Can AI make the mortgage market easier for lenders and borrowers? Matt Sanchez of Cognitive Scale explains how..

With interest rates rising and the economy slowing, the world of credit is getting a lot harder for both lenders and borrowers — and that’s especially painful in the mortgage market, given how high house prices are. 

Matt Sanchez, Founder and CTO of CognitiveScale thinks that Artificial Intelligence can help: it can help loan officers identify potential borrowers ahead of time, and it can help borrowers get access to the right mortgage offers. Can it also help make the lending process more fair? Matt , an IBM veteran and a board member of the Responsible AI Institute, believes it can. 
In this talk with hosts Michael Leifman and Marco Annunziata, Matt explains the value that CognitiveScale’s TrustStar solution brings to the table, and how. TrustStar is only one of the AI solutions developed by CognitiveScale, so you may hear more from Matt…

Transcript

[00:00:00] Marco Annunziata: Hello, everyone. I'm Marco Annunziata from Annunziata + Desai Advisors. 

[00:00:12] Michael Leifman: And I'm Michael Leifman of Tenley Consulting. 

[00:00:15] Marco Annunziata: And this is Macro Micro Michael Marco & Startups at the Edge. The podcast about startups with technology that changes how the economy functions. 

[00:00:25] Michael Leifman: Welcome back and thanks for being curious. So let's get going. 

[00:00:28] Marco Annunziata: Matt Sanchez, founder and CTO of CognitiveScale, welcome to M4Edge. 

[00:00:34] Matt Sanchez: Thank you. Glad to be here. 

[00:00:35] Michael Leifman: So, Matt, as you know, our show explores the micro and macro aspects of technologies that are changing our lives and our economy and we focus on those changes that feel especially urgent given what is going on around us. With you today, we wanna focus on finance and credit conditions. Before we get started, a few facts and stats to help us set the stage. 

[00:01:00] Marco Annunziata: And here it's over to The Economist for over a decade between 2009 and 2020, we had lived in a world where inflation was low, interest rates were extremely low, and liquidity was abundant. For both individuals and companies, getting credit was very easy. It was kind of like Lake Wobegon. All companies looked above average. Then things changed. 

[00:01:22] Michael Leifman: Inflation jumped from less than 2% to more than 8% in the US. The Federal Reserve raised short-term interest rates from about zero to about 4% and it's not done yet over the last 12 months. We're now in end of November as we are recording this of 2022. For the last 12 months, 30-year mortgage rates jumped from about 3% to 7%. 

[00:01:48] Marco Annunziata: And banks and other financial institutions are becoming a lot more careful in how much money they lend and who they lend to, and life for borrowers has become a lot harder and more expensive. 

[00:01:59] Michael Leifman: And with that as the background arrives CognitiveScale, your company, which specializes in artificial intelligence solutions, one of which called TrustStar, seems especially well placed for this moment, so.

[00:02:15] Marco Annunziata: So, Matt, we want to discuss TrustStar, but can you first give us some context for our audience and spend a bit of time talking about what CognitiveScale is and what it does and, in particular, what your Cortex AI platform is? 

[00:02:31] Matt Sanchez: Absolutely. So, just a quick background, I started CognitiveScale in 2013. Before that, I was at IBM for about seven years. In the last three years of my career at IBM, I led the Watson Labs team, which was the commercialization unit around the IBM Watson technology. So really, the first group taking the research project that was IBM Watson in 2011 and creating commercial solutions with it in the marketplace. And that was a great experience. Learned a lot about what we now call "AI" in the enterprise. But back then, we didn't even call it that 'cause AI was somewhat of an old term. Nobody really used it. They called it other things. IBM came up with a new term called "cognitive computing." But ultimately, "AI" became the term everybody started using again. And now, here we are today and AI is really everywhere. And in the enterprise context, AI has seen a huge resurgence in the last decade. 

