The mortgage industry has already embraced automation and AI through optical character recognition (OCR), automated underwriting systems (AUS), and verification of income and employment (VOIE) tools. But generative AI and agentic AI systems are poised to offer even greater efficiencies and opportunities. This expert panel will be a deep dive into the AI technologies that are reshaping the mortgage industry. Learn how to implement these advanced systems without catastrophe, mitigate risks and transform your operations, leading the way with intelligent automation in an increasingly demanding market.
Transcription:
Transcripts are generated using a combination of speech recognition software and human transcribers, and may contain errors. Please check the corresponding audio for the authoritative record.
Suha Zehl (00:08):
Well, you haven't heard about AI yet, right? It's brand new. Everyone's really excited about the potential and the possibility of AI, but really when you strip it down, it comes down to three things: people, the "who"—I could have said resources, but you'll see why I chose people, and that would be your human or your AI; it's also the process, the "what," your workflows and what you're trying to do; and then the third thing is the product, the "how." That's your AI. That's what you're going to be leveraging. They only gave us 30 minutes and we're already down to 29:25. So we're going to dive right in. There's been a lot of talk about AI obviously, but not a lot of focus on the human side of AI. If we hear the headlines, AI is going to replace all of us. So go out, update your resume, and look at how you're going to upskill. But as business leaders, how do we prepare our companies to foster an AI readiness culture? I'm going to give that to you, Joe. I was looking at you, but I didn't say it.
Joe Welu (01:27):
Great question. I tend to pay very close attention to the leaders in any category and what the best in the world are doing to get the most results. I would say that when I see an organization that is very clear and intentional about both the importance of AI initiatives, where you're going to focus those initiatives, and setting the expectation that the world is never going to be the same, the way forward and the future is going to look very different than it did over the last 20 or 30 years in this industry. As an organization, communicating consistently about why it's important, what you're doing, and the expectation that everyone leans in, I believe, is really critical.
Suha Zehl (02:24):
TJ, anything you would add?
Taranjeet Kaur (02:27):
I personally feel that what Joe was saying about the people and the process and focusing on where you want to be is key, but can we just stop calling it AI? Can we just call it technology? This is the same technology that got us to the cell phones and that got rid of polio. It's the same technology that is setting us up for the future. It's just a piece of technology that will be used for the advantage of humans. I'm at a point where I've heard so much negativity on AI and what it can bring in. It's just technology. I was listening to Jamie Dimon—I'm a huge fan and listen to most of his podcasts—and he was like, "now my kids can live up to 200 years thanks to technology." AI is a piece of technology, so we better embrace it. The people part of it, the human part of it, is the first step in the process.
Suha Zehl (03:23):
I do need to take one step back because I made a mistake and didn't introduce my panel. We've got Joe Welu, who is the president and founder of Total Expert, in case you didn't know. We've got TJ Kaur, who is with JPMorgan Chase.
Jack Cavanagh (03:44):
Yeah.
Suha Zehl (03:44):
And then we've got Jack Cavanagh from ICE. I apologize for not introducing them earlier. Jack, I'm going to follow up on that and ask you: as roles begin to change—and we've heard that's going to happen—how do we communicate to our staff and teams that while AI is not going to replace us, the person who's using AI is going to replace us?
Jack Cavanagh (04:06):
I think one of the biggest challenges is getting people comfortable with the change. There's some unknown right now. I totally agree with what Joe was saying, that the world is going to change. This is real, it isn't just hype. How it's being implemented today is not necessarily how it's going to be implemented five years from now. You have to get used to the unknown and the adjustments that will have to be made. To build on what TJ was saying about the fear of AI taking away jobs, I see it more as an augmentation and an opportunity to strengthen roles. Will it reduce the number of people for some tasks? Yes, it definitely will. But we've got to think about productivity, not just in terms of doing the same with less, but to do more with the same. There's more that can be identified there.
Suha Zehl (05:12):
That's a great point. TJ, I'm going to ask you this: we've talked about the human side and how it impacts adoption, but adoption starts long before implementation. For businesses looking to scale into newer solutions like Agentic AI, how do they prepare? Not everybody is a UWM or a Rocket with tons of money and resources.
