JazzX AI was proud to partner with HousingWire for Rewiring Mortgage for the AI Era: Why AI Is Now a Leadership Imperative, an executive discussion on how mortgage lenders can use AI to address rising costs, declining productivity, and growing operational complexity. As AI moves from experimentation to enterprise adoption, industry leaders are rethinking how work gets done across the mortgage lifecycle to improve efficiency, reduce costs, and create sustainable competitive advantages.
Featuring Eric Hart, President & CEO of Pulte Financial Services and Siddhartha Agarwal, CEO of JazzX AI, and moderated by HousingWire’s Allison LaForgia, this webinar explores practical strategies for deploying AI within existing mortgage workflows, balancing automation with human oversight, establishing effective governance, and driving organizational change. Attendees will gain actionable insights into what separates the lenders successfully scaling AI today from those at risk of falling behind in the years ahead.
Read the transcript of AI in Mortgage: Rewiring Mortgage for the AI Era : Why AI Is Now a Leadership Imperative:
ALLISON LAFORGIA (HOUSINGWIRE):
Welcome, everyone. We are about to get started on today’s webinar. I’m Allison LaForgia, the managing editor of the content studio at HousingWire. And today’s webinar, “Rewiring Mortgage for the AI Era: Why AI Is Now a Leadership Imperative”, is produced in partnership with JazzX.
Now, today we are joined by two experts who are going to lead us through this conversation about AI implementation in mortgage. We have Siddhartha Agarwal, the Chief Executive Officer at JazzX, and we are also joined by Eric Hart, the President and CEO of Pulte Financial Services. Eric, Siddhartha – thank you for joining me today.
Now, I am excited to get started with today’s conversation. AI is at the top of mind for everyone in the housing industry, and I am excited to hear both of your perspectives on it. We’re going to start with you, Eric. You’ve said that the mortgage industry is facing a structural productivity and cost problem, not just a cyclical one. What do you think is fundamentally broken in the current operating model?
ERIC HART (PULTE):
Yes, so it’s a great question. I think probably most of the folks listening to this are familiar with how expensive cost of production in mortgage has gotten. The MBA just came out with their most recent quarterly statistics — I think $11,000 to make a mortgage.
And it’s actually higher in some segments. Depositories, for instance — there’s a really funny stat that it costs about $15,000 for depositories to make a mortgage, which is roughly the same cost that it is in the United States to have a baby when you factor in for somebody who has employer-paid health insurance. So I think one of those should cost more than the other. I’ll let you guess as to which one should — but there really is a structural cost challenge.
When you look at other areas of consumer finance — credit card acquisition, auto lending, personal lending — particularly the manufacturing of the actual loan or credit product costs has come down, mostly due to automation. Home lending has gone the other direction. There’s a phenomenon in economics called the Baumol Effect, also called the cost disease, and it talks about how in lower-productivity industries or segments where you don’t have technology driving innovation and higher productivity, you actually see wages rise because in order to attract talent from industries where productivity is strong and wages are rising, you just have to keep spending more and more even though the productivity doesn’t get any better.
The classic example that two Princeton economists used to articulate that was — think about the idea of a string quartet. It takes a string quartet 30 minutes to play a Beethoven quartet piece. You can’t make it faster. There are no efficiencies you can gain out of it. So over time, you have to keep spending more and more to hire those musicians because you can’t get any efficiencies out of it. And if I were to extend that analogy, think about the quartet in the mortgage space. It’s a loan officer, it’s a processor, it’s an originator, and it’s a closer — four people holding the bow on those instruments. And as long as it’s sort of a human-mediated process, it’s hard to drive significant cost and efficiency improvements.
So over time I think we’ve seen this space just effectively grow with the cost of labor even though you haven’t gotten any efficiencies. It doesn’t get better, it just keeps getting worse, those costs get passed on to consumers, and I think it’s one of the big drivers of why home affordability continues to be so challenged.
ALLISON LAFORGIA (HOUSINGWIRE):
Absolutely. Now I want to talk a little bit about why AI is becoming a strategic imperative in mortgage and not just an innovation topic. Siddhartha, let’s start with you.
SIDDHARTHA AGARWAL (JAZZX AI):
Yeah, it’s a wonderful question. And I love Eric’s analogy of musicians to the loan processing ecosystem — I would not have thought about that. But I think the thing that has changed most recently with generative AI is that AI is no longer just about “oh, it can automate tasks” — it’s actually about reasoning, being able to think about things, being able to act and execute and get things done. Sowe’re now moving from this notion of automating to actually being able to think about outcomes, and I think that is what is really driving why AI should be an imperative for folks to think about.
