. When we gave that report to the client, he understood and went and renegotiated all his contracts to avoid that. Now, instead of needing a line of credit, he's still funding most of his operations and having a good bundle in treasury management products to accomplish that.

Mike Horrocks (17:18):
So you maybe lost something on the line of credit, but you gained it in the treasury management side of things. For the institution, when you think about that particular borrower in Montecito, they're never going to leave you.

Abdel Garcia (17:33):
Correct. And we're here to make relationships. Exactly your point.

Mike Horrocks (17:36):
So fantastic. Again, it's really about freeing up the time so that they can have more insights into your borrowers and understand what's happening there. Also, when we look at it from a risk perspective, especially when we talk about the gray areas—and apologize guys, this is a little bit of an eye chart here, but if you want, just let me know and we'll send you the larger version—it's taking the financial analysis and pushing that all the way through from, "Hey, I got the tax returns, I'm letting AI classify those for me so I'm not spending any time 10-keying and entering the data," and then I'm giving you a first pass of what would look like in a credit memo.

(18:41):
At this point, we're introducing that analysis. The other thing that we're seeing is that it's not only the financials—what my quick ratios look like compared to last year, et cetera—but also the industry data. There's a couple of folks out in the hall that have great industry data that they can pull into the system,

(19:12):
And that's some of the stuff you're doing. But then also it's just the access, depending on whatever model you want to use, Cloud, ChatGPT, et cetera, can access and then have that be a part of the underwriting process as well. We utilize a ChatGPT model because we want to make sure that data is within your four walls. Microsoft with their Azure structure has a fantastic way to make sure the data is not going outside because you absolutely do not want to—while you might get insights that come in, you don't want trust going out.

Abdel Garcia (19:59):
Correct.

Mike Horrocks (20:01):
Yeah. So that's the model that we use. But what we see as we take this credit memo is I can now go ahead and ask for additional summaries and insights by just asking ChatGPT, "What else would you say about this?" It's just prompting the right question. At that point, we get back these responses. We get back whether or not, based on the credit policy it's learned, this aligns with your needs ultimately.

Abdel Garcia (20:38):
Our risk tolerance.

Mike Horrocks (20:39):
Yeah, exactly. And at no point is it removing your credit department. I know your credit guys well.

Abdel Garcia (20:48):
Yes, you do.

Mike Horrocks (20:49):
And they would never allow that, but it does give them a first pass so that it's, again, back to that concept of augmented intelligence instead of artificial intelligence.

Abdel Garcia (21:00):
Absolutely. Copilot is wonderful. We have some lenders who can write an essay when it should be a small sentence. With Copilot, they're able to highlight what they wrote and modify it into terms that we as credit people like and get to the point, simplify it. Now they're really getting used to it; they just give the high level and the system does what it needs to do to create it.

Mike Horrocks (21:26):
In what other ways are you seeing that when you think about the credit manual write-ups, you mentioned the industry data? What other insights are you gaining? Is there a particular example you can share with us as you looked into that? I love where you're going with the succinctness of the write-up. What other benefits have you seen in credit as you've expanded the AI use?

Abdel Garcia (21:55):
Well, we're able to track our default probability. By doing that, we can say, "Hey, what SBSS score was this client at where they are fine, or the metrics were not where we needed to be? Maybe we had an exception, but they're fine." We're able to use the data and adapt our policies because of it.

Mike Horrocks (22:15):
Okay. So you're doing that entire learning and relearning and making sure your policies adapt to marketing conditions and your risk appetite in general. Fantastic. Let's talk just a little bit about hallucinations and biases. Have you guys had anything like that come up yet?

Abdel Garcia (22:42):
We have. One of them was a perfect example in a credit writeup, and we're testing a lot of this as well. It was a credit writeup where we got a brand new owner guarantor who owned 33%. We had a very good analysis, but it was all made up. The owner was not part of the company; they had nothing. It was probably a fictitious person. Again, we have to know our clients. We have to understand we cannot let AI do the work for us. It can simplify it, enhance it, and get it done faster, better, and cheaper, but we still need the human interaction to review and ensure that it's accurate.

Mike Horrocks (23:25):
Yeah, no, exactly. I mean, some of you are trying to figure out why Nena and the "99 Red Balloons" are out there. How many children of the 80s remember this song? Okay, thank you for admitting. My daughter is not a child of the 80s, but she loves this song. Anyway, the entire premise is that somebody took 99 balloons and launched them in the air. All of a sudden, German air defense believed that it was 99 jet fighters coming from the Soviet Union. Everyone scrambles the jets because the data is telling me something's wrong. I know they weren't thinking AI at the time, but the result wasn't good—it was basically World War III, we all die, but it was a great song during the Cold War. You have to think about that. A modern-day example with data that I think most of us trust every day is like our Google Maps or Waze.

