How mortgage lenders are looking to use AI in 2025

Partner Insights from

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.

Heidi Patalano   (00:37):

Hi, I'm Heidi Padano, editor in chief of National Mortgage News. Here with me today to discuss how lenders are planning to use AI in their tech stacks in 2025 is Leo Price, senior financial Technology and innovation specialist with the Federal Housing Finance Agency. Thank you so much for joining me today, Leah.

Leah Price (00:57):

Hi Heidi. Thanks for having me.

Heidi Patalano   (00:59):                                                             

So you are coming at the subject from a very interesting position. Prior to your time with Fannie Mae and Freddie Mac regulator, you were at figure as the vice president of the lending ecosystem, and before that you were the manager of product management with Fannie Mae. I know that over the past summer you also ran an AI tech sprint with some of the top lenders in the field to draft some forward-looking ideas for how artificial intelligence can be used, which I'd like to talk about in more detail later on. But first I want to ask, in what processes have you seen the most pickup in adoption of AI powered technologies Ever since chat? GPT made such a huge splash two years ago.

Leah Price (01:43):

So Heidi, I love that we're diving into the splash of generative AI right out of the gate, but before getting into the specifics, I want to start by taking a step back if that's okay. I know the tech is super cool, but it's really important for lenders to identify the problem they're trying to solve. First, are they having trouble attracting customers? Does a process take too long? What exactly is the issue? Well, technology like gen AI is really exciting. No one should be lured into deploying tech just for tech's sake, and I'm saying this because I saw this impulse also in the blockchain space, which many have criticized for being a solution, looking for a problem, and that's just the wrong way to think about any new technology. So the first step, focus on the problem you're trying to solve and then whatever the challenge is that will dictate the kind of technology that's best suited to solve it.

LeahPrice.JPG

(02:40):

Maybe that's ai, maybe that's not ai, maybe that's blockchain, maybe that's something else. Super cool. Maybe it's a combination of all of the above. And so after you figure out what the problem is, then you have to define what you want to achieve. What are the goals? How do you measure success? Is it improving things for staff or customer service? Enhancing operational efficiency, reducing costs whereby how much, all those fun questions. And then once you've prioritized the use cases and have clear goals, then lenders should be experimenting in some kind of a controlled environment. So test things out, understand what the new risks are. Obviously everybody needs to keep up with the regulations and ensure that everyone's compliant with all the existing laws. And I think we all know that with ai, that means you got to pay coast close attention to data privacy and fair lending to make sure you're not introducing new risks.

(03:42):

So what I just described is aligned with what we on our team called responsible innovation, which just means while you innovate, you have to also consider risks. And so for anyone who wants to know how lenders are thinking about AI in their tech stack, great place to look is shameless plug at FHFA's YouTube channel. We have an awesome YouTube channel with all kinds of content. We have a demo day video there that you can watch and that spells out exactly, it was mostly lenders participating in that event. So that's where you conceive 12 different use cases that lenders came up with. And there I'll point out what we saw lenders propose was mostly internal applications for AI rather than consumer facing.

Heidi Patalano   (04:32):

That's interesting because there have been so many co-pilots that have been launched by a bunch of lenders over the past two years. Rocket UWM, a lot of them are launching these AI power tools that are really just meant to help answer questions for their own employees on originations and all parts of the business.

Leah Price (04:52):

Yeah, yeah, I mean that definitely seems to be out of the gate what people are thinking about using gen AI for, I mean chat GPT that we all use in our personal lives. It really, it's so easy to see that application for chatbots both definitely internally, but I think lenders are just starting to tiptoe on what that could mean externally because there is, I mean we read about in the media some terrible examples about chatbots saying completely wrong information to consumers. And the thing to think about is that we're in the mortgage industry that's considered further behind in terms of technology adoption, but meanwhile, big tech every week they're improving their models. There's a breakneck development going on, so the models are just going to get better and better over time. So things could look different even in a couple months.

Heidi Patalano   (05:50):

It definitely seems like people feel the most comfortable trying to use it for internal processes than anything facing because of course there's a lot of unknowns in terms of proving there's a lot of safeguards that need to be set up that maybe are still

Leah Price (06:07):

Exactly. And it's interesting that there are so many think about things like guideline searches and state regulations that a human being can't possibly maintain all of that in their brain. It makes sense that you'd want to relieve some of that and put that into a model to be able to handle.

