- 21 minutes ago
- #aimortgage
- #mortgagetech
- #housingfinance
For mortgage executives, the challenge with AI is no longer whether to invest, but how to scale responsibly and effectively. Join Chris McEntee of ICE Mortgage Technology and Tela Gallagher Mathias of Phoenix Burst for a strategic discussion on ICE’s AI execution and the broader state of AI adoption across the industry. The webinar will highlight where lenders and servicers are realizing value, the risks that often emerge during execution and the strategic decisions that shape long-term success.
What you’ll learn:
Mortgage leaders are past the question of whether AI matters. The real challenge now is how to deploy it in ways that scale, manage risk and deliver measurable value. This webinar offers a grounded, strategic look at what that takes. =
In partnership with: ICE Mortgage Technology
#AIMortgage #MortgageTech #HousingFinance
What you’ll learn:
Mortgage leaders are past the question of whether AI matters. The real challenge now is how to deploy it in ways that scale, manage risk and deliver measurable value. This webinar offers a grounded, strategic look at what that takes. =
In partnership with: ICE Mortgage Technology
#AIMortgage #MortgageTech #HousingFinance
Category
📚
LearningTranscript
00:00Welcome, everyone. We are about to get started with today's webinar, From Vision to Value,
00:07How AI is Taking Shape in Mortgage. I'm Alison LaForgia, the managing editor of HousingWire's
00:14content studio, and today's webinar is featuring two fantastic experts. We have Chris McEntee,
00:23the VP of Product and Corporate Development at ICE. And we have Tila Mathias, who is the Chief
00:30Technology Officer at the Phoenix team. And I'm going to give all of you a moment to get settled
00:37before we jump right into today's content and give you a few housekeeping notes. This is meant to be
00:42an interactive webinar. We want to hear from you. If you have questions for Chris and Tila,
00:46we will be hosting a Q&A at the end of the webinar. Submit your questions at any point in
00:52time. We'll
00:52get through as many as we can, time permitting. And yes, today's webinar is being recorded.
00:58You will get the recording early next week. If you registered, you will get an update in your inbox.
01:03If there's anything you want to forward to someone or any points you want to watch again,
01:07you will be able to do so in the on-demand version. Now, we have a pretty significant list
01:14of topics we want to get through today. So I'm going to jump right into where lenders and servicing
01:21or realizing value, the risks that often emerge during the execution, and the strategic decisions
01:27that shape long-term success. So let's start with why we think AI has reached an inflection point
01:37in mortgage. Tila, I'm going to toss this one to you. Tila, what a meaty topic. Thanks for having us.
01:45So at least on the Silicon Valley front, so I just got back from Nvidia GTC, which is kind of
01:49the
01:50Super Bowl of AI from a Silicon Valley perspective. And what you hear on the circuit is kind of this
01:56progression. So 22, 23, we had chat GPT, came out November 22, the world went bonkers, right? And that
02:03was AI that could generate. Then we look into the 24 timeframe, and we saw AI that could reason,
02:09right, that could sort of think about a problem. And then we had the progression in 25 to AI that
02:14could actually do work, right, with the agents and agent mania. And then now what we're seeing,
02:19at least from Silicon Valley perspective, is we're seeing, you know, sort of autonomous agent
02:24communities or ecosystems that can self-improve. And that is what the inflection point, as you called it,
02:32is about, at least in that kind of broader ecosystem. I don't know, what do you think about that,
02:38Chris? I'm a little bit of a contrarian. I think generally your point, Teela, is well taken in
02:44the sense that Silicon Valley, as usual, tip of the spear and a lot of experimentation,
02:49the ability for that kind of greenfield activity to occur around startups, well-funded background,
02:56certainly technology infrastructure that's there. I think mortgages, most of us who are on the call
03:02probably recognize, we're always bringing along a lot of legacy infrastructure. And the question on
03:08that is to how do we retool? And then we have business processes as well and compliance requirements.
03:14Those tend to put a little bit of speed bumps on it. Now, I certainly recognize a lot of startups
03:19with AI capabilities coming into mortgage because they see this really super complex, in some ways,
03:25maybe even convoluted process that can be automated. And I think that's the spot where we start,
03:32which is when you're talking about AI, do we start with automation? What are you really trying to
03:37accomplish in terms of less touches of a loan, higher loan quality and things of that nature?
03:42So I think in the continuum and inflection point, I'd say some of the early adopters might be hitting
03:48some of those phases that you described, Teela, where the ability to be able to have set the stage
03:53properly, get enough of their team and staff trained and capable in the new tools are starting
04:00to see some of those improvements. But I'd say it's hard to say as the industry at large, certainly
04:05in comparison to Silicon Valley is kind of at that inflection point, certainly driving in that
04:10direction. But that's been my observation.
04:13I do think one of the things that I'm seeing that I haven't seen in prior years is really a
04:23very
04:24intense awareness of the need for education. You know, from the very beginning in the November
04:3222 moment, 23 timeframe, there's always been a disconnect around, well, what is AI? What is
04:38generative AI? What is reasoning AI? What is agentic AI? And it starts to go like this,
04:43right? Wah, wah, wah, wah, wah. And we call that phenomena semantic satiation, right? You say this
04:48word over and over again, it loses all its meaning. So I am seeing a lot of interest in education
04:54and
04:55getting everybody on the same page. And that is kind of a tipping point that I've seen in the past
05:00little while.
05:02Yeah, that's really a good point. And I'll prompt you a little bit, Teela, because you and I have
05:06talked about some of these things. And I think there's almost, and I don't want to get too
05:11esoteric, but more of an existential quality to being able to say, I want to reinvent this process
05:18myself, how I interact with customers. When you're talking with clients, how are you seeing them adapt
05:25to that? Because some people react to a threat, and maybe people perceive this as a threat rather
05:31than as a benefit. And it's like, how do I react to it? And some are like, oh, I go
05:36all in. Others
05:37are, I'm going to observe. You know, this is, again, from our client base where we're seeing various
05:42people, many, many customers, again, looking to us given our deployment of different products. But
05:47what are you seeing?