(24:18):
And again, I don't know why it was German-based. It must be a thing about Germany, and also it was 99 cell phones, but an artist took 99 cell phones, put maps active, and threw them in the back of a red wagon and started going across one of the busiest bridges in Berlin. Now you can imagine the feedback Google's getting is that there are 99 devices and 99 cars going at a snail's pace across this bridge. People would literally get up to the bridge and it would be empty except for this guy in a wagon, but Google would tell them, "Oh, you better reroute across the rest of Berlin to get there." So again, you have to be thinking about the biases and hallucinations. We were talking a little bit about the importance of verifying and trusting the data. When you think about verification, what are some steps that you're putting in place to make sure that as you use AI on credit, it's also not producing a hallucination that would be ultimately bad for the bank?

Abdel Garcia (26:13):
Correct. Well, again, it's knowing our clients and knowing who they are. Also, the partnerships we create with our vendors, such as Baker Hill, so that we can trust them and know that they're doing their due diligence to ensure that the data they're giving us is also accurate because that's really what it comes down to. We're a community bank. We don't have endless funds to be able to fund it ourselves, so that's where the partnerships come in.

Mike Horrocks (26:35):
Yeah, absolutely. Okay. And while we're thinking about risk, this is where we're going to shift from lending into portfolio monitoring. Again, when you look at any renewal, during that first 12 months, if there's nothing going wrong, there's no reason why you should be asking that small business owner. She's probably already overwhelmed trying to get new vendors and suppliers and customers. There's no reason why I should be asking that small business owner to do an entire renewal, basically.

Abdel Garcia (27:24):
Give me your financials. Give me everything.

Mike Horrocks (27:26):
Exactly. Meanwhile, if there's some slight degradation, maybe I should do a limited review. I don't necessarily need to get all the financials or tax returns, et cetera. But if I'm getting a lot of different triggers, maybe that would say, "Hey, this is a great time that we need to talk." You've already elaborated on this. What are some of the rules you're putting in place? What are some of the things you're looking at from a small business owner to help maintain that journey?

Abdel Garcia (28:03):
Well, all of our lines of credit have a 30-day out-of-debt covenant. If you meet the 30-day out-of-debt, your FICO and SBSS scores are still fine, you're utilizing the line as you should be, and you're repaying it, we see that. We require operating accounts for all of our clients. We analyze their operating statements and see how they're doing. If everything's fine, we manage by exception. If there are no exceptions, it's renewed. We go. It takes probably 10 minutes to review that and credit admin then sends a letter saying, "Yep, you've been renewed for another year." To your point, if there's an area in the yellow, then we need to reach out to the client and find out what happened. "You didn't revolve the line." "Oh, well, I purchased equipment and I ended up using the line to get it and I couldn't revolve it."

(28:54):
Okay, well, now we have the opportunity to cross-sell a product and help the client term that debt over a longer period of time and then get back to compliance. If they are severe and we saw deterioration in their SBSS or FICO, or they didn't revolve—it was fully maxed for the entire year—then we have a much deeper conversation with that client to understand what is happening with their business and then make the right course of action with our partners.

Mike Horrocks (29:24):
Yeah. What's been the interaction or the benefit for your borrowers when they have that kind of treatment from you?

Abdel Garcia (29:32):
They love it. Especially for those that are renewals by notice later, they go, "Wow, this was easy." Again, we stay in touch with them. The majority of the time we're able to see a problem before it becomes a problem and we work with our clients. A lot of banks don't do that. We're able to work with our clients irrelevant of loan amount and have all the data at our fingertips.

Mike Horrocks (29:59):
And you mentioned a couple of other things. You mentioned that you have to have the operating account. I used to sit in loan committee way too many times where it's like, "This is a great deal because we're going to have the compensating deposits as well." And I know for a fact that very rarely did all the operating accounts come over, but that's something you guys monitor.

Abdel Garcia (30:23):
We monitor that, correct.

Mike Horrocks (30:24):
You look at that very closely.

Abdel Garcia (30:26):
And to compensate for that, we offer some discounts. If you move your account, you get 50 basis points off the rate. So that's another incentive to accomplish that. Now the relationship manager is always on top of those clients who say they're going to move it and don't. Maybe at the time of renewal, we end up having a different conversation with those clients. We're a relationship bank.

Mike Horrocks (30:48):
Yeah. And you mentioned also the out-of-debt covenants. I know for a lot of folks, they may be thinking, "Well, that's just not my market." I know that one of the things you're able to do is use more of a percentage, like high-line utilization. Is that something you guys have looked at or are you pretty firm on your out-of-debt complete?

Abdel Garcia (31:13):
We are exploring that. Right now it's pretty concrete: 30 days out-of-debt. That's our credit administration department talking right there, but we're trying to twist their arm and see if they can change that rule.

