The new standard for credit decisioning
Why Mike Lewis believes non-agency lending needs an underwriting standard built on deterministic rules - with AI helping lenders improve the system, not deciding the loan approval.
Automated underwriting is so embedded in conventional mortgage lending that the decision itself can feel almost invisible. Data goes in, an AUS response comes back, and an industry built around agency execution knows how to act on it. Mike Lewis sees a very different operating reality across non-agency and portfolio lending, where credit policy can still live in PDFs, shared drives and email threads. Here a complex loan may depend on manual reviews before anyone can say with confidence whether it fits.
For Lewis, that gap explains why the newest chapter of LoanPASS feels less like a product extension and more like a return to the company's original thesis. When he began his work with Bill Roy in January 2020, Lewis says the foundational idea was a non-agency automated underwriting engine. The company went to market first with pricing because pricing was a practical layer on which the broader architecture could be built. Underwriting was always further down the road.
Our goal is to become the standard - just as in the conventional space.
- Mike Lewis
Now the road has reached that point. But the most important design choice may be what LoanPASS did not try to automate. Lewis says the Automated Underwriting Engine is intentionally rules based. Consistent and deterministic so it can be audited. AI can sit above that system - helping lenders analyze, refine and improve their policies - but it should not guess its way to an approval. In an industry racing to add AI everywhere, LoanPASS is making a different argument: the final credit decision should remain explainable by design.
The standard that never formed
Lewis has spent nearly three decades on the technology side of mortgage lending, beginning with Flagstar Bank's management development program in 1998 and later working with technology partners including Mortgage Cadence and Encompass as well as lenders of different sizes. Across that career, he has watched conventional lending organize itself around familiar automated underwriting responses. He points to DU and LP as standards that have shaped agency lending for decades.
Meanwhile, the non-agency side never developed the same common operating model. Products can vary by investor, portfolio strategy, channel and credit appetite, which makes a single static credit box a poor fit. LoanPASS is not attempting to recreate an agency AUS around one set of rules with the Credit
We intend to have LoanPASS become the standard as our clients develop their own nonagency AUS engine for their products.
- Mike Lewis
Decision Engine. Lewis describes the goal as giving each client the ability to build its own non-agency AUS around the products and policies it actually offers.
That makes the word standard more nuanced. Lewis is not claiming that LoanPASS has already become the industry's non-agency standard; he is describing the outcome the company wants to earn. The ambition is for configurable decisioning itself to become routine - a common way for lenders to translate their own credit policies into repeatable decisions without forcing every product into an agency-shaped process.
Underwriting still lives in documents
The problem begins with where credit policy lives. Lewis describes underwriting environments in which PDF guidelines sit on shared drives and policy updates are distributed by email. The knowledge exists, but it is separated from the transaction. Someone has to find the right document, interpret the language, reconcile the latest version and determine how it applies to the loan in front of them.
Today, underwriting is happening against PDF documents that live on shared drives within these organizations.
-Mike Lewis
That document-centered workflow becomes more consequential when a lender offers portfolio products. Lewis says those loans can require internal reviews, committees and manual underwriting while agency products move through an established technology stack for pricing, decisioning, documents and compliance. In his view, that difference creates an operational pull toward conventional volume even when a lender may see stronger economics in parts of its portfolio business.
The Automated Underwriting Engine is intended to move the decision closer to the point of origination. Once the relevant data is captured, the user can check the loan against the configured rules, determine whether it conforms to the guidelines, see whether it is approved and identify the conditions that must be satisfied to support that approval. The underwriter does not disappear; the process changes from searching for policy to working from a structured decision and a defined set of conditions.
Pricing was the first layer
The sequence matters because automated underwriting was not the first product LoanPASS brought to market. Lewis says the company began with the product and pricing engine because it was a useful commercial product and part of the larger strategy. Pricing allowed the team to encode lender and product logic in a configurable environment before asking that same architecture to carry the additional weight of an underwriting decision.
Understanding the people that interact with the technology was our starting point.