05:49Yeah, I see kind of every part of the spectrum. There does continue to be a lot of fear, especially
05:57because the pace of change is unthrottled, not in mortgage. In mortgage, we do have a natural
06:04throttle. And what we're actually seeing, and we saw in 25, and a lot of the studies proved it out,
06:08is that it isn't the technology that's setting the pace. It's the people that are setting the pace.
06:13And the human beings involved tend to be a natural throttle, as well as some of those
06:18technological challenges and heritage ecosystems that you were talking about early. They are also
06:23a natural throttle. I see a lot of fear. I continue to see that. I see a lot of questions.
06:30What does this
06:31even mean? Like, am I even smart enough for this, is sometimes what I hear. And I think there's a
06:37little bit of a misconception that this is just a, okay, this is a run-of-the-mill change management
06:42activity. You know, we use the Cotter framework, you know, and create some change champions, and
06:47celebrate some quick wins, and we'll be off to the races. And really, it's much more akin to a
06:53grieving process, or like the, you know, as strange as it sounds, the Cuba Ross grief framework, right?
07:00And we go through all these stages and all these feelings, because this really is like a,
07:06it's a loss of what was, whereas it's a gain of what will be. And that's a difficult, it's very
07:13existential. Just, you know, for me on a personal note, you know, I've made my, I've been in this
07:18business for 25 plus years, and I learned how to do stuff a particular way, and I had all this
07:24experience. Well, now my experience is cheap, isn't it? Intelligence is cheap. So what matters when
07:30intelligence is cheap, when digital is cheap and AI stuff is everywhere. But we can't underestimate
07:35the degree of impact on the humans involved and how existential those feelings are.
07:39Yeah, that's a really good point. But, you know, wisdom never goes out of vogue, right? And so I
07:44think in that respect, right, intelligence is there in the machine. But I think from the, and when you
07:51talk about wisdom, and not to be, you know, highfalutin, but I think it's more of how precise do we
07:56think
07:56about what we can, you know, how, how critical are we about what we really need to get into,
08:03into the effectiveness of the tooling. And that's where I think, you know, now we're, you know, maybe
08:09we just do a shift down into what, where I, the way I deal with anxiety when I have it
08:15is I get, I go
08:16roll up my sleeves, and I get into it, and I start learning and learn by doing. And rather than
08:22sit
08:22there and say, I'm going to hit some point where I've absorbed enough information, it's going to click in
08:26my head. So a lot of that tinkering, which is amazing, I'm seeing across the whole enterprise,
08:33we've, we've got an explosion of people going, I need more credits, I need this, I need that
08:38tooling, I'd like to get that. So what's been really interesting, and this is just ice as a
08:44leader, I mean, ice has been a leader in machine learning and different, different permutations of
08:50call it, you know, artificial intelligence. And we all know that there's, you know, neural networks
08:55and a lot of things that have been developed. And this goes back even to the 70s. And I think,
09:00interestingly enough, the applications almost always seem to come to trading,
09:05right, because that trading mechanism, certainly at the New York Stock Exchange,
09:09is always the bleeding edge of technology, you think of fiber optics, you think of algorithmic
09:15trading, you think of all those things, you know, the bots fighting against each other in terms of
09:19pricing a particular security. So based on that heritage, it's been really interesting to see,
09:26one, us as an enterprise company, see champions emerge in business product lines, we do have an
09:33overarching kind of governance model, which is really dynamic, and is trying to help people go,
09:39okay, I want to raise a use case, how's the use case treated? Where's the data? What's the data
09:44governance? But to get back to the, how does a large enterprise, you know, more than 10,000 people
09:50really start to, to be more dynamic. And a lot of that, what we've done is be able to drive
09:56it
09:56down to the product level. So I have a number of product leads who, who I'm responsible for.
10:02And what I said to them is, when I've come into a role, is I've said, show me your roadmap,
10:06and they'll show it to me. And they'll say, you know, we're at, you know, we need, you know,
10:11excess capacity to be able to do this R&D work. And I said, no, let's pull that out. So
10:16we've
10:16actually looked at some things where we say, can we can, you know, there's features, but is there
10:22the feature selection, can we create some capacity where we can experiment? So a good example we'll
10:27have is we've been doing like everyone looking at the automated chatbot engagement from a consumer
10:32perspective to try and help LOs. That's something that we had prototyped maybe six months ago, and
10:39already the iterations are coming in very rapidly. So it's quickly improving. But then it gets a little
10:45bit of a plateau level, right? And what I mean by that is the functional operation of the tooling
10:50and the engagement with the consumer is good. But then we have to go, well, we don't want to leave
10:55the LO user assistant admin encompass, if it's an origination behind, we need to figure out how they're
11:03going to put that into their workflow, right? And how are they going to engage with the technology?
11:07So I think that's a little bit where you can prototype kind of in the Petri dish,
11:12but then you have to bring it out to the, you know, ultimately our customer and say,
11:16how do you want to interact with this day in day out? Is it going to be a time saver
11:20for you? Is it
11:21going to be a scale function for you? Do you still want the human touch? So I think that's the
11:26last
11:26benefit we see of this that, okay, in theory, we're going to be able to free the person up to
11:32develop the
11:32relationships. They're going to help them build their book of business, have repeat customers
11:36and engagement. And so it's really interesting to see how it's going to embed itself into, you know,
11:43the day-to-day operations and workflows. Well, let's dig in there. Let's talk about where
11:48lenders are actually deploying AI today.