Mike Horrocks (31:27):
Yeah. Because I think what that does is allow you to have a different view of the relationship. Maybe it expands your green band a little bit where you're still addressing the risk, but you're not doing it in a way that compromises the portfolio of the bank.

Abdel Garcia (31:47):
Absolutely.

Mike Horrocks (31:48):
And I think the beauty behind this is your lenders aren't having to look at every single loan.

Abdel Garcia (31:56):
No.

Mike Horrocks (31:56):
So tell us about that. What's the benefit?

Abdel Garcia (32:00):
Once you get the report set up and we have all these data points coming in, I have a report that tells me, "Hey, these clients are coming up for renewal. Everything's fine on these. They're going to be renewed by notice. These are the ones where we noticed something wrong. Call them." Then we pick up the phone and make calls. We're able to tell the clients at that point in time, "Hey, I noticed this. What's going on?" specifically, rather than collecting financials and gathering all that data that takes time. We're able to do it in a matter of minutes, quite frankly.

Mike Horrocks (32:36):
Yeah. And really the goal again is to be able to use that superhuman power of relationship.

Abdel Garcia (32:42):
Exactly.

Mike Horrocks (32:43):
To get back to the point where I'm now doing more strategic, add-value interactions versus—

Abdel Garcia (32:49):
And we spend more time on new deals.

Mike Horrocks (32:51):
Yeah, and actually growing and building out your portfolio. One of the things that we have at Baker Hill along with the lending is this concept of portfolio monitoring. It's leveraging all the different data that's out there. It's leveraging your loan data, looking for those "I'm completely out-of-debt" moments. It's looking at high-line utilization, severity of past dues, and it's building rules and situations that look at behaviors. It's utilizing third-party credit data because we realize you don't have every insight into those small business owners. They may have relationships with other institutions that only Equifax or Experian or D&B can let you know. So it's getting that third-party data. It's looking at the deposits. I know that's something that, back to the operating accounts, is absolutely critical. I know you guys are also looking at the number of overdrafts, the frequency of the deposits, and the size of the deposits.

(34:02):
Is there anything in the behavior that makes it so that all of a sudden that green-status client is now a yellow so you can get ahead of it?

Abdel Garcia (34:12):
Correct. And you mentioned the overdrafts. That has been a number one indicator for us. All of a sudden they start having overdrafts, something's going on with their cashflow. We get on the phone and talk to them before the issue comes or they come to us with a default on our hands where we have to act. We are proactively looking for those solutions for our clients.

Mike Horrocks (34:33):
Yeah. I think that's so important to call out: that proactiveness. Because, again, if I bought some equipment—and there have been multiple folks on the stage that have said speed is so critical—granted, in true credit, I shouldn't be using my line to purchase something that I'm going to have for the next 10 years. But it does give you an opportunity to term that out and have a better, deeper relationship with them because you're addressing that. I was talking to one institution and they were running the same thing with an overdraft situation.

(35:14):
It was a dentist who never had an overdraft situation at all. All of a sudden, all the flags are up. What it allowed them to do is make a phone call and have a conversation with that dentist. Come to find out, she was going through a divorce. Her husband was doing some bad things to her business and disrupting the payments and the overdrafts. It was just a real simple phone call. That wasn't the first thing on her mind; the first thing on her mind was her family. The bank at that point was able to help her through that process. I think it goes back to the relationship. A small business owner at that point knows that you're there for them.

(36:04):
You've got their back.

(36:05):
There's nothing that's going to move them away from your institution. It's absolutely critical. Any other tips you would give folks when thinking about the portfolio monitoring side? I know we talk about small business lending a lot here, but—

Abdel Garcia (36:23):
Correct.

Mike Horrocks (36:24):
Again, once it's on the books, you still have responsibility.

Abdel Garcia (36:26):
You still have to... yes, correct.

(36:28):
Honestly, it's an awesome tool to have. It allows us to become more effective and efficient with our time. We're able to see what's happening in our portfolio and track it.

Mike Horrocks (36:39):
Yeah. And you talk about the efficiency and the use of time. We've seen on average institutions that use a portfolio monitoring tool like this—we had one client where it was about 200 accounts per portfolio manager. I don't know how many your portfolio managers were using, but that's a full-time job. With this, because we're now only looking at outliers, they were able to move that to about 2,000 accounts per portfolio manager. It's about 10X.

Abdel Garcia (37:18):
Scalability is awesome.