So when you think about large loan packages, underwriting guidelines, lender overlays, or evaluating conditions — all of these things can actually now be done by AI. And if you think about evaluating conditions, you need to understand Freddie and Fannie policies and guidelines. AI can actually do that. You need to be able to understand these large loan packages and map those loan packages and the data inside those loan packages into the policies and guidelines, so you can say “oh, this is why this condition is passed or failed.”
Now all of this was very manual in the past, was error-prone, and caused rework of things. And now with AI you can actually have this done in a more automated fashion. And at the end of the day, why does it matter to the business? The economics change. The productivity of the quartet that Eric talked about changes completely. The loan processor and underwriter are actually doing the same thing multiple times when they’re doing a condition evaluation. Now that can exist end to end — the cost structure changes because of what has to be done by different people and how fast that could be done.
Flexibility as interest rates go up or down makes it very easy to scale up and scale down without necessarily needing to think about the humans in the process. And then responsiveness — when you think about the front end of the business, the borrower and the loan officer, they’re getting a completely different responsiveness, which means conversion rates could go up and your top line could go up. So those are some of the economics that make it very important for people to think about AI first. And those organizations or lenders that think early about redesigning how the mortgage work gets done will probably stay ahead of their competitors.
ERIC HART (PULTE):
I’d add on to that, Siddhartha — extending the metaphor. The humans have been intrinsically placed in the center of the mortgage loop, originally because you didn’t have any other way to do it. But I think even the current regulatory environment has sort of enshrined that — there are human beings doing each role, they have an NMLS license, and there are specific expectations about where they sit and how many licenses they hold. It’s all been designed around four people sitting in a chair, each one of them holding a very specific instrument.
And I think that’s going to be one of the really interesting things to see how it develops. The work is there. When you change how much of the work is automated and how much of it has to be intermediated by a human, you change the roles, and then you start to transform what the quartet actually is — what instruments are there, what music are you playing. It looks entirely different. I think the technology is probably going to sprint ahead of maybe some of the regulators and their understanding of it, and that will create some interesting conversations.
It’s funny — one of the things when you look at Dodd-Frank, which I think everybody’s been talking about a lot given Congressman Frank’s unfortunate passing of late — it doesn’t actually anywheremention in many of the parts of Dodd-Frank that you have to have a loan officer. It just says here are the things a loan officer has to do and what the requirements are. I don’t think anybody ever envisioned a world where some of these roles don’t exist. There might still be people doing the work, but the role itself could look entirely different. So it’s an interesting — we’re at an interesting inflection point right now.
ALLISON LAFORGIA (HOUSINGWIRE):
[Audience Poll] To our audience, we have a question for you: Where do you expect AI to have the biggest near-term impact in mortgage? Is it in processing and underwriting, closing, borrower or LO experience, or compliance and audit? Let us know, and we’re going to give everyone a moment to send in their answers. I’m not going to ask either of you to give us your preview so that we don’t skew any of the results.
All right. Eric, what would your answer be?
ERIC HART (PULTE):
You know, it’s going to be a different question for different lenders based on the nature of their business model. I think certainly the core production space — processing, underwriting, closing — it’sa perfect use case for AI, where it’s a combination of some structured data, a generally rules-based decision hierarchy, and the need to connect multiple pieces of information together using intelligence. I think that’s the perfect use case for it.
I do actually think compliance is going to emerge as a real opportunity. We spend so much time and effort on the front end trying to make sure that you manufacture a loan perfectly — you have stops in your system, all these things to ensure that you don’t make mistakes while you’re manufacturing it. Well, I think AI is going to reduce the cost of being able to go through and do wholesome reviews end to end on 100% of your files. So a world where your post-production QC becomes much cheaper and more automated — that changes the cost structure you have to put in pre-production. Those are two areas I think are going to have tremendous opportunities right out of the gates.
Long term, I think the front end and the human interaction part will be very exciting, but I think that one’s going to take a little longer to flush out as we see where people get comfortable when they deal with AI from a consumer standpoint. On the back-end user — I think the impact is going to be fast and furious.
ALLISON LAFORGIA (HOUSINGWIRE):
To our audience, you can see the results at the bottom of your screen. I’m going to ask Siddhartha if he is surprised by these results. We see processing and underwriting take the two lead positions with about 30% of responses there, followed in a close second by the borrower and LO experience — and by close I mean 4% close, so very, very close results — and then the next most popular result is compliance and audit. Does that surprise you?
SIDDHARTHA AGARWAL (JAZZX AI):
No, actually. I think what’s really interesting is that we have to think about this era of transformation as an operating model change more than a technology project.
And when you think about the operating model change, the question we should not be asking is “what task can we automate?” — rather, we should be asking ourselves “what operational bottlenecks are we trying to solve?”