11:58I'll start and then I think we'll have an interesting dichotomy of perspectives. So
12:06I'll talk in two ways. One, I'll talk about what I would call kind of your point solution, your use
12:13case-based approaches. And then I want to contrast that with some of the really mature agentic
12:19solutions. Agentic just being AI that can autonomously or semi-autonomously pursue an
12:27objective, right? So instead of I ask a question, I get an answer, you know, go and give a machine,
12:32just like you might give a really smart intern, go take this job and go do it for me. Go
12:37meet this
12:38objective for me. So we see routine adoption of code generation. If we're not doing code generation,
12:43for those of you out there that are builders, you know, you're definitely behind. We're seeing a lot
12:48of mature use cases in the call center, call summarization, sentiment analysis, voice agents
12:54are, you know, in the next year going to become, I think, very, very table stakes. That technology is
13:01really good. We're also seeing, of course, a proliferation of knowledge bots, right? Engaging
13:07conversationally with bodies of knowledge and then intake support and quality control.
13:11So if it can come in and be, you know, the sentiment or the idea behind it can be
13:18figured out and then it needs to, say, be routed somewhere or simple task-based automation,
13:23you know, we're seeing that as well. From an agentic or holistic enterprise perspective,
13:28the things that are really mature are development, end-to-end. Call center is starting to see end-to-end
13:37agentic networks. And anywhere where you have a well-documented, mature set of KPIs,
13:44those are really ripe to be implemented. The thing we're also seeing that I'll close with,
13:49and then I'd love to hear from you, Chris, is quality control. Large language models,
13:55vision language models are just excellent at, you know, helping us perform quality reviews. This is
14:04also an area where if we look at the future of mortgage in, say, two to three to four years,
14:09you know, we're in a very reactive defense type of a framework, right? First line, second line,
14:14third line, fourth line. You know, what if we could look at 100 percent of everything all the time in
14:19line? You know, what does that do? What if we move to proactive offense rather than reactive defense?
14:25So I don't know what you're seeing, Chris. That's a really good observation. So I'll tick off on
14:31some of those because I think they're all really good observations. I think all the areas are the
14:37ones that you talk about. So, you know, maybe to kind of give a little more dimensionality to it,
14:42you know, is it a point solution, as you mentioned, where somebody's coming in and say, look, I have a
14:46task. I could experiment, say, if I'm an IMB and I could get a vendor even who can help me
14:52kind of code
14:52some of this. And then they'll go, okay, I can get some benefits here. But if I'm seeing benefits across
14:58my workflow, how are they going to play nice together? Then you have an integration problem,
15:02right? So you're kind of creating a different challenge that you'd encounter. I think what's
15:08been really interesting is the call summarizations, sentiment, those things are kind of, again, to your
15:14point, table stakes. But then you start to have, well, how do people grade things, right? Because
15:19some people will say, I think that sentiment is this, I think. And that comes to model training.
15:24So then you have this skill that some lenders might not have innately, right? Because then they're
15:29saying, I want the output and the outcome and the business results from the tool. But I don't want
15:34the ongoing upkeep, maintenance and investment that's required. Because this is such a dynamic
15:40and moving target. So I think you have those preliminary, how deep into the tech do I want to
15:46be fast forward, right? Do I want to be a tech enabled lender? Do I want to be a tech
15:50first lender?
15:51You know, those type of decisions that are really challenging. Because they all come with cost,
15:56right? None of this is free. So you're going to go, look, I'm going to be fighting for good quality
16:00talent. I'm going to be obviously paying licensing fees to the extent that that becomes part of my
16:05cost of goods sold. What has been really interesting observation for me, I've been in the mortgage
16:10industry 25 years plus, is it came at a time where we're suppressed in volume. So think about it.
16:18A number of years ago, we were north of 10 million loan production. Now we're down to five,
16:23right? The actual, you know, kind of equilibrium might be seven and a half to eight million units,
16:29right, coming through the infrastructure. And so you've seen over the past year, a lot of
16:34consolidation. So that consolidation is also requiring people to do more with less, right?
16:39And then you have the large IMBs and depositories now potentially coming in with some of the capital
16:45changes wanting to get those scale economics, right? That they might see in say trading or
16:50different business mod lines that they have. And they keep on saying, well, God, mortgage,
16:56I've just, I've never been able to get to that profitability level that I want.
17:00Right. And why is that the case? Well, certainly a lot of expenses on commission.
17:04And so there's, you know, if you look at the biggest expense there,
17:08I think that's back to the fear factor, right? Where people are going to say,
17:11I need LOs. I'm not going to create an agent. Would it be possible for me to be able to
17:20create
17:20a model where I license? And you and I have talked about this, Teal. I'll give it back to you.
17:24Would
17:25you license an agent? What would that look like? Yeah. Well, you know what I'm going to say here,
17:30right? SaaS is dead. And, you know, long live outcome-based pricing. Now I say that a little
17:38bit tongue in cheek. I think we're going to see a long tail on the transition from SaaS-based pricing
17:44to outcome-based pricing. And we're already starting to see it in mortgage. We're seeing professional
17:49services and software companies coming into mortgage that are saying, hey, you know, pay,
17:53I'm going to pay this upfront. And then I'm going to take a, I'm going to take a percentage of
17:58the value
17:58that is actually realized. And so that's going to be a trend that we're going to see. The reality
18:08of the technology ecosystem in mortgage is incredibly complex and it's based on a lot of
18:13heritage technology that has, you know, sprouted arms and legs and tails and tentacles post 2008,
18:22right? The ecosystem was never designed for the regulatory pressure
18:26that came in 2008 and the technology ecosystem got weighed down then with a lot of ancillary
18:32solutions, core solutions. And that takes a long time to untangle. And so I'll be honest,
18:40I forget what the question was, but did I answer it? No, I think you were going there. I think
18:45what
18:45the main point being that the macro environment, as it informs investment, I think that's what you're
18:51getting at is I know I'm going to have to invest something, but is it a cost of goods sold
18:56or is
18:57it an outcome based business model? As I'll quote Mark Twain, you know, the news of my death is
19:02greatly exaggerated. I was somebody defending the SAS model, I will say, and the irrationality
19:09associated with it and the presumption there though, in some cases is there's no value to the core
19:15infrastructure. But, but, but, but, but, you know, not, not to be defensive, but I think this is more
19:20evolutionary because if you think about this, right? So yeah, we're, we're creating, we're creating these
19:26machines that evolve at a different pace than humans do. And so there's going to be this rapid
19:32acceleration and, you know, not, not say go dystopic, but I think more of how can humans adapt and use
19:39the technology. And I think, you know, going back to the audience a little bit, you know, news you can
19:44use,
19:45you really have to ask precise strategy questions. Okay. It's easy to chase the next new technology.