-Mike Lewis
From there, the company could extend the framework into underwriting. The progression is less about bolting an AUS onto a pricing tool than about using a common rules foundation across decisions that have traditionally lived in separate systems. Price, eligibility and credit approval become related expressions of the lender's defined product policy rather than disconnected interpretations of the same guideline set.
Lewis describes the underwriting interaction as a straightforward sequence. The value is not in making the flow visually complicated; it is in making each step explicit and repeatable:
Once the rules return an approval and the associated conditions, LoanPASS can coexist with other technologies that help validate the documentation used to satisfy those conditions. Lewis is deliberate about that boundary. The engine does not need to perform every downstream task itself. Its job is to create and maintain the approval logic and conditioning so that the rest of the ecosystem has a consistent decision to work from.
Built by operators
Building that kind of engine required more than software engineering. Lewis describes a team with experience as loan officers, underwriters, secondary marketing professionals, capital-markets staff and technology administrators. His own background includes system administration and project sponsorship as well as years spent operating origination and servicing platforms. The starting point, he says, was understanding both the technology and the people who would have to interact with it.
The team then brought beta customers into the development process and asked practical questions: What features must exist? What value must the product create for a lender to see a return on using it? Vendor partners were part of the conversation as well because the mortgage technology ecosystem depends on systems handing work to one another. LoanPASS did not have to be the best tool for every function; it had to know where its responsibility began and ended.
We want the engine to be the brain of the organization as it pertains to their product offerings.
-Mike Lewis
That operator perspective is visible in the product philosophy. The goal is not automation for its own sake. It is to convert policy knowledge into something the people inside a lending organization can actually use, manage and trust. That means a decision engine has to account for the realities of product creation, pricing, underwriting, conditions, lock management and the handoffs that happen around them.
The engine as the brain
Lewis uses a simple metaphor for the role LoanPASS wants to play: the brain of the organization as it pertains to product offerings. That does not mean the platform replaces every system around a loan. It means the configured product guidelines become a central source of decision logic that can support the different products and channels a client chooses to offer.
He contrasts that model with technologies that describe themselves as automated underwriting but primarily validate documents against conditions that already exist. LoanPASS is focused earlier in the decision chain - on creating and maintaining the approval and conditioning logic itself. In Lewis's framing, the Automated Underwriting Engine should be able to tell a lender why a loan fits, what conditions support that result and how the approval aligns with the pricing that has been set.
Lewis also sees the concept extending enterprise-wide rather than remaining confined to one mortgage product. The important idea is not that every credit product is identical; it is that an institution can use a configurable rules framework to express different policies without rebuilding the decision process from “We want the engine to be the brain of the organization as it pertains to their product offerings.” -Mike Lewis scratch each time. The architecture follows the lender's products instead of requiring the products to conform to a single hard-coded model.
A system with boundaries
The breadth of that ambition makes boundaries important. Lewis is clear about what LoanPASS does not intend to become. The company is not positioning the engine as the manager of a lender's secondary department, and it does not provide hedging advisory through the solution. The focus is guideline approval and lock management, while specialized partners can remain responsible for the work they do best.
That division of labor matters in a mortgage stack where no single platform can realistically own every function. Lewis describes partnerships with technologies that can help automate the validation and sign-off of conditions after the decision engine has produced them. In that model, the value of the Automated Underwriting Engine is not that it eliminates the ecosystem; it gives the ecosystem a more structured decision to orbit around
The same discipline also applies to the lender's internal control. LoanPASS can encode the policies the institution defines, but the institution remains responsible for those policies and for how the technology is used. The engine is powerful because it can make a configured rule repeatable. It is not a substitute for deciding what the rule should be in the first place.
AI belongs above the decision
That distinction becomes especially important as AI moves deeper into financial services technology. Lewis says LoanPASS made a conscious decision not to have AI generate loan conditions on a loan-by-loan basis. This is a rule-based engine designed to be consistent, deterministic and auditable. If an auditor needs to evaluate how a product works or how a guideline affects a borrower segment, the decision path should be traceable to the rules that were configured.
You don't want your AI engine to guess on a loan decision, ever.