Heidi Patalano   (06:34):

Yeah, that makes sense. So yeah, that was one of my questions that I was wondering about. That's the most attractive thing, kind of taking this huge amount of information that they need to know and making it just easy to digest. It's not really so much in the actual process of origination itself and underwriting itself because those, well, I mean I know AI is used in those bases, but they're checked and checked and checked, so that's not really,

Leah Price (07:06):

Yeah, well, so that's an interesting point. We started off talking about chat PT and gen AI and AI has been used for many years, so there's a lot of this that isn't new at all. Let me share with you, so I was looking at some research that Fannie Mae put out in October, and what's interesting about that research is that they conducted a survey in 2018 and then they conducted that survey again in 2023 and they published the results recently. So it's like before chat GPT, here's how all the lenders were thinking about AI and then post chat gp, what are people, did people say, and just so I make sure I cite this so that a listener could find it, it was a perspectives blog by Fannie Mae's, VP of modeling and data science, Peter gmi. And so the main takeaway from the survey, and this was all mortgage lenders of all different sizes.

(08:06):

They cited improving operational efficiency as the primary motivation behind adopting ai. So in 2023, it was 73% want that benefit, and it was 42% in 2018. So that's what lenders are telling us they want to use AI for. And that's AI broadly, not necessarily gen ai. And specifically they said they're most interested in automating compliance review, especially depository institutions said this. The second most appealing idea was anomaly detection automation to help with identifying fraud or defects Early in the underwriting process. That survey, Fannie also asked lenders what they think the GSEs should do with ai, and they said they wanted to see more appraisal automation, more borrower income and employment verification options, data documentation, reconciliation and standardization and compliance management. So everybody is really honing in on this compliance risk space. In that study also our tech sprint, you had to compete in a different category.

(09:27):

So each of the 12 teams had to select a category and one of them was compliance and risk management, and every single team wanted to compete in that compliance and risk management category. This is interesting because in the broader context of technology, there's so much chatter in the media about the existential threat of AI and how this is the end of humanity and robots are going to take over the world, but in the mortgage industry, we're really inclined to use this technology for actually managing risk and compliance. So it's a really different kind of position to be in versus what everyone else out there in the world is talking about. I'll say that one of the questions in the survey also was for lenders who have not used ai, which there are many lenders that are not, what are the barriers? Why not? And they said integration complexity with the current infrastructure. I mean we're just in an old industry, lack of proven record of success and high costs.

Heidi Patalano   (10:50):

Yeah. Well that's interesting because I did want to look at it through the lens of IBS of all different sizes and what a smaller IMB can do versus a mid-size or large one. Obviously previously we were talking about UW and Rocket creating these copilots, but then with the smaller, the mid-sized ibs, if they want to be competitive next year in 2025, let's say, should they be trying to create these resources by using notebook LM from Google to load up all their rules and then you can use that to search Should they be trying to find these ways to use the same kinds of tools that the Rockets and uws are using so that they can stay competitive, but also if they're not building them themselves, they're not ensuring certain safety measures? I was just wondering what your thoughts were about

Leah Price (11:47):

Yeah, so we talked to a lot of lenders. We talked to a lot of vendors in the industry and then we obviously talked to Fannie and Freddie as well as big tech. So I'm going to say I don't think lenders are there yet. I don't think the little guys are chomping at the bit to be competitive with their use of ai. And there's a lot to be learned. Some companies are really forward about disclosing or promoting their AI tools. Some are more quiet. So it's hard to say when somebody puts out press releases and somebody else doesn't, who is actually ahead because it is a very competitive space and some of these players want to keep their competitive advantage.

(12:38):

So I want to point to something though that was interesting in the survey related to this. Interestingly, in 2018, the top reasons lenders wanted to adopt AI back then was to improve the consumer or borrower experience. However, in 2023, that faded significantly. So in 2023, 7% said they want AI for consumer experience improvement 41% in 2018. That's pretty significant, and it just points to the increasing fear of the risks associated with putting that tech out there in front of the borrower. We also saw that data security and privacy have grown significantly since 2018. The other thing that's interesting from that survey related to consumers is that there's an increased stressing of the importance of the human touch to the mortgage business, particularly as it pertains to customer interactions. And I think this is very interesting thinking about our industry versus other industries is that there's a fear of robots taking over our jobs, but we as an industry or these lenders that we surveyed said no, we actually really value the human element to everything that we do. So that'll be a really interesting development for how we move forward relative to how other industries move forward.