19:50I've been around the block. You know, it was cloud computing. It was blockchain. It was now it's
19:56tokenization. And I'd even argue tokenization is really more of a legal market structure construct,
20:02less, less a, a technology, but certainly the idea of 24 seven trading. I look at it at our broader
20:09business in large part. And you said it earlier to you meet the customer where they are and they're on
20:14a continuum, but the best customers and the ones we're seeing who are most successful in crossing
20:20this bridge have been really precise around. That's where I want to take costs out. I want to
20:27take costs out. That's where I'm willing to take some risk in terms of scaling. There's a spot where
20:32I'm not a domain expert. I want you to take care of it. And having those conversations with your core
20:38vendors, your core, you know, implementers, to the extent that you have a third party that's doing that
20:43is, is really the precursor step. I, I, and I think, but I think it runs parallel. I think it
20:48goes,
20:49get your people up to speed to understand what the technology is capable of, but then also imagine
20:54what that reinvention looks like. What is that reinvention? Where do you want to be as a lender
20:58two, three years down the road? Uh, do you want to carve out a niche? Uh, and you know, we
21:03see that,
21:03or do you want to be agency, but just scaling, you know? Yeah. And, and we do have to meet
21:07the,
21:08the, we have to meet clients and customers where they are as, as we've talked about,
21:11there continues to be a place for SAS. You know, I have a commercial product that,
21:15um, that we do license in a SAS model. So, you know, it's a little bit tongue in cheek in
21:20terms of,
21:21well, SAS is dead. So for example, uh, I tinker all the time, uh, and I built myself a handy
21:26dandy
21:27little Salesforce, um, over a weekend. Uh, and it's perfect for me. Uh, you know, I got 140 people.
21:33Um, I've got, you know, a pipeline that's, you know, maybe 150 things that are in it and it works
21:39way
21:39better and way cheaper than Salesforce. Now, if I wanted to try to deploy that at scale
21:43in an enterprise as complex as the mortgage ecosystem, uh, you know, what would happen?
21:48Very little. That's what would happen. Uh, and when I do take it into my product team,
21:53for example, and I, you know, and I, you know, and I go, Hey guys, I was in the lab
21:58this weekend.
21:58Look at what I built. How come we're not done yet? You know, how come this isn't finished?
22:02If I can do it in a weekend, why can't you, right? Well, there's security,
22:05there's enterprise scale. Um, there's regulatory compliance, there's guardrails, there's evals.
22:10There is, you know, the very real question of, of security, right? Securing data, securing the
22:15ecosystem, securing the application. So you're absolutely right. It is going to be a gradual
22:19transition. There is a place, um, there continues to be a place for all the models, um, today,
22:24but what we do see. And again, when I, when I go out into Silicon Valley, like it's like stepping
22:29into the future. Right. Um, and we absolutely see, um, uh, AI native companies, um, are, you know,
22:40eating everybody's lunch. Um, and the, that, and I, I wrote something about this, you know, earlier
22:46this week, you know, when I, when I came out of GTC, honestly, I was kind of depressed. Um, every,
22:51you know, every software provider, service provider, product provider, um, which by the way,
22:57that's everyone needs to be really concerned, um, about the, the pace of change and the way that
23:04uber large consultancies, AI native providers and foundation model providers, um, are coming for
23:11everything. Um, and it's, um, it's a good wake up call for all of us to make sure that we're
23:17focused
23:17on what does the world look like and how do I fit in that world in a year's time and
23:21two years time and
23:22three years time. It's a very existential question that goes into the fear,
23:27but if we can take that fear and use it as a catalyst for action, if we can use that,
23:30take that fear and say, you know what, I'm going to eat the bear. I'm not going to let that
23:35bear eat
23:35me. I'm eating the bear. I'm going to decide what the future is going to look like. And maybe I'm
23:39right. Maybe I'm wrong, but at least I will have had a good time in the process. Yeah. Well,
23:44and you'll learn a lot, right? And I think I'll learn a lot. Oh my God. It's crazy how much
23:48we learn.
23:49You'll, you'll learn, but extend the, the, the, the bear analogy. You know,
23:53the joke is always, why do you bring bear spray when you go to Alaska? Cause then you can squirt
23:58it in your eyes right before you get mauled. Um, and so, so, you know, I, I think, I think,
24:04but, but your point is really well taken as to be, um, fearless in some respects and be on the
24:10offensive because I think as any of these things know, they're going to evolve in directions that
24:16we don't really know. Okay. So, so we go back to core principles, right? What are the core principles?
24:23We're going to organize differently around this. Um, the human in the loop, as we've all talked about
24:28this, the deterministic logic of this is, I mean, ultimately digress for a moment. Ultimately,
24:34what's alone alone is the decision to lend money. That's what it is. It's somebody deciding I want to
24:39take this risk. So everything that leads up to it, good information, gathering, uh, calculations,
24:46and then, and then decisioning, I think we can look at the front end and go, a lot of those
24:51activities
24:53could, could obviously benefit from being artificial, you know, driven by artificial intelligence.