And then you think about the audience questions and the place where you want to address the operational bottlenecks first — where are places where there’s high manual effort? For example, the loan processor and underwriter doing a ton of work evaluating the conditions and mapping the documents. There’s complex document interpretation both on the policy side and on the loan package. And there are significant operational bottlenecks in terms of how things move from one phase to another, from one player in the quartet to another player. That is why fulfillment — with processing and underwriting — are coming across as a very strong initial use case.
The processors are spending so much time reviewing documents, validating information, interpreting the guidelines — and the guidelines are changing almost every month. Freddie and Fannie are releasing things almost every month — and then evaluating the conditions. Imagine how much knowledge needs to be in their heads to be able to evaluate those conditions and move the files forward.
So I think because generative AI can now reason over these things and orchestrate workflows — and workflows that are very long, not just things done right now but over weeks or months — that is the right place. At JazzX, what we’ve built is a reasoner that can reason over loan packages, evaluate the condition against the policies, identify the missing information, and — to Eric’s point on compliance — explain exactly why the decisions were made and give that provenance.
So, I think if business leaders are looking at prioritizing based on business impact, repeatability, governance readiness — that’s what will be the highest-value outcomes. Think more about workflows that span multiple systems and roles, because that’s where AI can improve the end-to-end operating model, not in a siloed functional area.
ALLISON LAFORGIA (HOUSINGWIRE):
Eric, I see you nodding. Do you want to jump in?
ERIC HART (PULTE):
I mean, I think Siddhartha hit on a number of the key points. This idea of asynchronous work that can be done in the background — so much of the heartache in mortgage production is how do you effectively build workflow. It’s a nonlinear process for most loans, right? Occasionally you get the loan that comes in and it’s very simple and you ask for this and you get this information and it moves to this step and you can just move step-wise through it. But in most cases it becomes very bespoke. So the efficiencies around how and when do you pull information in, how and when do you make decisions, how do you route one task from point A to point B.
A lot of the leading lenders in the space — the Rockets of the world — have, through decades of work, figured out how to program workflows to try to predictively score, “should I do this, then do that?” and built a lot of efficiencies one inch at a time. And I think what we’re seeing with agentic AI is: you take a human being’s ability to use their own intelligence to organically decide where and how you prioritize, and now you create agents that can do that, and you can have as many as you need. It opens up a whole different way of thinking about the sequencing of work.
So even if you’re not getting automation — which I think there’s tons of opportunity in automation, we have automation today, there’s lots of places where you can use it — it’s how do you ensure that the automation is being sequenced in a way where it actually creates efficiencies? And that’s the part where I think humans and systems have struggled to maximize that, and where AI has the ability to supercharge it.
SIDDHARTHA AGARWAL (JAZZX AI):
Rules-based systems are very brittle, and reasoning-based systems — which are now very possible — are not structured in that “this is exactly what you’re going to do when this happens and you do this next” way. Because to Eric’s point, exceptions are always happening in that mortgage manufacturing process. And it’s how do you deal with those exceptions and yet be able to deliver a deterministic outcome that can be justified.
ALLISON LAFORGIA (HOUSINGWIRE):
Now, there are a couple of questions in the audience that I think we’ve touched on lightly, but I want to get an explicit starting point out there. Where are the most practical starting points for AI in the mortgage process, and how should lenders prioritize those?
ERIC HART (PULTE):
I’ll jump in here, Allison. This is the best response to a good question, right? It depends. Look, I think it really does depend on your business model. So if I’m a retail lender, maybe a small broker — so much of my time and effort is around getting eyeballs, converting those into applications, pulling them through, keeping them in the process. I think there’s probably a lot of opportunity to start with point of sale, top-of-funnel, marketing efficiencies, so that the time and money you’re spending to get your business is efficiently spent.
I think for larger lenders — those with maybe built-in distribution channels, particularly banks, credit unions, or as a builder affiliate where we don’t really do retail lending, we only serve the clients who are purchasing homes from our home builder parent — I think you start with production efficiencies. Because that’s where the biggest opportunity is, the dollars are there, you can learn faster, and if you’re working with a counterparty they’re going to see transferable lessons that they can help you take advantage of as well. And it’ll create cost reductions you can use to pay forward into your pricing model and into your service levels for customers.
So there probably is a fork in the road. For some lenders it really is going to be about how do I, in an environment like this where business is hard to come by, use AI to make the most of that? And then I think those that have a business model more focused on executing the business they do get — core processing and underwriting is probably going to be the fastest road to making some impact.
ALLISON LAFORGIA (HOUSINGWIRE):
All right. Well, now there’s of course the balance between lenders thinking about benefiting from AI while still preserving their existing systems and technology investments. How should they be thinking about this?
