top of page

AI in Mortgage Banking: Underwriting & The Long and Winding Road to Success (Part III)

by Ruth Lee, CMB

A paved highway through the woods with squiggles and loops showing a difficult process
Unnecessarily Long and Windy Road
Mortgage underwriting has long been a field ripe for automation, yet the incorporation of Artificial Intelligence (AI) demands a thoughtful, strategic approach.

In my last blog, we talked about FinTech trends in mortgage banking. Now, let's zoom in on underwriting. It's the backbone of origination, where the magic (and the headaches) happen. No, we’re not close to showing seasoned underwriters the door—they're still the heart and soul of credit risk, but AI is stepping in, not only to streamline time-consuming tasks but also to enhance decision-making, complementing the irreplaceable expertise of seasoned underwriters.

Early Automation Efforts

The transformation of mortgage banking into a domain where automation is pivotal has been a study of careful, deliberate change rather than sudden upheaval. Thinking back to securitization's emergence in the 1970s, it's evident how our dealings with investors, regulators, and borrowers were profoundly altered. The shift necessitated new technological requirements emphasizing standardized data and streamlined processes. Spoiler alert - current advancements with AI in underwriting presage the technological requirements of the next frontier in mortgage banking - tokenization, disintermediation, and globalization.

It's essential to acknowledge the understated yet impactful advances that have helped dismantle longstanding technological barriers. So I wanted to acknowledge each incremental but pivotal era in automating mortgage underwriting.

Before the Digital Dawn—The Analog Epoch (Pre-1980s):  In the days before digital transformation took hold, the mortgage banking landscape was virtually submerged under a flood of paperwork. Documents painstakingly produced in triplicate, the slow dance of snail mail, and the ubiquitous yellow notepads constituted the essential toolkit. To the modern observer, these practices might seem as archaic as using stone tablets and chisels, particularly when we consider the streamlined efficiency of today's productivity apps at our disposal. Data "analysis" was decidedly low-tech, driven more by instinct and copious amounts of caffeine than any systematic approach.

The Advent of the Beige Box—Dial-Up to Digital (the 1980s to mid-1990s): In this period, the fax machine reigned supreme, offering a glimpse into a future where communication could transcend the limitations of physical mail. The personal computer, clad in its iconic beige, emerged as the cornerstone of office productivity. Word processors began to phase out the traditional typewriter, setting the stage for software like WordPerfect and Microsoft Word to revolutionize document handling. Meanwhile, Excel introduced a new way to approach data analysis, moving us beyond reliance on intuition to a more structured, spreadsheet-driven method. The emergence of loan origination systems signaled the start of a shift away from the overwhelming paper processes of old.

The World Wide Web Weaves Its Web—The Enlightenment (The late 1990s to 2000s): As the internet wove its web, it did more than democratize access to information and the occasional cat video—it changed how we took mortgage applications. Software systems burgeoned, morphing with increasing complexity and prowess. The scene was set for major transformation with the launch of Fannie Mae's Desktop Underwriter (DU) and Freddie Mac's Loan Prospector (LP). These Automated Underwriting Systems (AUS) were transformative, where borrower information could be swiftly inputted and assessed, yielding near-instantaneous loan approval recommendations. This wasn't just a shift; it challenged the status quo.

And with any advancement in mortgage banking, we were soon to point out the limitations. These pioneering AUS versions were tethered to a purely rule-based logic—rigid, unyielding, and devoid of the capacity to learn from past decisions. While groundbreaking, their initial framework was a double-edged sword; they lacked the dynamic adaptability that modern AI and ML technologies boast today. This limitation is noteworthy, not as a flaw, but as a pivotal stepping stone, laying down the initial tracks for the journey towards more nuanced, intelligent systems capable of learning and adapting.

The Era of Talking Software—The Age of Reason (Late 2000s to 2010s): With MISMO acting like the Rosetta Stone of mortgage data, disparate systems suddenly could understand one another, making sense of tangled data formats. APIs became the new lingua franca, allowing mature and nascent software solutions to speak, creating a more integrated and seamless process for confirming data and assessing credit risk. This era underscored the importance of communication and integration among disparate systems... another critical stepping stone.

Robotic Process Automation (RPA) - 2010s: When Robotic Process Automation (RPA) came on the mortgage scene, it was billed as though it would give lenders superpowers—tasks that used to be eternal, like retrieving purchase advice and starting the double and sometimes triple data entry (because who needs an accounting platform integrated into a mortgage business?), suddenly became lightning fast. RPA was the industry's new caffeine rush.

While RPA brought unprecedented speed and efficiency to routine tasks, its rigid, rule-based logic (much like early AUS systems) struggled with the complex, often unpredictable scenarios that mortgage underwriting presents. Plus, they had the learning curve of a flat line—no adapting, no evolving, just doing the same old dance, even when the music changed. This realization sparked a pivotal shift towards embracing more sophisticated technologies capable of not just performing tasks, but understanding and adapting to the intricacies of mortgage underwriting.

Towards Cognitive Automation: Today, progressive companies are exploring more sophisticated solutions to incorporate artificial intelligence (AI) and machine learning (ML) into the underwriting process. These technologies promise not only to automate tasks but also to bring a new level of cognitive understanding and adaptability. By learning from vast datasets and making informed decisions and recommendations based on patterns and trends, AI and ML are set to address the shortcomings of early RPA and AUS technologies, offering a more dynamic, efficient, and accurate approach to mortgage underwriting.