Taranjeet Kaur (05:47):
Agent AI is the new buzzword in the industry right now, and everybody's jumping on it rightfully so because Agent AI addresses the problem we have in mortgage: "Can I run my operations better so that I don't have the turns and the lows due to the cyclical nature of mortgage?" The change starts with Agent AI or embracing other sorts of AI through the contextual awareness of your process. Up until this point, you were taking a document, ingesting it, and making sense of the data. Now that ingestion is part of a bigger workflow with Agent AI. You need to understand your workflow, as someone called out in the prior panel. Have contextual awareness and a mindset of continuous improvement. Ask yourself, "Why am I doing this?" and maybe you improve from that point of view. Do I need to do this in the future state?
Once you get the hang of it, you determine if you need agents for it. There are agents for underwriting and compliance. 1 There are super agents who can guide your agents to learn from the data. You need to have contextual awareness; that is where AI can uplift the mortgage world. Prior to this, it was just point-in-time: I can read a document, a picture, or a credit report and make sense of it. Now it is part of a bigger workflow. That's where it becomes really transformative. We need to embrace understanding our process end-to-end rather than just piecemeal solutions. That's where it will start to differentiate.
Suha Zehl (07:36):
I see you nodding, Joe. You want to add something?
Joe Welu (07:39):
She's got some great sound bites. I agree that this is the greatest innovation wave in human history. AI is how we describe it, but it is really about innovation. If you think of other technological transformations, this is the largest one. As you're thinking about where to put AI agents into your business and workflow, my favorite analogy is that humanity just discovered fire. Everybody is running around lighting everything on fire all the time with AI, and we're like, "oh, maybe we shouldn't light some things on fire." Maybe we need to learn how to leverage it, harness it, and be thoughtful about it. I'm in a lot of boardroom and C-suite conversations where people want AI everywhere, all at once. Actually, there are certain workflows where you're going to compound the value of your organization and force multiply the productivity of your people. Those are the areas we focus on. There's an insane long tail of opportunity, but I think it is going to be iterative.
Taranjeet Kaur (08:58):
Another soundbite in my mind: when I grew up back in India, we used to have phone booths. I would go to a phone booth to make an international call. They don't exist anymore. We have evolved and those families are perfectly fine. We have evolved to a new way of working. The same applies to video game parlors; very few people go to them these days because we have compact tools and technology at home. It's going to be something of that nature—transformative, but we have to embrace it and make the best use of it.
Suha Zehl (09:42):
I'm going to ask you, Jack, because you and I have had this conversation. A recent study said 95% of AI initiatives are failing. 2 As organizations, how do we make sure our initiatives are not going into that AI graveyard?
Jack Cavanagh (10:05):
That's a good question. To give context, that 95% failure study involved a smaller sample and a shorter period of measure. It was more qualitative than quantitative. Thinking about it broadly, we have to look for opportunities for low-risk, high-reward wins. Look at specific areas when you're testing rather than trying to change everything at once. I totally agree that you don't want it to be piecemeal, but you've got to look for smaller pieces you can identify. There's a lot of antiquated processes and plenty of opportunity. Finding one of those higher-reward, lower-risk areas, like the back office, is a way people can start moving in that direction. Tying into the last question, get your feet wet and understand the opportunity before you set everything on fire.
Suha Zehl (11:15):
I think what you're talking about is finding those quick wins and use cases. Where are you going to get the most ROI quickly to demonstrate success and build trust and credibility?
Jack Cavanagh (11:27):
Absolutely. Lower risk is key, especially as you're starting, particularly in terms of compliance.
Suha Zehl (11:34):
I'm going to stay with you because Agentic AI thrives on data, and you're the head of data science.
Jack Cavanagh (11:53):
Yes, correct.
Suha Zehl (11:54):
I use the analogy that AI is like a high-rise, and a high-rise needs a solid foundation, which is your data. How do organizations ensure their data is high quality, break down silos, and create an environment for initiatives to be AI-ready?
Jack Cavanagh (12:18):
This has been a challenge and an opportunity even for traditional AI before we even get into agent or Gen AI. Whether I'm talking about robotics, RPA, vision (OCR/IDP), or natural language processing—those are your chatbots or voice bots—they are all powered by data.