24:58But at some point there is a obligation responsibility that is, you know, legally and
25:03codified and, you know, how quickly do we want to go in there? So I use an example. We were
25:08talking
25:09about it when we were in the green room was just the idea of does, does somebody want to purchase
25:13loan approval in three days? I don't know if it's your first time home buying in three days and saying,
25:20hey, you're, we're, we're going to lend you three quarters of a million dollars, half a million,
25:24and what average home price, half a million dollars. That would be pretty daunting, right?
25:29So I think that there's some embedded, not delays, but checks and balances in the legacy workflow.
25:35So like, again, what's your appraisal come back, but we all know some of these things,
25:39given the tooling can be instant. And if they're instant, it's, are you ready for that instant
25:44decision? So I think there's a lot of expectations. Maybe that's where the benefit of the LO comes in.
25:49I'm coaching people. I'm telling them how they work on their budget, how they can, you know,
25:53fulfill their obligations. What are the opportunities in the event they do lose jobs or, you know, all the,
25:58you know, can't play, can't, you know, can't pay, can't stay. So in theory, the ability to be able to
26:03accelerate
26:04the menial task and work with the client to try and get them a customer prospect to understand the,
26:11the complexity of the, of really the loan obligation, the legal obligation might be a benefit. And again,
26:17that comes to an outcome. The outcomes, let's get back to more of the specifics, the scalability, right? I have
26:23a LO,
26:25and that LO, I want to do 10, 15 loans, you know, rather than two or three. Underwriting is a
26:31good
26:31example. Three loan files, one and a half a day. Can I get five, six? Those are the KPIs that
26:36seem to
26:37be driving some of the experimentation. Last thing I'd say, which you brought up Tila, which is really
26:41interesting. I mean, think about a hundred percent loan review, right? Most of the problems that came out
26:46of Dodd-Frank securitization. I mean, it's literally in the law of the load samples. I'm going to take 10
26:51% of
26:52the loans. And presumptively, I got the 10% that are right. They're going to be reflective of the population.
26:58Now you can have almost instant QC and have things kicking out of securities. In theory, and again,
27:06whether we've seen this in practice, that should result in a pay up. But when you go to the investors
27:11and go,
27:12I'm going to pay up because this is higher quality. Our experience has always been that that's what you're obligated
27:17to
27:17do already. So it's going to be a little bit of a challenge to kind of wedge those use cases.
27:23Like, hey, we did
27:24100% loan file review. Well, yeah, but I expect I'm going to get clean loan files already. So now
27:29it goes back to
27:30how do I get the infrastructure to be able to review those files? The files that are accepted go to
27:37the expert who can
27:37make those determinations. So it's a lot of re-engineering.
27:42Yeah. And I want you to think of this one last thing. So when something that was previously
27:50unimaginable is able to be done, that changes everything, right? So let's think about like,
27:57air travel or train, right? Before the train, it would have been a terrible idea. Nobody would ever go
28:04from Washington, D.C. to say, Westlake Village, California. You wouldn't do it. It would have
28:09been unimaginably difficult. And then we had the train, and then we had the air travel. And all of
28:14a sudden, these previously unimaginable things were able to be done. And that changed everything.
28:20We now have technology where what would have been previously unimaginable, 100% review in line,
28:28every decision, every time, that would have been an insane concept. Now, it's like, oh my god,
28:33this previously insane thing, I can actually do it. And that's like the mushroom cloud that is happening.
28:40It's like, wow, well, what else that I couldn't do before can I now do? And that moves us away
28:49from
28:51a conversation about ROI heads and dollars to like, wow, what could the future be like?
28:58Yeah. No, look, I think it's great. And ultimately, it's about the imagination. So I think that's the
29:04other thing, the creative aspects. And what I mean by that is more importantly, do the work both with
29:12your organization. I presume most of the people on here are in some leadership role or they're driving
29:17technology. I think there's an element of empathy to go back to it, understand how people are anxious
29:23about what it means for them or long term for their industry, certainly for their customers
29:27and their success. So I know that I think we're pivoting over, Alison, you might be,
29:33I heard a few chirps. So it sounds like we've got some Q&A.
29:37We have some audience questions. I do want to talk a little bit as I parse through the audience
29:44questions about some of those misconceptions and get both of you to touch on them as I review those.
29:50Because I think it's really important because I love that you both just mentioned about being
29:57able to be creative and imagining how you can very intentionally change things that previously
30:04might not have been possible and how that can really redefine some experiential pieces as you go
30:11to manufacture these loans. But I do think going back to both of you mentioning some earlier fear,
30:17that that fear and perhaps those misconceptions do persist.
30:21Oh my gosh. And I love this question and I love this idea, right? So they're one of the
30:26most common misconceptions is AI is coming from my job. Okay, look, the reality is there are going
30:31to be jobs lost. Every major technological revolution in the history of technological
30:36revolutions has changed the job. If you look, there's a study recently, fully 70% of the jobs that are
30:45currently recognized by the United States Census didn't exist prior to the 1940s and they wouldn't
30:49have even made sense. So that's just reality. The thing that makes a human so unique, right, is our
30:56ability to aspire, our ability to dream, to have ambitions, to set a vision and have people go around
31:05that vision. And that unique humanity is even more precious and beautiful today because of its scarcity.
31:14You know, and that is something that I find every day to be very inspiring is that beautiful human
31:22experience that makes humans unique. And what generative artificial intelligence has done is it has 10x,
31:2920x, 30x, those things that are innately human. Yeah. And I think, look, it's a new frontier. I mean,
31:39that would be my perspective on it. And when you're entering those domains, what will keep you alive?