An excellent illustration of this advanced approach is Marcus, Goldman Sachs' online personal loan service, which leverages AI and ML to evaluate credit risk and decide on loan approvals meticulously. It analyzes a comprehensive dataset, including unconventional data, to assess a borrower's creditworthiness more precisely.

Here, we're not just talking about understanding borrowers; we're about deciphering the financial DNA that makes each person unique. It's a financial detective adept at spotting the clues hidden in plain sight, revealing the richness and complexity of individual money habits that a human might miss. This ability to read between the lines sets apart a truly insightful lending strategy capable of navigating the labyrinth of human financial behavior with elegance and insight.

Another significant advancement brought about by AI in mortgage underwriting is its ability to detect and mitigate potential fraud risks. AI systems can scrutinize patterns and anomalies in application data that human underwriters might miss, offering a formidable defense against fraud. These systems continuously learn from new data, improving their fraud detection capabilities and adapting to evolving fraudulent tactics. This not only protects lenders from financial losses but contributes to a more secure and trustworthy lending ecosystem.

Embracing the Future: AI in Mortgage Underwriting

The evolution from Robotic Process Automation (RPA) to Artificial Intelligence (AI) and Machine Learning (ML) signifies a profound paradigm shift in mortgage underwriting. This transition heralds a future where the process is not just accelerated and more efficient but also inherently more intelligent, personalized, and equitable. As the mortgage banking industry embraces AI and ML, it stands on the brink of a new era where the nuanced art and rigorous science of underwriting converge, guaranteeing every borrower a mortgage experience that is fair, transparent, and tailored to their unique needs.

Risk, Fairness, and Beyond

The advancement of AI/ML brings to the forefront a spectrum of considerations for mortgage banking. Implementing these technologies requires a delicate balance, emphasizing the elimination of historical biases, enhancing decision-making transparency, and ensuring absolute accountability. Within a framework of evolving best practices, the industry aims to address these challenges through several strategic approaches:

  • Transparency as a Cornerstone: Integrating AI into the mortgage underwriting process necessitates a demystification of the algorithms' decision-making processes. It involves clarifying the considerations AI systems prioritize and the rationale behind their conclusions. This transparency is pivotal, cultivating a culture of trust among applicants and regulators, encouraging them to engage with and understand the AI's operational logic. It further entails a commitment to using diverse data sets that reflect the broad spectrum of human experiences, thus preventing the perpetuation of existing biases.

  • The Imperative of Explainable AI (XAI): Adopting XAI practices is crucial. It allows for the decoding of AI's sophisticated algorithms into narratives that are comprehensible and relatable, bridging the gap between complex technology and human understanding.

But not to worry, there are tools and strategies to counteract bias within AI systems, such as:

  • Data Analysis and Pattern Recognition: Utilizing AI to detect and comprehend data patterns that may indicate bias.

  • Predictive Fairness Assessments: Leveraging AI to project outcomes and evaluate the fairness of these predictions across diverse demographics.

  • Feature Importance Analysis: Identifying and assessing the significance of variables considered by AI, ensuring they do not contribute to biased outcomes.

  • Bias Mitigation Algorithms: Implementing advanced algorithms aimed at reducing or eradicating bias within AI models.

  • Fairness-aware Modeling: Creating AI models that intrinsically prioritize fairness, striving for equitable outcomes for all users.

Expanding the Horizon: Efficiency, Fraud, and Risk Mitigation

Beyond fairness and transparency, the integration of AI/ML into mortgage underwriting also promises significant efficiency gains, alongside enhanced capabilities in fraud detection and risk mitigation:

  • Operational Efficiency: AI and ML streamline the underwriting process, automating routine tasks and analyses, which allows mortgage professionals to focus on more strategic activities. This not only accelerates the mortgage application process but also improves the overall customer experience.

  • Fraud Detection: With the ability to analyze vast quantities of data, AI systems can identify anomalous patterns indicative of fraudulent activity with unprecedented accuracy. This proactive approach to fraud detection safeguards both lenders and borrowers, contributing to a more secure mortgage ecosystem.

  • Risk Assessment and Management: AI’s predictive analytics offer a more nuanced assessment of risk by considering a wider array of variables, including some that may not be immediately evident to human underwriters. This results in more accurate risk profiles, better lending decisions, and ultimately, a more robust financial portfolio for lenders.

As we stand at the cusp of this transformative era, it's clear that AI and ML will play instrumental roles in shaping the future of mortgage banking. By harnessing these technologies' full potential, the industry is poised to offer not only more efficient and secure lending practices but also a more equitable and personalized borrowing experience. The journey ahead is one of continuous innovation, learning, and adaptation, as we redefine the boundaries of what's possible in mortgage underwriting.

To Access the Previous Parts of the Series, Click on the Image:

Revolutionizing Mortgage Banking with AI: A New Era Begins Part I

Five Top Trends in FinTech: AI in Mortgage Banking Series Part II

Industry Sources & Resources Used:

Disclaimer: This article was augmented with artificial intelligence (AI) to ensure comprehensive coverage of the latest FinTech trends impacting mortgage banking. While AI has contributed to the research and drafting process, the insights and analyses reflect the professional judgment and expertise of Ruth Lee, CMB, and have been reviewed for accuracy and relevance to our audience. Our commitment to providing high-quality, informative content remains paramount, and we embrace innovative technologies like AI to enhance our ability to deliver valuable insights to our readers.

54 views0 comments


Rated 0 out of 5 stars.
No ratings yet

Add a rating
bottom of page