SIDDHARTHA AGARWAL (JAZZX AI):
Yeah, and I’ll bring my technology lens and I’m sure Eric will bring his business lens. I don’t think one should think about AI as ripping and replacing — that if you have AI then you’ve got to replace your core systems. I think what AI is doing is introducing a new layer — a layer that we might want to call a system of intelligence — that can sit on top of existing platforms like the LOS. And that system of intelligence can handle reasoning, workflow orchestration, document understanding, policy interpretation, etc.
I think back to my Oracle days, and this is somewhat similar to what happened in the ERP world. We used to have something called EBS, E-Business Suite. One of the biggest challenges was that business logic became deeply embedded inside it — customers would customize it and put all the business logic inside, and effectively changed the DNA of that E-Business application. So then over time, every customization made upgrades slower and took two to three years. It became more expensive and harder to manage.
I think there’s a similar pattern emerging in mortgage today, where too much of the business logic is getting indoctrinated into the LOS or fragmented across the point solutions. So I think the opportunity from AI is to decouple the intelligence from those core transaction systems and have a system of intelligence that’s residing on top of your system of record — where the LOS, for example, or the servicing platforms, are storing the transactions and maintaining compliance — and AI sits on top and helps preserve existing lender investments. Moving that business logic to that layer lowers the risk, improves the adoption, and allows organizations to transform incrementally instead of trying to do a big-bang replacement approach.
ERIC HART (PULTE):
That’s a really good point, Siddhartha. Running a mortgage company from a technology standpoint is sort of like being the athletic director of a D1 college sports program — it all comes down to who’s your football coach, right? The center of everything.
And what is your LOS, what are you going to choose for your LOS? And what is an LOS, to your point? It’s a data repository, it’s a business rules engine, and it’s an interface that your team members use to transact. The data repository and the rules engines — they have to work in line with the requirements of the products that you’re making. But all of the time, the money, the effort, the intelligence is built around the transaction interface. It takes you months to train somebody on how to use your system — here’s how we navigate from this screen to that screen, which buttons we push. And I think what agentic solutions can do is add that orchestration layer where you can continually upgrade and in some cases customize those interfaces to the needs of individual team members or their portfolio of work — and you don’t have to worry about maintaining and managing that technology platform.
The metaphor I would use for how I think about our legacy technology stack and what we can do with AI — it’s a bit far afield here, but if you think about the United States military right now and the decisions they’re having to make about how they invest in their future technology — we’re watching in these horrible conflicts in Ukraine and Iran, the pace of change around the battlefield is remarkable. Drone technology, missile technology — every single month there seems to be some new innovation coming out. These folks in Ukraine are inventing whole new weapon systems in garages and putting them in the field a month later.
So you might think, okay, that means countries like the United States with big legacy military systems — planes, boats, etc. — they’re going to have to throw all that out and move to something new. And what we’re seeing is actually the opposite. Last month the U.S. Air Force put an order in for like 300 F-15 jets — a jet that first flew in 1972. And they’re doing a modernization on their B-52 bombers, which first flew in the Truman administration and haven’t been built since the early ’60s. And why? Because when you equip them with new data uplink systems and better radar communication protocols, these old legacy systems turn into bomb trucks that you can equip with whatever the newest weapon is.
It’s a far more consequential use case for modular technology than mortgages, but I think the same lesson applies. There’s value in a lot of these legacy tech stacks that we’ve built. They work well. They have traceable data. Our regulators understand them. We know how they work. Being able to bring agentic AI in to supercharge those platforms with new capabilities — I think that gives you the best of both worlds over time.
Look, all of us are going to be replacing those legacy applications with more AI-native software tools. But the ability to do that incrementally, with thoughtfulness and change management — that’sone of the most exciting things about AI. It doesn’t feel like we have to throw all of our old work away and do something new. It’s rather that we can take what we have and make it better.
SIDDHARTHA AGARWAL (JAZZX AI):
I love Eric’s metaphors. Metaphor after metaphor after metaphor. We need that skill, Eric — you’re going to have to teach us how that comes into your mind.
ERIC HART (PULTE):
They’re not mine. I’ve got a live Claude feed next to me, feeding these to me as we talk.
ALLISON LAFORGIA (HOUSINGWIRE):
Okay, that’s what I need. A live Claude feed. Okay, got it. There we go.
*[Audience Poll]* So to our audience, you’re going to see a question at the bottom of your screen. Now that we’ve talked about getting started and balancing legacy systems while implementing AI in your tech stack — what is the biggest barrier to deploying AI in your organization? Is it change management, integration with current systems, trust and governance, unclear ROI, or executive management? Let us know what your barrier is before we get into our next section.