[00:03:24] And so, CognitiveScale was formed to help enterprise customers generate value, generate lasting business value using AI. And we do that through helping them with a platform and a set of tools in a space we call AI Engineering. So being able to apply the right engineering principles around building AI systems, deploying those systems, monitoring the business value that they generate, and being able to do that at scale. So using the resources they have within the enterprise IT landscape to be able to participate in that. We look at AI as a multi-role team sport in the enterprise and there's really been no playbook for how that team operates. And AI Engineering really is about creating that playbook and allowing all of those roles to participate and to drive efficiencies of scale around building AI systems. And there's many facets to that problem, including how you compose different types of models and data; how you govern a system like that because it is so dependent on data and maybe probabilistic models and other elements that are foreign to the typical IT environment; and there are, you know, the need to align to the business. So we call the business observability of AI — how do we make sure that people understand if it's generating value or not and what does that look like and how do you calculate it? 

[00:04:39] So, CognitiveScale is a platform of Cortex AI platform that is designed for customers to build, to use, supply those AI engineering principles, and to build AI systems. We also have, over the years, our focus, early days of CognitiveScale, our philosophy really always was lead with an application area and a vertical and then differentiate ourselves with our platform capabilities. And so, we started in the healthcare space and really looking at applications like care management and other areas where AI we thought was really a good opportunity to apply AI. And over time, we decided that we also wanted to have our own applications, TrustStar being an example of that where we saw an opportunity in the mortgage space in the US in particular about a year and a half ago, where we saw that customers, buyers of homes, as well as the loan officers and the real estate agents, a lot of the participants when you think about home buying, there's a lot of data out there. There's a lot of websites you can go to and search for things. There's a lot of information that's opaque as well to these different actors. We wanted to make that all simple, particularly for the sell side of this with loan officers and lenders being able to identify quickly the referral network that's gonna help them identify qualified buyers, connect those buyers to the lenders, and really drive a lot of streamlined insight in that space without the end-user having to deploy an AI platform or build their own data science team or find their own data. 

[00:06:10] What we sought for with TrustStar was really how do you do this in a way that, you know, within a couple of clicks in a simple application, you can find that next referral partner if you're a lender that you need to go work with to to identify new buyers? So, we brought together, we use our own platform to do this, of course, and within a relatively short amount of time, we brought this to market as a pure SaaS application that is directly sold to the end-user, in this case, loan officers. And we have other things we're doing with it as well, but that's where we started with TrustStar. So we introduced that into our business model last year — sorry, earlier this year. But, you know, that's relatively new in the journey and it's been really exciting to see the uptake of it, and it's a way for, you know, real end business users to get the advantages of this AI technology without having to do any of the AI work essentially that's required and to do that, introduce that into the market quickly. And that's how we thought about TrustStar. So, different way to monetize their technology but something we're really excited about.

[00:07:09] Michael Leifman: So that was very useful. Thanks. It seems like what you initially recognized or what you initially hypothesized as the value proposition was facilitating, if I got this right, facilitating lenders finding new loan applicants. On the website, from the material we've read, it seems like TrustStar does a few other things that has some help with affordability index and from risk reduction and a few other things. Can you tell us a little bit about what TrustStar does? It seems like it's gotten these additional features and I'm curious. In addition, are those things newer or was that part of your original hypothesis of how you could help? 

[00:07:48] Matt Sanchez: Good question. So our original hypothesis was very simple. What is it that the lender needs that they don't have access to today? Or if they have access to, they haven't been able to bring it together with a bunch of other data to find the insights that are going to help improve their business, find buyers essentially in the market? And as we evolved that vision, we started to realize that there was a bunch of other challenges. Like for example, which loan officers should I be hiring to help me improve that business? What kind of compliance issues do I have, and can I get ahead of those faster and easier? Every lender has to deal with fair lending compliance challenges, but it's typically done on an annual basis. It's somewhat done in hindsight. We've heard a lot of lenders want to think about it more in a real-time basis. Can we do that? It's kind of expensive to do that analysis. And again, it's something we can do very easily really with very little effort with the data that we had. And so, we basically said, look, we have to pull together a lot of data to make this work. We can't ask our customers to have it to pull together if they don't have it. So, we've done that as well. 