Heidi Patalano   (14:34):

Well, I think that's interesting to note that fewer lenders were interested or smaller share. We're interested in how to put this in front of consumers going, and I know that you're part of a governmental agency and you can't comment on things, but we're recording this during following an election and there will be a new administration that may have a different philosophy on enforcement of some of these rules. So I think it'll be interesting to see how this changes in the future. But speaking to that, I wanted to ask, we have this AI technology, it's hearsay in mortgage lending even if we are rather slow to adopt it as an industry, how do you think lenders can future-proof themselves against making costly investments in technologies that can become, in a sense, a money pit once there's new regulations or there are new safety protocols that they have to do or things like that? What does financially responsible innovation look like for a lender?

Leah Price (15:48):

Maybe we'll take that on. So I mean, I think this situation regarding lenders future proofing their technology investments, it hasn't changed just because it's this super hypey new technology. I mean, the truth is that it's important for lenders to pilot the technology. First you got to explore, understand, and test it, and that takes time. It's not like you can just dive into something you've never done before, spend a bunch of money and expect that to work out well. So you got to see if the investment will even be worth it after you've taken some time to test it out. And no matter what's going on, there will always be rules and laws with business. There will always be rules and laws with technology and those can change no matter what's, no matter who or what is leading the way. From our perspective now, we want to help lenders be able to identify and mitigate technology risks, whatever they are, which is why the use cases at our Tech Sprint were judged as much by the impact or the benefits as they were by how well their control measures mitigated the risks. And one thing that we heard is that Tech Sprint participants and the lenders and the vendors and the GSEs, everybody wants clear regulatory guardrails. And you hear about that in the news also, I've heard that big tech is saying we want clarity. And that's not different to our industry. Everybody wants to avoid a fragmented landscape of regulations or any kind of supervisory guidance that could actually complicate adoption. And that's very well aligned with our perspective as well. We want to help technology accelerate innovation, not stifle it. And that's a really delicate line I

Heidi Patalano   (18:20):

Could you tell us about how as the regulator of Fannie Mae, Freddie Mac, the FHFA is trying to get the mortgage industry to advance, to meet the expectations of the average consumer today who expects a very easy transaction, they expect everything to be as easy as ordering something on Amazon. How is the FHAA helping Fannie and Freddie in terms of getting the mortgage origination process to that point?

Leah Price (18:52):

Yeah, and you talk about the buy it now, click here, buy it now, Amazon consumer, but now think of the consumers that are just used to having a conversational experience with chat GPT or whatever interface. That's a very new trend that I think we'll see more and more of because you asked how are we trying to get the mortgage industry to advance? So the tech sprint is a big way that we do that. Besides that, we talked to big tech companies, we talked to Fannie and Freddie about how they're using gen ai and then we're also, I think the phrase is, eat your own dog food. We're also working on our own internal AI journey at FHFA along with the rest of the federal government. So we're learning just as everyone else is learning. So in terms of what as a regulator, what have we already done?

(19:52):

We already issued an advisory bulletin about AI machine learning in 2022, so that was a while ago. Everything that's in there still holds and is relevant to new gen AI developments. That bulletin, we highlighted the benefits, we highlighted the risks, and we encouraged in the bulletin Fannie and Freddie to implement a flexible risk based approach. And that's relevant if it's gen AI or not. So our intent is to support the leveraging of technology and data to promote efficiency and cost savings and the mortgage process not only for Fannie and Freddie, but for lenders and ultimately consumers. And that means homeowners and renters also.

Heidi Patalano   (20:44):

Yeah. Well, a question I had about the tech sprint, I remember hearing about it at the digital mortgage conference where we had you on to talk about some of the results of that, which was great. I think one of the most interesting results that you had mentioned was in the multifamily space in terms of innovation there, the ideas that came out of a sprint about multifamily, but I wanted to ask, what's the go forward after a tech sprint like that? So for example, does one of the team members in one of those sprints then take it and develop it a step further or what happens after you? They've had all these great ideas, so what happens next?