31:46You know, alertness, curiosity, some degree of fear, but not that it makes you in mobile. And then
31:54community, right? At the end of the day, I mean, people, people, there's a reason communities get
31:59created is because there's a foe out there and whether it's another tribe or whether it's another
32:04country. In some cases, this is something that is not, you know, it may be created by humans, but
32:10that's where I think it's really interesting to see. Silicon Valley, as usual, has and, you know,
32:18has a view. I don't think it's shared by the general populace. I think that we've all kind
32:24of seen this concentration of both wealth and decisioning among a bunch of tools. I think you
32:30could have this opportunity for it to be more democratic, certainly. What I mean by that is
32:35being able to have things that were my tooling and even my personal decisions, I can, you know,
32:41asset management and things that I want to do, I can bring into my life with relative ease and
32:45customize. That's the other piece, highly customized decisioning. So I think it's a
32:50really good world, but it also, you know, gets people anxious. But again, how do you get over
32:54it? You just learn. Yeah.
32:58Yeah. And I think some of the other misconceptions to your question,
33:02AI is good for everything. AI is not good for everything. If you need 100% accuracy 100% of
33:07the
33:07time and you need it to be 100% evidenceable, not a great, not a great, not a great solution.
33:13AI is easy. That's another common misconception, which goes with the next misconception, which is
33:18AI is too hard. So like, you've got them all across the map. Another one is there's a best
33:25model. I get this asked a lot, all right, what's the best model? What's the best model that I should
33:29use? And it's like, well, there's no one best model. There's the best model for the particular
33:34thing you're trying to do. We're seeing this incredibly rich pantheon of models available,
33:39small models, big models, open models, closed models, vision models, language models. I mean,
33:47there's like, there's literally hundreds and hundreds of thousands of models out there. And
33:51this, what we call mixture of experts is, is definitely, you know, the place to go. And then
33:56I think the last one I would say is, you know, everything fails and there's no ROI. There's all
34:02these studies, studies, and I'm going to put that in quotations. There are these online surveys that
34:08get a lot of airtime that say, oh, 95% of all AI initiatives fail, or, you know, 70%
34:14of CIOs say
34:15there's no ROI. You know, it is true to say that these things are difficult. And there are many wildly
34:23successful implementations. If you, if you stick with it and you really get there. So those are
34:31some of the misconceptions I see. Yeah. And I think, I think also to a
34:35last point, I guess, well, I would just say, everybody's a BA now. I'm laughing. I said that
34:40into a meeting with our group, you know, and I'll look at documentation and, and, you know,
34:45I'm not calling out anybody specifically, but I would say, God, the documentation I used to get
34:51from a particular employee was week. Now it's like, suddenly it's improved a lot. And I'm going,
34:56did you really understand what you were prompting? And so ultimately, right, the interrogatory,
35:03the asking the right questions is important. I mean, this is fundamentally what we're at,
35:07right? Well, what's a prompt, but a interrogatory, what is in the world? How is the world sequence?
35:14So I think, you know, challenging people to think about the art of the possible, but be practical.
35:20And I think that's where you're getting at with the failures. People have too much ambitions,
35:24just get there and mix it up. And I hear this from our customers all the time, which is where
35:29are you going? And we say, look, we're on this journey with you. We're going to talk about the
35:34areas where we're investing. I had a question come up from a customer. They said, Chris, I want to know
35:40where you're going. This was in a recent event. And I said, if you got a couple of days, because
35:45I have
35:46127 plus products within the surface area of ice mortgage technology, each one of those on multiple
35:53dimensions are being impacted. So it's a factor of 127 times N, right? The developers are changing.
36:00The QC team is changing. The program management is changing. So, you know, rather than I'm saying
36:08to the customer, hey, let me explain this all to you because it takes too much time. Tell me where
36:15you feel that you're going to get the biggest lift from us building this on your behalf. And what I
36:19mean, building on behalf, building it into the product. We didn't even talk about this around the
36:23economic issue. You brought it up a little bit, Tila, which is what's commercial. What is really
36:28commercial at the end of the day is the improvements on this. I know from a product that I'm responsible
36:32for
36:32all regs, which we have out in the marketplace, and it's got AI embedded in and generative wise.
36:38And it was very early stage development. I had people telling me, hey, I don't get the same results
36:44you get. I go, of course not, because I have 900,000 highly curated, reviewed by legal team
36:50and content management. We're training off of two different libraries. So, you know, you might have a view
36:56that that is better because I'm training on the internet or because I have different prompts or
37:01whatever that experimentation internally. How much do I want to say I'm trusting that we as a
37:09provider, service provider, are getting up every day to train that model and make it the best possible
37:15model for your business use case. So, I think those are the decision trees people have to focus on.
37:22All right. I'm going to pivot into audience questions. And I want to start with the first one,
37:28which is getting back to an earlier point, which is we all want to get a loan, evaluate,
37:36and then close. We have to consider it's still a very emotional transaction. How do you balance
37:42too much tech slash AI versus people?
37:49I think it goes back to, I'm going to jump in there just from the standpoint of,
37:56does the ability for technology to take, say an LO or LO assistance is chasing down a pay stub.
38:02Hey, Chris, we want to close your loan, but you need to document this.
38:06In theory, the agent would be having that dialogue with the consumer. The consumers still have to
38:12provide the data, but the question would be, to your point, how do you get that person emotional?
38:17I mean, if somebody was to say, I give you an instant approval and you can go have this obligation
38:23to borrow, again, use the number, half a million dollars. I don't know how, I think people need
38:28to sleep on that. I think getting an instant decision is emotional and they're in a reactionary
38:35mode. So, do I say we take what used to be 90 days and concentrate it down to 30? The
38:41bigger question
38:42is, with the time savings, what are we going to do at that time? Does it go to profit? Does
38:47it go to
38:47time available to coach the customer to put them in the best position as a borrower to be successful?
38:54There's policy objectives on this overlay. Does the lowering the cost create incremental
39:02affordable opportunities? There's so many questions there, but the main thing is,
39:06at the end of the day, we're still human. So, the users are going to be emotional, react to the
39:11extent that they do. The consumers are all your stakeholders, your investors. You know,
39:15you told me you're going to sell me A, I got B. Those are always challenging conversations,
39:22independent of the technology, always will be. Yeah. And what I would say is,
39:26there is absolutely unambiguously no substitute for authentic human connection. And as human beings,
39:33we seek authentic human connection. And that isn't going to change, in my opinion. And humans,
39:45engaging with humans is like the joy and the beauty of life. And it is, you know, one of the
39:50most difficult parts of the home buying experience is exactly getting over that fear. It is very emotional.