[00:08:50] And as we've built this set of data in the space, which includes loan information, it includes real estate information. You know, for example, we have what we call "profiles," "entity profiles." Around 21 million properties in the United States, 2 million real estate agents, probably about 600,000 real estate offices, every single lender and every single loan officer in the United States — we have data on all of this. And what we've been able to do is create a high fidelity data model using the Cortex technology and something we call "profile of one" where we can link across these different data sets. And so, if you're basically, let's say you're a hiring manager at a bank and you wanna figure out who should I hire, I want to answer a bunch of different questions. I wanna know where, what loan officers are in the regions that I want to go after. What areas should I be going after to identify talent and to bring them into my organization? Where are the buyers today? Do those loan officers have those buyers? Are they bringing new buyers, new purchase business into the bank? All of these questions they have to answer and it takes a lot of work and it's a lot of guesswork. In today's market, as it turns out, that is probably one of the most important questions that the lenders are trying to find, trying to answer. Because if you think about the market and how fast it's changed, a year ago today, if we were sitting here talking about this, this would be still an important problem. But if you look back in the past year, refinance because of the low interest rates was the name of the game. Most of the lenders did tremendous business in 2021, 2020 because of refinance. Well, now, it's all changed, and it's changed faster than it's ever changed in the industry with the interest rate increase and the spike. Couple that with the year-over-year increases in housing prices and, of course, other inflationary pressures. It's a very different market. It's probably one of the most rapid changes in a market we've seen in a long time, especially in the mortgage market. And so, now the key problem is how do I find those buyers? How do I find the qualified buyers? Where are they and who's aligned to those buyers? Which loan officers are there? Which realtors are? Where are those opportunities? 

[00:10:55] So your question about affordability. Affordability is an important component of something we call the "loan officer opportunity score" in our system. Think about affordability as one of the components. I wanna know where there are affordable housing, where is affordable housing? And when we say "affordable," we're talking about conforming loan limits. We're talking about where conventional mortgages are available, which just changed yesterday and is changing, changes quite often actually in various ways. And it's regionally, it's really specific to different regions, right? So the conforming loan limit in the US Virgin Islands is different than the conforming loan limit in Texas. And so, you have to understand all of these dimensions and then you have to be able to present a view to a loan officer that says, here's where your opportunity is today. Here's where there's affordable homes. There's lots of supply. There's good buyer demand. And make that very simple to them; has to be very easy for them to understand. We can't show them a bunch of statistics and numbers and charts. They want to just see a heat map, a simple heat map that says this county's very affordable, it's got a lot of supply, and you don't have any competition here right now. This is where you should be digging, right? Here's the top five realtors that are trending in that particular area. This is who you should be talking to right now. That's, you know, essentially what TrustStar does for the loan officer. That's one example. So, we use all that information to surface exactly an actionable insight that gives them value immediately.

[00:12:18] Marco Annunziata: That's a very clear exposition, Matt, of the value that you bring to your customers, the loan officers in particular. Now, a clarification for our audience. When Michael was saying something changed yesterday, Michael, you were referring to the increase in the size limit for mortgages eligible for backstop from Fannie Mae and Freddie Mac exactly, for mortgages exactly, which happened yesterday exactly. And Matt, you've talked about buyers and the importance of buyers in these two-sided markets. So, can you tell us how does TrustStar in the end help borrowers? Because presumably to generate more business for your loan office or clients, some additional value must be offered to borrowers as well.