The first step is understanding the data you currently have. You have structured database data from different applications and systems; you have to make sure it's connected and clean. That's your base foundation. Next, where you really build value is adding additional structured data you can buy on properties and borrowers—purchasing habits, demographics, property history. Unstructured data is where you pull information from website history or chat logs. Really, a big value is being able to take the data coming out of these large language models and incorporate it into existing models or build new models on top of it. It's about combining structured internal/external data with the large amount of unstructured data from chats, call logs, and document reading. That is where there's a huge amount of power for prediction.
Suha Zehl (14:18):
I think you're building the case to make sure that everybody takes a step back and looks at their data. Make sure you have governance in place and a data strategy. Understand who has access to what and why they need that access.
Jack Cavanagh (14:38):
That's what I mean by connectivity and cleaning. It's making sure, without getting too heavy into data governance and stewardship, that you understand and organize information to extract value. It's not necessarily about building large enterprise data warehouse projects. It can be data lakes where you're just making sure data is available, consistent, and clean in the same area.
Suha Zehl (15:16):
TJ, Joe?
Joe Welu (15:18):
Regarding the data side of things, it is an evolution of making sure your systems of intelligence and your insights around customers can be properly accessed. Related to getting value and improving ROI quickly on AI, the thing we are most excited about is how quickly you can actually generate an ROI. Historically, if you were implementing large core systems, it took one or two years to see value. What if you could control the blast radius with a smaller set of data and get value in four or eight weeks? Those are the things we see happening. Most data projects have failed to yield the needed ROI primarily because they don't connect into the workflows in the systems of action. Agentic AI gives you the superpower ability to connect that intelligence to a system of action. AI agents, in the simplest way to think about it, are going to do a task that the human doesn't have to do, or maybe doesn't want to do. It's amazing potential.
Taranjeet Kaur (16:44):
AI is not immune to "garbage in, garbage out." 3 The better data you serve to AI, the better results you're going to get. Joe and I are saying the same thing: with Agent AI, you're getting more context. You're going to build AI on top of AI. You're learning from your data and that data is automatically improving your AI. That is a really exciting advancement in the AI world.
Joe Welu (17:28):
To dovetail on that, one of the key lessons I have learned from working with organizations is the importance of really good data architects and software architects. I could bucket 80% of the problems I see out there into just not having proper architecture understanding.
Suha Zehl (17:49):
You've actually hit the follow-up question I was going to ask. If you had a magic wand and could change one data challenge in mortgage lending right now, what would you fix?
Joe Welu (18:09):
Who wants to take that one? Data guy? I'll take a stab. Real-time access to all core systems for any of the data you need, and the ability to stream that into a central place. When I say real-time access, a lot of the legacy systems you're running are a major hurdle. The speed of AI has accelerated everything, including consumer expectations for things to magically appear. The only way to do that is to be as real-time as humanly possible. I would love to see solved real-time streaming access to any core system data needed to make better, smarter decisions.
Taranjeet Kaur (19:11):
I have a software developer spin on it. One data problem I would like to see solved is defining what data is. Is data the value sitting in the database, or is data the requirement coming to me? In the prior session, they mentioned using data for AI for requirements writing. We need to broaden our horizon of what data we want to work on. If that problem gets solved, we'll be better and more efficient.
Joe Welu (20:01):
It sounds like people get confused between raw data and the insights you need from that data.
Jack Cavanagh (20:14):
You're basically going from data to information to knowledge, and you're leveraging the AI to actually get you to that level of information. Exactly. Data is raw, and you're using the AI to help you build on it.
Suha Zehl (20:22):
We talked about legacy systems built on old infrastructure. Our systems are not AI-ready. Some people want to just bolt AI on to get going, but bolting is not really the right way to proceed. This session is talking about re-imagining how we do things. What would you add to that?