All right. Here we go. So our top result is integration with current systems, followed closely by trust and governance, and then followed by unclear ROI, then change management, which is tied with executive alignment — which gives us some good places to move on to. Eric, you just touched on change management at the end of your answer. I want to bring back that topic and get your perspective on what lenders should be thinking about on the change management side of AI transformation. How does it actually impact roles, workflows, and the organization?
ERIC HART (PULTE):
Yeah. I think there are two areas of change management that we’ve observed. The first one — and you could almost say this is part of trust and governance — is how do you bring these solutions into your organization? How do you move towards even deploying new AI tools?
Most of the innovation on AI we’re seeing right now, unfortunately, is not coming from established service and product providers. They’re doing some incremental things, but the more disruptive ideas are coming from new players in the space who don’t have track records — younger companies, more nimble. But if you’re a large lender or bank, you see counterparty risk everywhere, and you certainly see it with newer startups. Add to that the fact that these are new kinds of contractual agreements you have to work through — getting your arms wrapped around how do you share risk, data ownership, as the system learns how much of that is your IP versus their IP, how do you cost-share on computation resources. There’s a whole onboarding change management piece that I think we’re going to have to get better at.
And that’s before you even get into: now that you have these tools, how do you actually change within the organization to make the most of them? I think what we’re seeing is the biggest change management — headwinds isn’t a strong enough word, but headwinds — that we’ll have to work through is the tendency to want to take these new tools and just apply them to your existing processes and your existing way of doing things. Going back to the string quartet example — it’s still four people sitting in a chair playing this kind of music, and now we’ve given them better instruments, and that’s where we’ll get the efficiencies. No. You actually have to deconstruct some of the roles and responsibilities you have. Think about your processes differently.
And that’s harder for mortgage companies, which really do have these very clearly defined, regimented, industry-standardized roles. That’s going to be the hardest thing — to break those apart in a way that you might not see in other areas where it’s not so cut and dry. This person does this and this person does that.
So if you can get through the change management of how do you actually partner up and work with some of these new technology and solution providers, get past that — then you’ve got the exciting but challenging work of how do you envision your workforce differently? How do you think about the roles you have? And then, oh by the way, how do you think about your agentic workforce? Because you’ve got a virtual workforce now, not just the folks that you’re writing paychecks to.
SIDDHARTHA AGARWAL (JAZZX AI):
You know, that last statement about how do you manage your agentic workforce — the digital workers that you have — I think that’s wonderful. And I take a little contrarian lens to what the audience might have put out as the biggest barrier. I really think that change management is very important in the AI world, and where that change management stream should run in parallel to the technology project.
Because AI is changing how workflows happen. AI is changing how decisions get made. AI is changing productivity expectations. For example, if you take an underwriter today, they process about 1.8 loans per day — maybe different lenders have different numbers, but let’s say 1.8 to 2.2 loans per day. Well, what if AI handles large portions of the guideline interpretation, evidence gathering, and condition evaluation? Suddenly the human is operating at a very different productivity level. So what is the human supposed to do? What do you want the human to do?
And then other things that AI changes: organizational structures, and also how institutional knowledge gets managed. And I think that’s probably the most profound thing. Today, a tremendous amount of mortgage expertise lives inside individual processors and underwriters. In an AI-enabled organization, every time an underwriter overrides what the AI system recommended, that now becomes a learning opportunity. So with that learning opportunity — how do you decide whether what these five underwriters said should become part of the policy of the organization?
This might mean you need new roles. For example, a policy supervisor role — and this does not have to be new headcount, but a subject matter expert to whom JazzX is bringing all the changes that underwriters said yes or no to, and why they said that. And now it is this policy supervisor answering questions like: should these overrides become part of my organization’s policy? Should we add this information into our knowledge base? Or should the AI be given these policy updates so it can reason differently? Should we add these into the reasoner thought-process?
So this whole notion of — what portion is AI assisting humans, maybe 25%? What portion is AI and humans collaborating deeply, such that maybe 75% automation can happen? And what portioncan be completely automated without human oversight? That is where change management is required, so that leaders can think about: how are my roles evolving? How are humans and AI collaborating? How does knowledge get codified? How does governance work? How is trust built? And I think trust is the most important thing here.
ALLISON LAFORGIA (HOUSINGWIRE):
Governance was the second biggest result in the audience poll, which means it’s a really important piece to delve into today. So what does AI governance look like in a regulated mortgage business where accuracy and auditability are critical? Siddhartha, you just touched on this — let’s delve into it more.