Leah Price (21:25):

Okay, so my dream would be that the followers of this podcast go watch the demo day video. Somebody is like, wow, idea number three was amazing. I'm going to go build that. And that transforms the mortgage industry collapses, costs better, manages risks. So we would love nothing more for industry to take some of those ideas and build them out. A lot of them, because we had so many lenders at the table, their primary market, many of them are primary market, so Fannie Freddie might not be appropriate to take those on, but anyone should feel free to take those, go gangbusters and tell us about it. We would love to hear the best way to reach us is email fintech@fhfa.gov.

Heidi Patalano   (22:12):

Yeah. Yeah, I love it. I love it. Okay. Well, and then speaking of that, the blue sky thinking, where do you think there's room for innovation where you're not seeing much of it yet?

Leah Price (22:23):

Yeah, I love the blue sky question because a big part of my job on my team is to keep tabs on what is at the horizon. Where are we? Not yet, but what's ahead and how do you think about risk of that big tech? They are on this very aggressive route to develop the most cutting edge technology that there is. One of the things on the horizon is sophisticated AI agents, so sometimes it's called agentic ai. Basically the idea there is you would have AI that is able to act somewhat autonomously. And a good example is I could tell my Siri, I want to go on a trip to the south of France and she can just go off and pull together my whole itinerary and she knows my budget, she knows where I like to sit on an airplane. She could just go do it and then come back and let me know what my plans are.

(23:26):

We are so far from that being the reality right now, but that's where everyone's trying to go. And so there's really cool stuff up ahead. But the mortgage industry, we're still grappling with the chat bot thing and which is fine, we should be. And so our industry has lagged behind other industries and technology adoption before probably with good reason. But to me, we need to be thinking about those emerging use cases because if we don't even explore them and we don't consider that these could reduce a lot of cost and complexity, we'll be totally unprepared once those technologies have proliferated and become mainstream. So let's say AI agents are abound, and I am doing as a consumer, I'm taking care of all the stuff that I don't want to deal with all my shopping, all of my commuting is handled by AI agents, but my dinky mortgage process, I'm still, I got to call a guy and then I got to go show up and sign a bunch of papers. So at some point, the consumers are going to be used to having that. The mortgage industry is going to be dramatically unprepared for. We won't even have thought about what that looks like for us. And everyone is suddenly going to go rush to adopt the technology just to catch up without being at all prepared to implement sufficient controls.

Heidi Patalano   (25:03):

Well, that's so interesting because that makes me think of Rocket Money and a lot of these tools that are like, we're going to manage all of your finances and we're going to tell you exactly where you need to do this and that, and that's their way in as a feeder for getting these people to become mortgage clients or refinance. And I know it makes me also think of Blend and the CEO Nima Ari was talking about how he wants this to be kind something that's applied to every kind of aspect of finance in your life. It seems like that could be the kernel or the foundation of something like this where it becomes like we are everything financial app. And then from there, maybe that is something that can be applied in that way.

Leah Price (25:57):

There are,

Heidi Patalano   (25:59):

Oh, sorry. I was just going to say, I love that I learned this new word today, AG agentic ai, and I'm going to start throwing that into conversations all the time.

Leah Price (26:08):

Totally. Definitely. You should. There's a lot being published, open AI is blatant that that's their goal. I read a study that's probably out of date now. It was published a couple months ago. It was comparing AI agents used for engineering tools, and the best one was successful 13% of the time. So imagine you're going on a trip and you're like, I think my trip will be 13%, not a total disaster. That's not good. It needs to be like, I mean, I don't know. I don't want to go on a trip unless I know 90% is going to be okay.

Heidi Patalano   (26:50):

So we've got a ways to go. But to your earlier point, the speed at which all of these things are improving and getting better, I mean, who knows how soon we could finally get to that 90%.

Leah Price (27:01):

Exactly. And so my point is there's an impulse, and maybe it's just what people say when they talk to FHFA, they are putting forward this part of themselves that's all about risk management and oh, we would never do this. We would never do that. And they think that's the right attitude to convey to us. But the problem with that is that if you don't think about the possibilities, you won't know what the downsides are. So bearing your head in the sand and being like, oh, we're just not going to do anything. It's too scary. That's not going to work.

Heidi Patalano   (27:34):

Well, Leah, on that note, I'll close it out and say thank you so much for your time today. Thanks for sharing. This was so helpful and instructive to me, and I'm sure the audience as well. So thank you for your time. Thank you so much, Heidi.