A lot has changed about how people view artificial intelligence over the past year. Since the release of ChatGPT-4 in November 2022, almost every major industry has been abuzz with ideas for its uses and impact. Alongside talk of those possibilities, workers are grappling with ways their roles might be affected in the wake of technological disruption.
Understandably, the new era of generative AI has created excitement, confusion, anxiety, and wild predictions. Within the mortgage industry, lending professionals wonder how AI will impact their jobs and all its possible applications within lead generation, underwriting, borrower experience, and more.
What seems to be certain in the uncharted realm of AI is that this technology is here to stay. For lenders to stay ahead of the competition and market, they need to understand AI’s growing capabilities and connect with forward-thinking partners.
In a recent episode of the Clear to Close podcast, we spoke with someone who spends time with and works on artificial intelligence regularly: Maxwell’s very own Co-founder and CTO Rutul Davé. Here’s what he had to say about some commonly asked questions regarding AI in mortgage.
Why are new versions of generative AI so groundbreaking? How do they differ from past renditions?
According to Rutul, generative AI like ChatGPT and Google Bard is much more applicable in day-to-day settings because it can “learn” the complexities of human language and interaction via a massive amount of inputs.
“Think of it as a system that took a bunch of inputs and allows us to generate natural language responses from them,” explains Rutul. “Given a set of inputs, AI is able to determine the most statistically likely word that should come after the current word—it takes a bunch of prompts and generates a sentence.”
Still confused? A good comparison is Pandora: In a basic use of AI, Pandora is able to predict songs you might like based on the initial song you selected. Pandora uses around 100 parameters, including rhythm, pace, and melody, to choose songs similar to your choice. Now extrapolate those parameters to 1.7 trillion—the reported amount used by GPT-4—and it’s easy to see how this kind of generative AI provides such powerful predictive abilities across virtually any topic.
How might AI be applied within the mortgage industry?
Within mortgage, generative AI has potential to be leveraged in several capacities, from better segmentation and personalization within lead generation to providing faster answers to loan-related questions. In the underwriting process, for instance, AI might act as a sidekick to an underwriter, making data-informed decisions by taking all the information related to a borrower into account.
A low-hanging-fruit application for AI is within the borrower experience. Take a non-English speaking borrower, for instance. Studies show that their satisfaction is dramatically impacted by how much a lender is able to deliver an experience accessible in that borrower’s language. In this instance, AI could be used to generate the loan application and responses to their questions in real time in the language of their choice.
As another example, Maxwell recently released an AI-driven feature that automatically generates letters of explanation for borrower review. This feature is especially helpful for first-time home buyers, who might find writing a letter of explanation intimidating. Instead, AI delivers a draft to the borrower to hone, edit, and submit to the lender—easing the borrower burden and accelerating the lending timeline.
What about compliance? How can lenders ensure bias isn’t baked into AI models?
As with anything implemented in the mortgage industry, compliance needs to be a top concern, Rutul explains. That’s why humans need to be the driver of AI, no matter how it’s leveraged.
“The right application of AI in mortgage is to support the use case, not run on autopilot without any oversight,” says Rutul. “The ideal scenario is definitely not to eliminate people in mortgage, but to enhance their roles.”
Uses of AI within mortgage generally fall into two categories, according to Rutul: low-stakes and high-stakes. Low-stakes uses include efficiency-driving functionalities and communication-based features that improve the borrower experience, such as AI-generated responses to questions about the lending process and loan-related terms. High-stakes uses, meanwhile, include leveraging AI to facilitate the actual lending process, including making decisions in the underwriting process.
Within the high-stakes category especially, AI’s role needs to be closely monitored, with a close eye to the datapoints used to train the model.
“We often hear ‘trust but verify,’ but with generative AI, it’s more ‘verify then trust,'” says Rutul. “Crucially, we need to verify what data was used to train models in first place. That data needs to be incredibly transparent and used in a way where we know what leads to the outcome of the models.”
Recently, there’s been increased interest in regulation and plans for AI within mortgage, with Fed Vice Chair Michael Barr vowing to bring transparency to AI models and the CFPB exploring its role in making sure that transparency is followed. In other words: AI certainly won’t be implemented without serious thought to compliance and how inputted parameters impact lenders and borrowers.
What should lenders do now to begin taking advantage of AI?
As much as the capabilities of AI might be exciting, they can also be overwhelming—especially to the busy lending team who already concentrates their time and energy on connecting with borrowers and other important day-to-day tasks. For now, lenders curious about what AI can do for their businesses might find it valuable to dabble in the capabilities of publicly accessible tools like ChatGPT, reducing the intimidation factor and getting a feel for its possibilities within mortgage. Beyond exploration, though, lenders shouldn’t feel pressured to become technologists and figure out how to implement AI into their lending processes.
Instead, lenders should look to partner with technology providers who are making strides in AI and looking at ways to implement it for better efficiency, margins, and borrower experience.
“It shouldn’t be expected that the community bank or IMB figures out the latest in AI,” comments Rutul. “I see my role at Maxwell as offering the power of these innovations so lenders can continue to serve the communities they operate in.”