SIDDHARTHA AGARWAL (JAZZX AI):
Yeah. I think with the advent of AI doing a lot more work, you cannot have an opaque or a black-box system making those decisions and affecting borrowers, compliance, or investor risk. The AI system needs to have:
**Explainability** — why did it make the decisions it did?
**Auditability** — to be able to show, this is the piece of policy that was used, and this is the loan data package that was used, to come to this conclusion on whether this condition was failed or passed.
And one very important thing is it needs **deterministic outcomes**. Today, LLMs — large language models and generative AI — can be probabilistic, meaning the answers can change. In a mortgage, you cannot have answers changing based on something AI did. So you have to use a combination of generative AI and deterministic business processes. Then governance requires **human accountability**. At the end of the day, who owns the outcome? Who’s going to be held responsible for the outcome? I don’t think AI is responsible for the outcome — it’s humans. So what are the escalation paths? How are exceptions handled? What are the oversight mechanisms that are there? Who is able to approve what? Those are things we have to think about.
And last is **continuous monitoring**. When we used to take technology and move it into production in the past, yes, you did monitoring, but you didn’t expect the answers to change. Well, now with something called model drift — where large language models can drift in production — you have to actually continuously monitor the outcomes happening in production. Check for accuracy, check for drift, check for exception patterns.
And I think at the end of the day, people do not trust AI to do certain things. And this is all going to come down to how the right governance and the right approaches can create trust — for executives like Eric, for the regulators, for employees using the systems, for investors, and ultimately for borrowers. They’re the customers.
ERIC HART (PULTE):
Well, to quote the great Kent Brockman from The Simpsons — I, for one, welcome our new AI overlords, and I’m happy to help them as they force us to toil away in their underground data centers. It’sreally funny, and we were talking about this before we came on — I’ve got three kids, a 15-year-old, a 13-year-old, and a 10-year-old. The 13 and 15- year-old feel very passionately about things they have very little knowledge about, which is a pretty typical experience for kids that age. And there are lots of strong feelings about AI.
There’s power and there’s danger in something that you can describe or put an identity around, and AI is sort of this vessel that people channel all of their anxieties about technology into. From a governance standpoint, I think a lot of folks who run companies — if you’re listening to this, you probably know this — it’s treated as this whole separate magisterium, this thing that we have tomanage differently, AI is its own source of risk.
And I think a little differently. I don’t think the sources of risk that we’re trying to manage have changed. Think about what we manage in the mortgage industry from a risk standpoint: data security, privacy, quality control, fair and unbiased lending practices, counterparty oversight. Those haven’t changed. We still have to manage them. And as we bring AI solutions into place, regulators aren’tgood about considering counterfactual risk.
Yes, there’s risk when you bring AI tools in, but that risk exists already. Are you making it better or worse? Is an AI engine going to make a biased decision? It’s a risk you have to manage for. But Iguarantee you who’s really good at making biased decisions: human beings. We thrive at it. And in some ways — to the point that Siddhartha made — you’ve got 100% auditability, traceability, you can see the provenance of the decisions that are being made and why.
Sometimes you can’t always get that out of an underwriter. Hidden biases they don’t even know that they have. So I think over time, my hope is that we move from a world where AI is seen as thisseparate thing you have to govern, and rather it’s another use case of technology — and you’re managing risks that you’ve always been managing, just thinking about how you need to adapt those oversight regimes for the specific way that AI uses them.
We’re seeing that with regulators. Sitting here in Colorado — Colorado was the first in the country to come out with a very strident AI policy, and they went through a couple of rounds in the state legislature modifying those rules. I think even the representatives started to wrap their heads around — okay, is this totally different, or is this just the next technological horizon of risks that we are already asking people to manage?
It’ll take time for companies, for people, for regulators to get their heads wrapped around what’s different and what’s really just the same old risks wrapped in a new technology solution. But what we have to do as a mortgage company at the end of the day isn’t going to change. We get paid to identify whether or not money should be lent to somebody and at what price based on the risk. AI doesn’t change that. It just gives us different tools for solving that question and then creates different expectations for how we manage the risk.
ALLISON LAFORGIA (HOUSINGWIRE):
I think that’s a fantastic point of view that I hadn’t considered, so thank you,
Eric, for that perspective. And before we wrap up with our last questions, to Eric and Siddhartha, I have one more poll question for you guys — probably a very critical one: How far is your organization in adopting AI? Where are you actually at? Are you just exploring? Are you running pilots? Is there limited production use? Are you scaling across the organization, or have you not started yet? Let us know where your organization is at.
I have to wonder, Eric — considering your last statement — is this a similar inflection point to previous iterations of technological innovation? Are there always the same concerns? Are these similar risks to what we’ve been considering with new technology before? I think that’s a fantastic point to raise at this juncture. It doesn’t have to be as scary as people might be thinking.