Mike Horrocks (37:20):
Yeah. So it gets back to the efficiency

(37:22):
And again, really addressing just the outliers and letting AI handle the business-as-usual moments. When it comes to AI in terms of ROI on this, our friends at Microsoft shared this with us: for every dollar you spend, you get about $3.70 back in terms of efficiency and gain. So definitely something that you guys are looking at to increase performance. We're going to go now, in the last few minutes, into the third example, which was around automation and data assets. The use case here isn't necessarily around the concept of lending, but around compliance. Because I know that you love compliance, I love compliance, everyone in this room loves compliance, and we want to spend as much time doing compliance as we can—not necessarily the case.

(38:28):
What we've also seen is that if you allow data from your existing compliance, policies, and guidelines to leverage AI, we've had clients that have been able to take their original costs and get it down to about 10% of what that would cost normally. You can see some of the stats here: for a new policy creation, you're going to spend around 60,000 a year. But if you could compare that to, "Hey, I'd rather only spend 6,000," that's a considerable advantage. I never heard anyone say, "We are going to beat the market by being the most compliant." It just doesn't happen. Again, not saying anything bad about compliance—you have to have it—but that's the kind of use case that's out there. Just something to think about.

(39:30):
Our chief information security officer was talking with me the other day and said, "Think about the five things you hate doing and just AI the heck out of those." I think that's great advice: literally think about the five things that you hate doing around small business lending and how you can do that. I'm not going to ask you what you hate because you love everything about it, right? But think about that. All right, so let's start wrapping it up here and some final thoughts. I had this really good idea: I'm going to ask Copilot to come up with a beautiful slide about the five things we want you to avoid. And this literally is what it gave me. First of all, for those that want to keep doing the math, I got one, two, two, three, five.

(40:30):
Okay, first of all, there are only four columns. Somehow I got up to five. I don't know what "travelling" is, but I want to make sure that I don't ever travel. I want to avoid that at all costs. There are just so many bad things with this. I told it what to do. I said, "These are the five things I want you to avoid." And I don't know how it came up with this junk. I only show it because even if it's good at some things, it's not great at all things. You don't want to take your small businesses and put them at risk for this kind of treatment, and you don't want to hazard your bank there. Really, when you think about it, there are some common mistakes that you ultimately want to avoid. We talked about the lack of testing and how that could be a thing.

(41:31):
We talked about how it doesn't replace the relationship; it augments the relationship. I know there's a lot of talk about agentic AI and what I should be allowing agents to do, but I think it goes back to that concept of the five things I hate. But the biggest thing is: don't treat AI as a silver bullet. It's a tool, it's good, it's important, but make sure you have the guidelines for that.

Abdel Garcia (42:01):
We have to. We have to have the guidelines and policies around it. We also have a committee.

Mike Horrocks (42:07):
Yeah. Tell us a little bit about that as we wrap this up. What are you guys doing in terms of setting those guardrails and the foundations around AI utilization?

Abdel Garcia (42:17):
Well, number one, we follow the "crawl, walk, run" approach so that we make sure we're not going to jump into the deep end of the pool without knowing what we're getting into. So we have a very strict vendor management process and an AI committee that manages and actively does research on this topic. Before we engage with a vendor, we put them through the wringer. That's the honest answer. We make sure that our risk tolerance is not violated. And if it passes everything, we'll go.

Mike Horrocks (42:51):
Yeah. So again, it's making sure, because right now it is the Wild West. When we're looking at AI in terms of the regulations, they are non-existent. Guidelines—okay, there are some guidelines, but it's more common sense. I was talking to our chief information and security officer and the best advice he thought you could give is: find the guy or gal who was there when your institution said, "Hey, I have this thing, it's called the internet. How are we going to use it?" Because think about what they had to go through. "What sites can I go to? Who can use this? Why could they use it? What's the business case for it?" Go through that because that's about the most revolutionary comparison to AI. Just to kind of echo your points: focus on where it can solve problems today.

(43:51):
Go deep in some areas. Don't think you're going to have one model that addresses everything. Definitely trust and verify and don't do anything that your institution wouldn't be comfortable with. Again, on the portfolio monitoring, you guys are very comfortable with that. You're running it. We run millions and millions of rules every day against different portfolios and it lets you address the needs, but again, you've given guardrails to it.

(44:20):
All right, so we're wrapping up here. Any other things you would advise folks as they're thinking about AI being in a much more pragmatic environment?

Abdel Garcia (44:30):
Yeah. Again, follow the "crawl, walk, run" rule. Don't jump into it. Be mindful. Start maybe with some automatic data collection or documentation through portals, making sure that we're able to get that in a fast, efficient way. When we first started with that, we saved probably 90 man-hours a month by collecting financials from clients. Those small efficiencies are a big win for our organization, especially smaller financial institutions, and you go from there.

Mike Horrocks (45:04):
All right. Fantastic.

Abdel Garcia (45:05):
Yeah.

Mike Horrocks (45:06):
Abdel, thank you so much for sharing your time with us. Thank you. Thank you so much.