39:56I think there was a study, something like 50% of people cry during the home buying process.
40:01And so, we have to be there to help with that. So, in terms of the direct answer to the
40:06question,
40:07we lean into using technology to do all those things that can be automated easily, that are a big pain
40:17in
40:17our rear ends. And we, you know, we take friction out of the process and we really amplify that human
40:25connection component. Yeah.
40:29Tila, to your point, I read a study about the onus of the choice to actually purchase a house,
40:36which is also very emotional. And that there was an interestingly high percentage of millennials who
40:44are actually purchasing their house with a backyard for their dog. So, you're talking about emotional
40:49prompts from concept, from the conception of the idea through the actual clothes.
40:57Yeah. I mean, it's, this is like, this is the journey, right? This is like the joy and the sorrow
41:01of life. You know, you're like right in the thick of it. Biggest financial purchase of a lifetime in
41:07many cases. You know, I would like a human, I would like a human in the loop with me on
41:13that.
41:13Yeah. Yeah. And that's a great, great way to put a fork in it. At the end of the day,
41:19that's where
41:21humans come together, obviously, in this business to borrow money and to really create a home. So, that's
41:27a difference between a house. Can you buy a house with AI? Sure. Can you make it a home? Probably
41:32not.
41:33Well, I'm going to leave that answer there. I think we've buttoned that up pretty nicely with that ending
41:39statement. So, Chris, I'd be remiss if I did not ask you a question about ICE's near-term product
41:46deployment. Since I have you here and you've mentioned at a couple of points in time,
41:50different initiatives, what areas are you looking at near-term for AI product development over at ICE?
41:57So, across the entire workflow, and rather than just have like a medley to share with you, I will
42:03go off on a few points. All the things that we've talked about easier in this call. So, highly,
42:08highly structured, but certainly repetitive tasks. So, again, the call center, call center routing,
42:15intelligence around that. I just recently worked on a prototype that was internal. It was remarkable.
42:21I called and I used my own loan for a refi and the automated agent like nailed it. It was
42:29almost spooky.
42:30And that is something we're going to have out probably third quarter of next year,
42:34fourth quarter in our customer hands. So, that level of engagement. Again, the chatbots, a lot of
42:40these functionalities that are emerging are going to be table stakes and we'll have all those, right?
42:46All those capabilities. I think loan review, documentation recognition, you know, we talked
42:52about call center. And then I think also too, then you're going to have some execution side. So, you go
42:57back
42:57into the securitization side, being able to purchase bulk loans. I mean, those will always go through
43:02review, but talked about the 100% review side. That's more of an investor requirement, but overall,
43:09both an origination, I'd say at least two or three of the key areas, front level, customer acquisition,
43:15engagement, scalable, customer thoroughness of file. You know, do I need to do an inventory of file? We're
43:22in the products. So, having them native in the products is just part of us, our commitment to
43:28our customers. So, we're excited. But I think more importantly, it's almost when I have that
43:34conversation with the customer, I want to know from them what their priorities are. How are they
43:40evolving their model? Are they going to go to niche? Because even go so far as to say, look, I
43:45have,
43:45I might be, you know, having translations, Spanish language translations and real-time
43:51coaching that can happen now with this tooling and set that borrower up for success better. So,
43:57really, it's almost endless, the number of applications, but from a product and customer,
44:02they're going to be seeing these things coming at them pretty rapidly. And I think it's important
44:06that they get organized like we talked to earlier. All right. I have two rapid-fire questions for
44:13both of you guys as we get closer to the top of the hour. First one, as succinctly as possible.
44:21What does successful AI adoption look like in practice?
44:28Do you want to take that one first? Yeah. I mean, I'm seeing what I call green shoots around this.
44:35And what I'm seeing, and it's more descriptive, less a kind of, you know, call it like a model.
44:41We have all, like any organization, we have teams, you know, Slack channel, whatever. And the ability
44:48for people to come together and quickly share information that's actionable is amazing, right?
44:54Like what used to take, hey, let's go gather information, let's get a call, let's test the
44:59information. Now we're almost doing this on a real-time basis. Like, oh, I read the report,
45:04I see the recommendation, boom, let's keep on moving. So velocity, the speed at velocity,
45:10but also the confidence level where I know that the data is good. So really from a practical
45:14standpoint, when you're saying what I see success is higher velocity, which we're getting more
45:19confidence, and I'm seeing it. The ability to trust all of your colleagues, I know it sounds strange,
45:25but everybody's contributing a different perspective. And when this is run through the models,
45:30it kind of takes out maybe bias in some level. So that's been really fascinating from a decisioning
45:36and really kind of knowledge management side. So I think that's where I've been seeing success
45:41internally. Externally, what I've been able to see is people who do rapid prototyping,
45:46but then also don't forget the legacy infrastructure, human in the loop. So example,
45:51I mean, I've seen some people say, I'm going to push some of these things out over to
45:55review overnight and overseas. So I think those are some successful models we're seeing.
46:00You know, I would say as succinctly as possible, you will know, let's work backwards. You will know
46:07that your AI strategy is successful when it is just your strategy. Because AI strategy is just strategy.
46:15Technology strategy is just strategy. The other thing I would say is when you are not only realizing ROI now,
46:23but you are deeply confident that you will also be relevant in three years' time. That's what I would say.
46:30Yeah. That's a great, great description, Tila. I mean, I think that's table stakes and
46:34goes back to the existential point. And your strategy is to not only survive, but thrive. And these tools
46:41will help you do it. All right. And last question. How should people, or the leaders specifically,
46:52in mortgage, think about the next 12 to 24 months?