ERIC HART (PULTE):
The Journal had a great piece on this a couple of weeks ago, talking about using automobiles as a historical example — the disruption that automobiles had on society was probably even moreprofound than what AI is going to have here. You could die getting thrown off a horse just like a car — you just needed better brakes on a car. So I think your push is the right one. This isn’t that different than inflection points in the past. It’s just powerful and it’s here now.
ALLISON LAFORGIA (HOUSINGWIRE):
I was talking to someone who works on AI in the mortgage space and works on teaching people about AI at some of the MBA events, and her stance was that prior to the invention of airplanes, transcontinental travel seemed impossible for an overwhelming amount of the population. And now it’s something that is entirely possible, and there are even airlines pushing to make it increasingly affordable. So perhaps what we think of as impossible now is similar to what people had thought was impossible 80 or 100 years ago.
SIDDHARTHA AGARWAL (JAZZX AI):
Yeah. If you think about it — the LLM model you’re using today, the AI you’re using today, is the worst AI you’ll ever use in your entire life. It’s only getting better at a breakneck pace from here.
ALLISON LAFORGIA (HOUSINGWIRE):
That’s a great point. So our audience results are in. We are very firmly in “running pilots,” which is fantastic to hear. Running pilots followed closely by “just exploring,” then “scaling across the organization” — I love that that is a close third — then limited production use. And in a fantastic turn of events, there is nobody who responded “not started.” Love to see that.
So Siddhartha — you just mentioned that the AI we’re using today is going to be the worst application of AI we’ll ever use. If we have this conversation again in two to three years, what will be true about the lenders who leaned into AI transformation early versus the ones who waited?
SIDDHARTHA AGARWAL (JAZZX AI):
That’s a very tough question — it’s like asking do we have a crystal ball or not. I think those lenders who really think end to end about the transformation and the operating model change — all the way from borrower to loan officer to loan processor to underwriter to closer to funder — and by the way, that’s how we think about JazzX’s solution, because the same work is being done to some extent by the loan officer and the underwriter and the loan processor — when they think about it end to end, I think they will be able to operate at a completely different level. They’ll have materially higher productivity. They will lower the cost — that $11,000 cost of producing a baby versus a loan. They’ll have faster cycle times, which means the borrower experience is better and they’re able to get a higher conversion rate from borrowers into the funnel. And the organizations will be more scalable, there’ll be fewer handoffs.
So I think those who start now and actually focus on how do I move that institutional knowledge out of the individual employees into the system of intelligence — I think they will experience a completely different economic base. Their top line will grow and their bottom line will improve. The cost of production will improve. Those lenders who start now, in two to three years, will be at a completely different place in being able to manufacture that loan much faster and with a much lower cost base. Eric, I’d love to get your thoughts on this.
ERIC HART (PULTE):
I really like the point you made — I never thought about this, that today’s AI is the worst AI you’ll ever be using. It reminds me of one of my favorite comedians, Mitch Hedberg, who had a great joke where he says, “I hate it when people come up and say, ‘This is a picture of me when I was younger.’ And I’m like, every picture of you is a picture of you when you were younger.”
But look — this is the future. I don’t think anybody can avoid this, and if you don’t adopt it, you’ll get left on the side of the road. I’d maybe pivot the answer to — what’s going to be true about the people standpoint at the companies that adopt it now versus those that adopt it later?
We don’t have a crystal ball. We don’t know which solutions are going to succeed, how fast it’s going to go, what the savings are going to be, whether those savings will be offset by additionalcomplications. But given that AI will absolutely be the way that we interact with technology moving forward, I think those that start earlier will have an advantage in that they’ll have more opportunities to have their team members learn how to use AI, to become AI-fluent.
It’s honestly one of the most terrifying things as a leader — new technologies that come up that you don’t know. Most people in mortgage got to a leadership position because they were the best at being a loan officer, or because they understood the technology, and it was mastery of the tools and the systems that got you to places of higher and higher authority. Now this disruption comes in, and the 20-year-olds we’re going to be hiring into our companies are going to be native AI speakers. The rest of us — AI is going to be a second language, and I think we’re always going to have an accent.
So I think the sooner we can train and give our team members — even those 30 years into their career in the space — opportunities to learn how they can use AI and actually become fluent in AI, those workforces are going to be better positioned to pick up quickly on whatever the next iteration of technology brings. No different than 40 or 50 years ago, companies that taught their employees how to use computers and gave folks an opportunity to engage with digital technology were the ones that could move quickly as that technology continued to accelerate at an exponential pace.
Same thing with AI. Any mortgage lender that a year from now can’t point to at least one use case where every one of their employees is using AI in a meaningful way — I think you’re putting yourself at risk.
