Home >> Daily Dose >> The Evolving Digital Frontier
Print This Post Print This Post

The Evolving Digital Frontier

This piece originally appeared in the December 2023 edition of MortgagePoint magazine, online now.

As we stand on the precipice of 2024, the technological landscape is rapidly evolving, with some trends promising transformative effects on the mortgage industry.

Here’s a deep dive into the pioneering advancements that are poised to shape our digital future in 2024 and beyond.

Generative AI for the Mortgage Industry
Foundation models are massive neural networks trained on petabytes of unlabeled data through unsupervised learning methods. In other words, take a deep learning model, feed it a library of data, and let it create datasets of what it believes to be related/similar data.

When questioned, it then samples these datasets to craft a response that maintains the core integrity of the dataset.

Your basic generative model with encoders (dataset creators) and decoders (dataset samplers) have been around for years and are used extensively in financial modeling, statistics, etc. What took this to the next level is Google Labs’ Transformers—a mechanism to parallel process text so a general model could be created and then fine-tuned for a specific task.

Earlier language interpreters, like Recurring Neural Networks, processed one word at a time. Transformers, on the other hand, not only process entire sentences but have the capability to understand grammatical rules, positional context, and even relationship context, thereby far expanding the scope of data it can understand without human intervention.

This, of course, begs the question: what trust can we put in AI-generated output? With large foundational models, this can still be a gray area. Earlier ChatGPT models confidently spouted facts as fiction, but this will improve with time as technology scales. Prompt Engineering and Reinforcement Learning from Human Feedback are human supervised learning methods that are gaining in popularity to enhance AI’s black box learning styles.

These large foundation models can cost a fortune to train—for example, the training for Chat GPT 4 cost an estimated $100 million. But there is an advent of smaller, nimble models that cost significantly less that can be focused on a specific knowledge base. Think protein identification for cancer research, smart grid configuration and maintenance, or supply chain management.

For the mortgage industry, this opens up a slew of use cases—portfolio/pool analyses for the secondary market, default and foreclosure support to gain insights into profitability, and fraud prevention in Know Your Customer and Anti-Money Laundering spectrums.

It could even mean enhancements of existing products, for example, adding even more insight to an already robust Ask Poli. In addition to links to Fannie’s guides, perhaps we’ll see a summarized version that reads out as Steve Irwin. (“Crikey! That LTV seems awfully high for that credit score. You may want to approach carefully so as not to startle the compliance guidelines.”)

For today, it’s using AI-enhanced Google for more insightful searches or ChatGPT to fix code (maybe an Excel macro?). Baby steps to a (hopefully) well-regulated, AI-powered future.

Custom Data Endpoints
More and more institutions are getting into the practice of exposing specific data points through APIs. In the past, information was shared as paper only, which was clunky, slow, and insecure. With APIs becoming more prevalent, institutions can share data in real-time with confidence in who they are sharing it with and exactly what is being shared.

For the mortgage industry, this can mean not needing to strain your eyes over the fine print of a 4506T or an appraisal while still getting all the information need[1]ed to make an informed decision.

For a consumer, this means not having to share more information than needed. Take a corporate tax return submitted as part of a personal loan qualification, for example. The return may include information about business partners not tied to the deal at hand, or a divorce decree that exposes the ex-spouse’s information.

All currently required documents for the loan origination process may contain unnecessary data points. Back in 2018, as part of a privacy blockchain article, I wrote about zero-knowledge proofs, a method by which information being requested can link back to a verifying person or institution without conveying any additional information except that about which they are being asked. One of the examples I gave then was of a car rental agency needing to verify if you were over the age limit.

They don’t need to know your exact age, your birthday, etc.—just a yes/no response from a trusted agency. This is the future we should expect.

Privacy and Transparency
My data is my data, not our data—and definitely not your data. We say this out loud, we may even believe it at times, but in our heart of hearts, we know that privacy is but an illusion. Data brokers have been selling our most private thoughts, our late-night Google searches, and our locations for years. We are a willing/unwilling open e-book.

But government agencies are stepping up to put an end to this—or at least to provide guard rails. Last year, the Federal Trade Commission sued Kochava over selling sensitive location data. The CFPB wants to hold data brokers more accountable. Even the turbulent economic environment of 2022 did not take away from privacy spending by institutions.

The average spending was $2.7 million, up significantly from $1.2 million just three years ago.

Data management—the entire spectrum of data scientists, data analysts, data reporters, data handlers—is a continuously growing area. Data mesh—which is the concept of decentralizing data, so we have specific data owners for specific data sets, removing silos and constraints—allows for data to be seen as a business product, thereby controlled through business-level access yet structured enough for easy self-service. All of these are moves to not only protect our data but also provide more transparency on what is known about us and by whom.

I believe this uptick in vulnerability assessments, security controls, and data management will continue and only get stronger. Expect nimble, narrow-AI powered solutions.

The Thinking Machine
Between sociopolitical unrest, economic turmoil, COVID-19 variants, and deepfakes, who knows what 2024 will unravel for us? The one constant, like time, is the digital bits that power our world—always a zero or a one—relentlessly blinking away, connecting us, disconnecting us, but always ticking on.

The views and opinions expressed in this article are those of the author and do not necessarily reflect or represent the views, policy, or position of Planet Home Lending, LLC.

About Author: Aneeza Haleem

For nearly two decades, Aneeza Haleem, VP of Technology at Planet Home Lending, has leveraged technology to bring business visions to life in the mortgage banking industry. An innovator and entrepreneur, she manages and implements large-scale, complex technical systems for Planet. Before joining Planet, she managed new ventures for Cognizant across clients in the mortgage industry. Having lived and worked in multiple countries, Haleem brings a cross-cultural, cross-functional leadership focus to all her efforts. Her blogs have appeared in MBA Newslink and BAI Insights, where she has written about weaving inclusion and diversity into the fabric of our industry. In 2020, she was named to Progress in Lending’s Most Powerful Women in FinTech list.

Check Also

Survey: Homeownership Remains Elusive for Baby Boomer Renters

A recent look into housing affordability by NeighborWorks America has found that three in five long-term baby boomer renters feel homeownership remains unattainable.