[00:13:00] Matt Sanchez: Yeah. So, typically a borrower, you know, I think about it in my own experiences. You know, you're buying a home, you're working with your realtor, and unless you have a preferred lender in mind, you're looking for referrals, right? You're looking for, you know, who's a good lender? And you may ask your, you know, your social network, your friend network. You may ask your realtor. There's a lot of people. Do you have some recommended lenders you want us that you work with? Who would you work? That's referral business. And that is really what drives a lot of the industry is referrals. If you're a lender, it's business from your existing clients that you've worked with in the past and it's referrals you get from realtors and other partners in the network. 

[00:13:38] So, we came at it from a different angle. Instead of going after the buyer side, which there's a lot of tools out there to help buyers today in the market. Think of Redfin, Zillow, any tools like that. All sorts of tools that are out there for buyers to understand the real estate market and even tools to help them understand loans like who are the lenders. We came in from the other direction, which is we said if we can build higher quality referral networks, we can help our lenders build those referral networks. We could also eventually provide realtors the same view of that information and ultimately go after the entire ecosystem. But we came at it from the lender side. And so, we think the higher quality of those referral networks, the better it is for everybody. You're getting realtors who have a better understanding of which lenders they can work with. Lenders get a better understanding of which realtors have buyers that match the profile that they need from a product perspective in terms of the products they offer. And so, it makes the whole network more efficient. It's just that today we're working on it from the lender side. 

[00:14:36] We could eventually work at it from the buyer side. We've talked about that. It's a different animal, right, going and working with buyers. Instead of going after that directly and the one thing we've learned about the property tech space in general is it's a very rich ecosystem of different products and technologies and we are working within that ecosystem, partnering with others who perhaps have connectivity into the other parts of the network that we're not focused on. So, our strategy right now is to enhance those other ecosystem partners and what they do with what TrustStar has and that gives us a broader reach. So, our strategy is to stay focused in our swim lane, but then to have a reach, a network effect as we go out into the ecosystem working with other partners. 

[00:15:21] Michael Leifman: Yeah. No, that's great. I think having that network effect, it addresses a little bit of this next question I have, which is around fairness in lending. I'm gonna combine a couple questions here. One of the elements of fairness in lending has to do with bias in lending. And there's been a lot of discussion in the last few years as algorithmic lending scores have popped up and the concern has been that we are basically encoding human biases into hard-to decipher-algorithms. And where you once might have more easily found an existence of bias, it's harder to determine that because of the, quote, black box effect of some of the algorithms. So, how do you address that issue within TrustStar? How do you ensure fairness in the algorithms? How do you ensure against or protect against bias?

[00:16:11] Matt Sanchez: Well, there's multiple things in your question I'll point out. One is you mentioned something important, which is human bias. And when you think about making a decision, like a lending decision, at the end of the day, if you're the buyer, you're the one on the other end of that decision. It doesn't matter that it was a human or a machine. If you were treated unfairly, you know, you're unhappy and you wanna know why you were treated unfairly. So, that has to be addressed uniformly, meaning if we want to address fairness issues, we have to look at the entire system, whether it's algorithms, people, some combination of both. We have to address all of it, and we need transparency and explainability in that system to be able to address that. 

[00:16:48] So, we've taken an approach of really providing that. One of our technologies, we have something called "Certifai," which essentially looks at fairness within these different algorithms. We can measure it in terms of what we call "burdens." So, if you are, depending on how you define different protected groups of individuals, can you show that one group has a much harder time getting a fair decision or an optimal decision versus another group? And how do you measure that over time? So, we have technologies like that that we used to apply to different algorithms. But ultimately, the algorithmic assessment is only one ingredient. It actually then has to bubble up into something that is interpretable by the regulators and legally accurate as well. And they're very well established standards for fair lending where they do analysis of lending decisions, again, independent of whether they're algorithmic decisions or human decisions or combination of both. They have a way of analyzing that. 