SIDDHARTHA AGARWAL (JAZZX AI):
And if I can add to that — two things on the people part of it. One, in a recent conversation, the VP of Underwriting actually got tremendously excited because she saw the ability to change the process herself, because it could be done through a natural language interface yet governed by the system — rather than having to figure out who do I go to in IT, how long is this going to take, what are the IT resources, what are the IT skills required. And she said in a lot of cases she just didn’t make the change because it wasn’t big enough to justify all of those needs. But now she said “I can actually streamline my processes myself. I’m in control of my own destiny.”
And one of the tenets I’ve set for my organization is AI-first innovation — which means refuse manual toil. Ask yourself where you can use AI, and really try and answer the question: why can AI not help me here? But then if all you do is give back the answers from AI, then we don’t need you. So you also have to think about: where do I bring in judgment? Where do I bring in my knowledge and my experience? And how do I make the answer — whatever outcome I’m delivering — completely different and much more powerful, where the manual toil is gone but it’s so much better and done so much faster?
ALLISON LAFORGIA (HOUSINGWIRE):
Absolutely. There’s a joke that I think does mean you might have to give up your calculator. I’m impressed you still have an abacus.
*[Laughter]*
I do think it’s also interesting where this is probably similar to where the example of training on a computer came up, but also the transition from paper to digital. That was probably a process to get used to all of those systems as well.
SIDDHARTHA AGARWAL (JAZZX AI):
I think the good news is that everyone has gotten used to it. Meaning — who hasn’t used Gemini? I’m a Google fan, so who hasn’t used Gemini, or who hasn’t used ChatGPT, or who hasn’t used Anthropic Claude? Everyone’s used it in some way, shape, or form. So the great thing is people can see the power, they can see what it can do. But I think people are also realizing where they need to bring in their judgment. For example, students working on their papers — one of the professors said, “Look, the HBS case studies you need to analyze — feel free to use AI. You can get the answer to it. But what I’m going to do is I’m going to have you interact with personas that represent the people in that case study. And now you’ve got to figure out how to engage with them, how to partner with them — but delivered through AI. Those personas are AI personas simulating that Harvard Business School case study.” So it’s not enough to just get the answer. You have to think differently about how you approach that.
ALLISON LAFORGIA (HOUSINGWIRE):
All right. Well, we are rapidly approaching the top of the hour. Eric, Siddhartha — thank you so much for sharing your insights with us today, for giving us multiple points on governance, on how to get started, on the balance between AI and legacy systems. I want to turn things over to you guys to give final thoughts before we wrap up.
ERIC HART (PULTE):
Oh, nothing — just, um, I don’t have any deep thoughts. I guess just more — I think it’s easy with this new technology to feel like you’re behind. I loved that question you asked earlier: is anybody not doing anything? I can’t imagine anybody in their right mind admitting they’re not doing anything. Even if they’re not doing anything right now, everybody feels this anxiety like “I’m supposed to be doing more.”
As leaders, I’d say: don’t abandon your judgment. This is a new technology. Take a measured approach, find things that work. The opportunities will present themselves for when you need to hit the gas pedal. This isn’t — once again — the first time those opportunities haven’t presented themselves or you’ve had to navigate new technology or challenges. So just for any folks who are similarly wrestling with how do I play, where do I play, when do I jump in — this isn’t different from any other decisions you’re making. Trust your judgment, be smart about where you deploy resources, and you’re probably not as far behind as you think you are.
SIDDHARTHA AGARWAL (JAZZX AI):
And I would just say — if there are two takeaways from here, and I also heard this from Eric: one, think end to end about the operational transformation that this is going to drive. Don’t just think of this as a piece of technology in a particular silo that you can automate — but take the leap to think end to end about how you can transform your entire mortgage production.
And the second thing is: think about how you can leverage AI to institutionalize the judgment that today resides in the minds of the folks — the quartet that Eric was talking about. How do you institutionalize that judgment such that you can do better reasoning, better automation? And lastly — what is the role that humans should play around judgment?
Those are three things. I guess I said two, but those are three things that I would say are takeaways.
And Allison, thank you so much for being a wonderful moderator. Really appreciated all your questions and the polls — reaching out through the polls and helping us understand where those poll results landed. That was wonderful.
ALLISON LAFORGIA (HOUSINGWIRE):
Thank you both. I have several pages of notes, so I appreciate both of your perspectives. And to our audience, thank you for joining us. That wraps today’s webinar, “Rewiring Mortgage for the AI Era: Why AI Is Now a Leadership Imperative”.
Eric, Siddhartha — once again, thank you so much for taking the time and sharing your expertise. And to our audience — thank you for joining us. HousingWire will be sending a recording out of today’s session to everybody who registered, and it will also be available
