Clayton Collins: What does artificial intelligence mean for the world?
Andrew Maas: Artificial intelligence refers to a large class of technology that can be used for many different things. Sometimes when people hear artificial intelligence, they think of singular sentient intelligence, or what we call general artificial intelligence or hard artificial intelligence. For the most part, people like me don’t think about this every day—we don’t think it’s even close. Systems such as Chat GPT – It’s become clear over the past two years that we don’t see anything becoming as smart as a five-eared child any time soon. These are very difficult things in terms of general reasoning and common sense.
The truth is, these language models are very good at fooling us humans… we want to believe they’re smart, and they’re just as good as little chatbots. Now, we’re increasingly designing these different components around and on top of them to make them more useful and more reliable in terms of factual accuracy and interface with the outside world.
But in terms of broad artificial intelligence, you can almost think of it as the adoption of the internet. There’s not just one way for businesses to say, this is how we’re going to use the web to improve our business. And artificial intelligence, especially the new artificial intelligence systems that we can start building and adopting now, you should think about it in the same way, which is not to say that I would hire an artificial intelligence agent to replace an employee, which is the perfect employee, and Sell every mortgage to everyone. No, there are 100 different little use cases.
Rick Rock: Our industry… tends to be a little bit outdated. We tend to focus too much on human resources and repetitive tasks that may hinder scaling or may introduce inaccuracies or errors that in our business mean increased profits. Like we don’t even know how many inefficiencies there are in the mortgage chain, so we just price it out, right? This is why so many innovations, such as artificial intelligence or blockchain, help alleviate doubts about the quality of the products produced. So you have less need to price mortgages across the capital markets chain because you know you have greater confidence in what you’re receiving.
When I’m leading a team off-road, I said, this is really a necessary conversation, not just for Cross Country, but for all lenders in the industry. Because it’s clear that the big divergence that’s going to happen in the industry is lenders leveraging artificial intelligence, and a complete overhaul of the efficiency of their pipelines, and that’s going to be the difference between businesses that leverage fax and express versus those that embrace email. I sincerely believe that the efficiencies will be staggering, both in terms of pricing and consumer experience, and in terms of profitability for lenders, with a stark contrast between adopters and non-adopters.
CC: What do you think companies are building right now?
yes: This is where the big upheaval is happening now. Because basically, where we are today, there are a huge number of new use cases that are enabled by technological advances and now large public models from friendly commercial licensees. Honestly, we probably have five years of work to do in terms of enterprise adoption and integration – even assuming everyone is on board, everyone wants to do it, and depending on what the regulations say.
Now, in general, my advice would be to carefully research the vendors you’re working with because they may be building something for the first time, or promising you something they know will work, but they haven’t actually built it yet. It’s times like this.
CC: You mentioned that the model was trained on public sources. Are there two different AI markets emerging—for example, models trained around proprietary or internal datasets versus models trained around publicly available information on the internet?
yes: This is what people thought might be going on two years ago, but that’s not what happened. Initially, when chatGPT started coming out, there were a lot of people who thought that every big enterprise would use all their internal data and train custom versions of these types of models, which would just solve a bunch of problems. What we found is that if you just take all the internal data and train one of these big models on your private data, you end up with something that talks more like your private data, but it doesn’t actually solve some of the most important problems for you. Valuable use cases. You still need to design the components.
So lately, I think the consensus that’s emerging is that you think of a language model as being a little bit like a paintbrush – I might have a paintbrush from public data that’s really good at talking about mortgages and financial terms and things like that. If we use some public GPT model, they can discuss mortgage terms very well. They don’t really integrate with your back-end systems or structured materials in any useful way to say: This is the mortgage I’m offering and these are my terms. So whether you have a private model or use a public model, this is additional engineering work that you have to do.
So what we’re seeing now is people doing engineering work to integrate their proprietary workflows and proprietary data, but not necessarily training one of these LL.M.s.
CC: How can senior executives begin to lay the groundwork for the impact this will have on their business?
yes: Something I hear a lot in enterprise AI discussions these days is: When and where will the enterprise value of all this AI craze become apparent? Because we haven’t had much hype about ChatGPT in two years. Some senior executives spend large amounts of money on projects with unclear scope. For example, one of the very large private models that we just talked about was built, and then they deployed it internally and they discovered, well, we still have a bunch of use cases that this model can’t solve. Now, how do we actually solve these use cases?
Some of them may be out of reach because they are hard problems that cannot be solved by the language modeling part. So we’re seeing a lot of buzz, a lot of investment, a lot of startups being launched and product ideas being launched, and we’re now trying to figure out: who is actually leveraging artificial intelligence to do something that could save money or create new revenue? process? Now, it’s getting into more details. So the executives that I’ve been seeing lately, they’re not thinking about how do I invest in artificial intelligence? Instead, they are thinking: In which use cases will AI really make a difference? How does this affect my profits? Or what is the return on investment? I think this is the right place to start.
CC: Rick, what did you hear at the mortgage board? What conversations are happening in your C-suite circle about artificial intelligence?
RR: I would say complete chaos. All the boardroom focus is on production…there’s a lot of focus on the winners coming up in October, November, December. Now, the higher up the corporate and mortgage lender tiers, the focus is that demand will undoubtedly be extreme in 2025 and 2026 – demand for housing will penetrate to staggering proportions. We all see production opportunities in 2025, 2026 and 2027, which are not that far off.
So in some of the conversations I’ve been involved in, the bigger you are, whether you’re new america funding or a NFM or a sierra pacific, the conversation was: How do we triple production without adding any overhead? What do we need to do? Whether it’s outsourcing or leveraging artificial intelligence – I think artificial intelligence is really the next innovation. The focus of many of us in business over the past decade has been: Can we reduce production costs by outsourcing to India or the Philippines?
CC: That’s been a theme of the show for years. How do we build an accordion cost structure that can fluctuate with volume, whether it’s purchase volume or refinancing volume, that doesn’t mean you have to go out and do it with thousands of new processors, underwriters and Operational talent to scale every time we see volume flowing in?
RR: The thing is, outsourcing doesn’t change the paradigm because you’re still throwing in hundreds of people. I think AFN has 400 or 500 employees in the Philippines. I mean, we’ll expand it very easily. I know it’s easy to achieve this with cross country, but again, you don’t really change the paradigm, you just add cheaper labor. You add the same corpses, but they don’t live in the United States. You’re applying incredible talent at a third or 25% of the cost in the Philippines. I think artificial intelligence completely breaks this paradigm.
CC: How is artificial intelligence different? How must the mindset of housing managers change to truly create a better cost structure?
RR: Let me chime in and say I disagree with you that it doesn’t increase efficiency because it increases efficiency – the problem is the regulatory environment, which is not really built for automation. It’s not built around how to streamline compliance efficiency beyond auditing. OCR solutions were a big thing in 2008, 2009, 2010, 2011. But if it’s not 100% correct, then it’s really worthless because you still have to go back and preview everything. It’s almost like Siri – if I was texting while driving today, it really wouldn’t be that helpful. Because now I’m proofreading everything from speech to text mistranslations. Now I may be at greater risk when driving.
So I would say what technology does is it provides tremendous efficiencies by design…it has nothing to do with our ability to close loans faster. Artificial intelligence can change this.