[00:17:43] And so, we use a combination of our own technology as well as some new functionality we've created to be able to generate those analyses, those reports, and to do so on a real-time basis. And I mentioned that earlier because a lot of times that's a, it's more like an audit process where, at the end of the year, you provide all this information to a third party. They come back a couple months later with a report and it's expensive. And it's, somewhat, I would say a cumbersome process. We wanted to make that much simpler. And with the data that we have, we could make it instantaneous. And that's where we think providing that additional transparency adds a lot of value to the system as well. But our point of view is that, you know, the way that we think about AI systems comes from our building AI systems for the last many years is that you have to start with trust and transparency built in from the beginning. That has to be part of your development process. That has to be part of your infrastructure and your release process. It has to be as essential to your development process as security scanning or any other aspect of it that you think is key for quality software development. When you're dealing with AI, you have to incorporate trust scanning as we like to think of it into the system as well. And so, that's part of our development process internally. But we also think, you know, from a TrustStar perspective, providing some of these tools out of the box and these views, it's value added and it certainly helps the whole ecosystem.

[00:19:05] Marco Annunziata: That is excellent. One last question before we let you go and it's the following: How do you see the lending and the mortgage market changing in the coming three to five years, thanks to technology like yours? In particular, one thought that comes to my mind. Do you think that innovations like yours can help establish the lenders and loan officers content with the challenge coming from new FinTech companies?

[00:19:28] Matt Sanchez: Yeah, that's a great question. I think the next several years we're gonna see interest rates stabilize around something that's probably historically still relatively low but higher than it has been in the previous four or five years. I don't think housing prices in my view are gonna drop substantially. I think they may ease off in certain markets. They'll continue to be strong in other markets, so that's not gonna change. I think the conforming loan limits being adjusted to the extent that they were recently as an acknowledgment of that in some sense; that, you know, some of these increases are here. So, you're still gonna have a market that's got, it's kind of an interesting market. It's got very tight supply still and the demand has decreased as interest rate, you know, almost exactly as interest rates have gone up. The demand has also gone down. That's not gonna last forever. That's not the way that the housing market has ever worked. It will stabilize over, but it may take a while. It may take a couple more years for us to get to back to some sort of steady state where interest rates sort of stabilize and the prices stabilize and the market starts to provide, you know, have the right kind of dynamics to it again. But it's gonna be a difficult, challenging housing market, at least for the next couple years, in my view. I'm not an economist. That's just what I see in the data and what I'm hearing from others. But that's kind of how I see it. TrustStar is gonna help and tools like TrustStar are gonna help because as we've talked to loan officers, they don't have the time. I mean, they wanna spend time with clients. They wanna spend time working and finding and closing business. They don't have the time to even research the kinds of things that TrustStar provides to them. We wanted to set out to provide that level of insight to the end-user without having to go through any extra work. Make it very simple. Make it very actionable. The more actionable we get in this space, the more efficient it can be for everybody. There's just too much guesswork involved and too much work in general involved, technical work involved, to bring all the information together and get to that insight. And so, our vision has been to provide that on demand in a very simple way for the end-user so that we can make that entire system more efficient and a better experience for everybody involved. With TrustStar there's actually, we have maybe another half a dozen modules we have in mind that will in the future that will address different aspects of this. So, stay tuned for that. More to come. But we think there's a huge potential here for more TrustStar-like things in the mortgage industry and potentially other industries. I'll just leave it there. 

[00:21:50] Michael Leifman: Sounds exciting. We'll have to have you back once those new products are out. 

[00:21:54] Matt Sanchez: Absolutely. 

[00:21:55] Michael Leifman: Matt, we'll let you go. Matt Sanchez of CognitiveScale, thank you so much for joining us. Sounds like really interesting technology, so thank you for explaining some of it to us. 

[00:22:03] Marco Annunziata: Thanks very much, Matt.

[00:22:09] Outro: Hey, if you enjoyed this episode, which we hope you did, please leave us a review on Apple Podcast. Also, try sharing it with a friend. Post it on social media. Help us spread the word. Find other listeners who are curious like you. Thanks.