47:00I'll kick that one off. What I would love to see and how, and it's very, very hard,
47:08is, you know, what does a reimagined mortgage process that has been specifically designed from
47:15the very beginning to be accelerated, what does that look like? And I would really encourage
47:20the industry and I encourage my clients and anyone who will listen, you know, to,
47:26you know, our processes are not designed to go fast. And now we have this incredible magical
47:35tool that we already talked about. You can do these previously unimaginable things. So,
47:39you know, if you look at accelerated computing, what is accelerated computing? Accelerated computing is
47:43this idea that we can take a certain type of workload and offload it to a different type of technology,
47:48right? That's all accelerated computing is. Well, what if we had accelerated mortgage? What if we
47:52looked at mortgage as inherently native to be accelerated? What would be those workloads and
47:59parts of the process that we might offload to AI? And especially to that wonderful question we had
48:04earlier around just the emotional nature, you know, hot tip, that's not an aspect of an accelerated
48:11mortgage process where you don't have humans. That's not a good idea to accelerate. I don't
48:16want to, all right, I've got 32 seconds. Let me get with my homeowner and see if I can squeeze
48:21in,
48:21you know, like that's, that's, that doesn't make sense. So I think that's what we should be looking
48:25at in the next 12 to 24 months is how do I reimagine this process so that it is innately,
48:30inherently designed to be accelerated? And then what do we want to accelerate as a result?
48:35Yeah, it's a good question. I think, I think of it as almost parallel tracks, right? Most lenders,
48:44they have businesses to run, right? So you call, keep the lights on, RTB, run the business,
48:48whatever that looks like. But I think getting that willingness to go, what are we doing today
48:56that isn't adding value either internally or for our customers or for our ecosystem, which is what we're
49:00doing? Can I have the confidence that if I remove these things, I'm going to create a space, a vacuum,
49:07nature abhors a vacuum. What do I put in that vacuum? I'm going to put the experimentation, the
49:13collaboration, the rapid prototyping that we've all talked about, which is necessary to figure out
49:19what works for you as a lender. So I think that strategy is more driven as much about what am
49:24I
49:25doing with my time and where do I want to be? I mean, I reverse engineer practically everything in my
49:29life. I say, where do I want to be in three years? What are the steps I need to get
49:33there? And that's
49:34exactly what they have to be. And I think they have to say, what does the macro environment look like?
49:38I don't think we're going to bounce back to 10 million loans a year. I think the large INBs that
49:43are going very actively in the space are going to create more competition. I'm going to have to react to
49:48that. So I think it's just more of being able to be agile and leadership is about that every day,
49:54getting up and every day, you know, um, uh, being able to go, I might be afraid, but, um,
50:01fear is not going to, is not a success factor. It's the imagination and creativity. It's going to
50:06lead me. Yeah. Like, hang on guys. It's okay. Hang on. Yeah. I mean, I think about like the,
50:13the NASA, right. The guy, NASA, the guys who got, you know, strap themselves to a rocket and say,
50:18yeah, you know, I'm not really sure whether it's, you know, Chuck Yeager,
50:21I'm going to go break the speed of sound. I don't know the consequences. A lot of these people
50:25thought they were going to get atomized, but they didn't. And they broke new frontiers.
50:30Okay. So I have, I'm going to add one more rapid fire question before I let both of you go,
50:35because you both just touched on this beautifully. Chris, you just mentioned the people who are
50:41perhaps concerned about RTV running the business. What would you say is important right now to the
50:47people who are singularly focused on running the business? What is the benefit of working
50:53on your business while you're focused on working in your business? And why should we do that?
51:01That's a good, well, look, it's, it's, it's a contradiction or dichotomy paradox. The paradox,
51:06probably what it is, is that I know what I need to do, but I can't create the time to
51:10do that.
51:10And I ignore it at my own peril. So the key point is, what am I doing today that I
51:17can create space
51:18to do the things outside that? And that goes back to what Tila just said. It's strategy.
51:22You can call it technology strategy. You can call it business operation strategy, whatever it is.
51:27Ultimately, if you don't have a strategy, you know, in the famous saying, right, if you don't know where
51:32you're going, any road will take you there. I think having that precision, that is more than 50% of
51:38the
51:38work. Then all the technology and the decisions just thread in around it.
51:48What I would say is that life is a lot more exciting and interesting when we're learning
51:53things along the way. And how can, whatever my role is, I owe it to myself, to my families. We're,
52:01you know, we all have lives to live, families to feed, dogs to do whatever people do with dogs and
52:06plants to water. Right. And so I want to derive meaning out of what I do, whether I'm running
52:11the business or I'm designing the future business. It has never been easier than it is today to learn
52:17new things. You know, just ask ChatGPT, just ask Microsoft Copilot. And so I owe it to myself,
52:24to my kids, to my grandkids, to become that very best, most effective, most awesome version of myself.
52:31And that means learning all the new stuff, because it will help you. It helps me all the time. It
52:38helps
52:38me learn how to think. Even if the specific keys I punch, screens I look at, even if those things
52:44aren't changing, I can still learn to do those in a way that is inherently much more valuable to an
52:53organization. And that is something that all it takes is hard work and hustle.
53:01Well, I could talk AI all day, which is something that I stole to our audience from Tila earlier,
53:07but we are getting rapidly close to the top of the hour. So Chris, Tila, thank you both for being
53:14here.
53:14Thank you for giving your insights for our audience. To our audience, thank you for joining us today. If
53:20there's anything you want to rewatch, any points you want to share with people in your office,
53:26or people that you work with, you will be more than able to do so when we send you the
53:30on-demand
53:31recording in the next couple days. Thank you both again for being with us, and thank you to our
53:35audience for tuning in, and hopefully we'll see you all again shortly. Yeah, thank you, Alison.
Comments