Joe Welu (21:19):
Every organization is slightly different in its maturity and architecture. I would be incredibly honest with yourself as a leadership team about what skill sets you have internally. Do you have a traditional IT team pretending to be a tech team? You have to be honest about your team's ability to make progress. You can make progress on certain use cases by bolting capability on top of data, but that should not be the long-term strategy. Start with the lowest complexity, highest ROI use cases to prove value, and then have a roadmap to get to the perfect state.
Jack Cavanagh (22:23):
At some point, it's like 1947 and the jet engine. In some cases, it's like you're attaching a wagon to a jet engine; that's not going to work long-term. There is a benefit to augmentation using traditional technology, but we have to think about the larger picture.
Joe Welu (22:57):
You wouldn't take raw crude oil off the ground and put it in a jet engine. It's not going to run. It's got to be refined and processed.
Suha Zehl (23:12):
The lesson learned: don't just bolt AI on anything. You're setting yourself up for failure. Before we wrap up, I would love to hear from each of you. About six months ago, voice AI became the emerging thing. What emerging technology are you keeping your eye on today?
Taranjeet Kaur (23:45):
Something in the mortgage world that could be disruptive is the Palantir and Alex Carp connection with Fannie Mae, where they tapped into data for fraud detection. 4 I think that advancement can unlock so much more that we can do with data. I'm looking forward to that partnership and its results.
Jack Cavanagh (24:34):
It's not necessarily a new technology as much as the opportunity to leverage unstructured data coming from large language models. As that continues to grow, we can build it into fraud detection, climate risk, and things we have already been predicting. We can now expand by increasing the amount of data available by a hundred or a thousandfold.
Joe Welu (25:16):
If you look at voice AI and agentic AI, so much of the capability has been made possible by the underlying large language models built on top of chips and infrastructure. 5 The next wave of value creation is deep verticalization of these agentic capabilities inside specific verticals. You're going to have fine-tuning of large language models where people take those models and infuse everything in the universe about mortgage on top of them to create amazing value. That's what I'm paying attention to.
Suha Zehl (26:18):
I'll share mine since we have a couple of minutes. Blockchain was the thing six years ago and it didn't go anywhere, but now people are talking about the collaboration between blockchain and AI. I believe there was recently an underwriting decision made and a mortgage completed for an $8 million house using blockchain. I think there is a lot of possibility in how we tie blockchain into the AI infrastructure. We have time for a couple of questions. Julian?
Julian (27:23):
Joe, I'm going to put you on the spot. We had a conversation at lunch about your voice AI and what it's doing for some of your customers. I'm wondering if you're willing to share with the group the instance of lenders actually having to turn it off.
Joe Welu (27:50):
We have a voice AI solution live with quite a few customers. The intention was to use these capabilities to do things that humans don't want to do, aren't great at doing, or that aren't infinitely scalable. We deployed an AI agent on top of a database for multiple lenders. Last week, when rates dropped, their inside sales teams had to push pause because of the amount of opportunities being harvested from the database. An average top-producing originator might have 1,000 or 2,000 past customers; the probability of them reaching out to every one of them at key moments is low. It was cool to see lenders saying, "we actually cannot handle any more volume today." The AI agents were feeding that volume to the humans. It was really exciting.
Suha Zehl (29:14):
Other questions?
Audience Member One (29:25):
To what degree do you think brand remains important as every company levels up with AI and looks the same because AI lets them all be incredibly efficient?
Joe Welu (29:53):
Brand is incredibly important. It's what you stand for and your identity as an organization. That brand identity, if you properly use AI, should be extended all the way into consumer conversations. In the voice agents we're training, we infuse the brand voice so the agent understands how your company speaks about helping a customer. Brand is just as important as it's ever been.
Suha Zehl (30:39)
I would add that brand is a differentiator. AI is an amplifier that will help you amplify your brand reputation.
Taranjeet Kaur (30:49):
Brand is knowing the right problem to solve. You establish a brand when you know the exact problem you want to solve for your customer, colleagues, and business partners.
Jack Cavanagh (31:08):
Brand is the trust and relationship part of the understanding with your customer.
Suha Zehl (31:14):
They're blinking at us, so they want us off the stage. Thank you everybody for coming.
Reimagine What's Possible: AI & Tech Trends Shaping the Future of Mortgage Lending
September 29, 2025 10:56 AM
31:27