Last month, I wrote about the 30-year anniversary of the conventional automated underwriting systems; Fannie Mae’s Desktop Underwriter® (DU®) and Freddie Mac’s Loan Product Advisor® (LPA®) and the era of mortgage lending they continue to define. When these systems were introduced, they were widely viewed as the most revolutionary technology the mortgage industry had ever experienced. That statement mostly still holds true today. But it raises an important question:

What exactly made it revolutionary wasn’t speed alone, but trust.

I recently found and read an article that documented the early design and thinking behind DU® as an automated mortgage underwriting expert system. Reading it nearly three decades later is both fascinating and instructive, not only for what it explains about the past, but for what it quietly suggests about the technological future of the mortgage lending industry.

Desktop Underwriter® as an “Expert System”

What many people forget is that DU® was never positioned as simple automation. According to the article, Desktop Underwriter® was designed as an expert system, a class of artificial intelligence technology intended to replicate the decision-making logic of human experts. In the case of DU®, those experts were seasoned mortgage underwriters. Rather than replacing underwriting judgment, DU® attempted to encode it:

• Agency eligibility rules

• Credit policy interpretations

• Risk layering logic

• Compensating factor evaluation

• Consistency across borrower scenarios

This was not machine learning as we think of it today. It was rules-based intelligence, carefully modeled from Fannie Mae’s underwriting guidelines and refined through years of expert input and performance feedback. That distinction matters.

Why This Was Revolutionary in 1995

At the time, mortgage underwriting was:

• Highly manual

• Dependent on individual interpretation

• Difficult to scale

• Inconsistent across lenders and geographies

DU® introduced something fundamentally new: a centralized, standardized decisioning engine that applied the same underwriting logic to every loan, every time.

The article highlights several breakthrough characteristics:

1. Separation of Decision Logic Workflow

DU® isolated underwriting decision logic from loan processing mechanics allowing policy changes without rebuilding systems.

2. Transparent, Explainable Outcomes

DU® generated findings that could be reviewed, audited, and trusted, which is a critical requirement for secondary market acceptance.

3. Consistency and Fairness

By applying the same rules universally, DU® reduced variability and subjectivity in credit decisions.

4. Speed Without Sacrificing Discipline

What once took days could now happen in minutes, or in the case of the article, in as little as 15 seconds, without abandoning underwriting accuracy.

These are not small achievements. They reshaped mortgage underwriting, loan operations, risk management, and capital markets execution. Entire system ecosystems were built around the automated underwriting decision as it became the focal point of originating a loan file. That is precisely why DU® was so revolutionary.

Why That Design Still Matters Today

What stands out most when reading that article is how intentional the architects of DU® were about control, auditability, and accountability. Even then, the designers understood something that still applies today: Automation without governance introduces risk.

DU® was built to:

• Support regulatory scrutiny

• Enable post-decision audit

• Provide repeatable outcomes

• Grow at scale

• Maintain human oversight

Those principles are precisely why DU® earned trust and why it still anchors agency lending three decades later.

So What’s the Next Revolution?

Today, the obvious answer is artificial intelligence. AI is already reshaping many industries, and mortgage lending is no exception. But AI raises important questions that feel strikingly familiar to the concerns addressed in the original design of DU®:

• Should AI make the final underwriting decision?

• How do we prevent hallucinations in credit decisions?

• How do we ensure fairness, consistency, and explainability?

• How do lenders audit and defend decisions made by probabilistic models?

• How do we align innovation with regulatory and investor expectations?

AI is only as good as its inputs, its constraints, and the framework in which it operates. To illustrate this simply, while writing this article I conducted a small experiment. With no custom prompt or guidance, I asked an AI assistant to generate a summary of the article I read. It did so almost instantly and the summary was excellent. I then closed the document, reopened it, and asked the exact same question again. The AI produced another strong summary, and yet the two summaries were not identical, even though the source material and the request were unchanged. That variability is not a flaw; it’s the nature of probabilistic AI. But it highlights an important consideration for mortgage underwriting: without structure, guardrails, and governance, AI will not consistently produce the same output for the same input. In a lending environment that demands consistency, auditability, and fairness, that distinction matters.

A More Likely Evolution: Rules + Intelligence

The more compelling future may not be AI instead of rules — but AI informed by a next-generation rules-based system.

Imagine:

• A configurable underwriting rules engine that evaluates Non-QM, Jumbo, HELOCs, second liens, BPL, and portfolio products.

• Real-time product performance feedback.

• With AI layered on top to:

In this model:

• Rules ensure consistency, auditability, and fairness

• AI provides learning, guidance, and insight

• Humans retain decision authority

That feels much closer to the spirit of what DU® originally set out to accomplish, applied to a more complex, diverse, and dynamic lending ecosystem.

Looking Back to See Forward

Desktop Underwriter® was revolutionary not because it was flashy, but because it was disciplined, intentional, and trustworthy. As the industry looks toward its next wave of innovation, those same attributes will matter just as much as speed or sophistication. Because sometimes, the best way to understand the future of mortgage technology is to revisit the moments when we got it right the first time.

About the Author

Mike Lewis is President of LoanPASS and a seasoned veteran in the mortgage lending and mortgage technology industries. With more than 28 years of experience spanning loan operations, systems implementation, product strategy, and enterprise technology leadership, Mike has worked with banks, independent mortgage banks, and software firms to modernize mortgage origination and loan decisioning processes.

Over the years, he has designed, configured, integrated, and led the deployment of complex mortgage systems and platforms, helping hundreds of lenders improve operational efficiency and adapt to changing market dynamics.

As President of LoanPASS, Mike leads the company’s mission to deliver modern, configurable rules-based pricing and decision engine automation that empower lenders to compete in an increasingly dynamic market. Under his leadership, LoanPASS has achieved rapid growth, industry recognition, and broad adoption across diverse loan products and lending verticals.

Mike holds a BS in Health Promotion from Brigham Young University and an MBA in E-Business. Away from work, he enjoys family life, looking for ways to mentor, teach, learn and grow through volunteer service, and exploring how to make the industry better than he found it through the effective use and adoption of technology.