-Mike Lewis
Lewis is not anti-AI. He sees substantial opportunity for AI to sit on top of an AUS and help lenders learn from the information their decision engine produces. It could help identify where guidelines might be adjusted, where a lender could pursue more volume with an acceptable risk profile, or where product strategy could be refined. Those are consultative and coaching roles - ways to help people make better-informed choices about the credit policies they want to offer.
The line is the final approval. AI is probabilistic; credit decisions need repeatability and control. Lewis argues that human intervention must remain in the process so AI is not independently deciding who receives an approval. The distinction is subtle but consequential: use AI to interrogate the system, improve the system and advise the people responsible for it, while letting deterministic rules govern the production decision.
Start with one win
For credit leaders considering an AUS for non-agency lending, Lewis recommends a deliberately narrow start. His advice is to find an early win rather than attempt to convert an entire organization at once. Select one area, configure the rules, evaluate the results and validate the value before expanding. Adoption becomes a series of proven steps instead of a single enterprise-wide leap.
That phased approach also gives teams time to learn what it means to own a configurable engine. Lewis says clients can manage their engines themselves without relying on LoanPASS for every change. Once product guidelines are programmed into the system, the lender can continue defining new offerings and modifying the engine as its business changes.
The strategic benefit is responsiveness. Markets move, credit appetite changes and product opportunities appear faster than traditional technology projects can sometimes accommodate. Lewis's vision is for the decision framework to move at the speed at which the lender can define its policy - not at the speed of a custom development queue for every product change.
That is why the early win matters. A successful first implementation does more than prove that a rule can be automated. It teaches the organization how to use, govern and extend the engine. Lewis believes that experience is what allows lenders to recognize the broader growth potential in the tool they now control.
People are still in the loop
For all the emphasis on automation, Lewis repeatedly comes back to people. He expects the AUS to give lenders a better toolset and help them make more confident and consistent decisions, but he does not describe the goal as removing expertise from lending. In fact, he says some of the industry's biggest decisions still require knowledgeable people behind the technology.
Some of the industry's biggest decisions still require knowledgeable people behind the technology.
-Mike Lewis
That is an important counterweight to the language of automation. A deterministic engine can apply a defined rule consistently; it cannot decide what a lender's risk appetite ought to be, how a new market should be entered or when a policy deserves reconsideration. Those remain human decisions. Technology can make the execution of that judgment more reliable, but the judgment itself still belongs to the institution and the people responsible for it.
The question Lewis wants the market to ask
Lewis is candid that the Automated Underwriting Engine is still at the beginning of its market story. He talks about continued development, enhancements and the feedback LoanPASS receives from clients and partners across the lending supply chain. The product will keep changing as those users test where the engine creates value and where the surrounding ecosystem needs to connect more effectively.
Have you run your LoanPASS?
-Mike Lewis
The longer-term ambition is much larger than a feature release. Lewis wants the Automated Underwriting Engine to become to non-agency lending what other familiar AUS responses became to agency lending: a routine part of the workflow and a shared expectation. That is a vision, not an established fact, and it will only become real if lenders choose to adopt it.
His shorthand for that future is a question. Someday, he hopes, when someone is discussing a non-agency loan, the natural response will be: 'Have you run your LoanPASS?' The phrase captures the scale of the aspiration because it assumes the technology has moved beyond being a product that must be explained and become a behavior the market simply expects.
Lewis believes reaching that point would leave the industry better than LoanPASS found it. He describes nonQM and non-agency as a major growth opportunity and sees automated decisioning as infrastructure that can help lenders participate in that market with more confidence. The measure of success, in his telling, is not whether the software sounds sophisticated. It is whether better structure helps lenders make better decisions.
Confidence created through configurable and auditable decision logic
Lewis's career has been spent close to the systems that mortgage professionals use every day. That history makes his view of automation notably practical. The value of a decision engine is not that it acts on its own. The value is that it can hold a lender's defined policy, apply it the same way repeatedly and give the people using it a clearer basis for action.
LoanPASS took an indirect path to this moment. The original idea was a non-agency AUS, but the company commercialized pricing first, learned from customers and partners, and continued building toward underwriting. That sequence now gives Lewis a chance to test the thesis that began the venture: complex non-agency policy can be translated from documents and institutional knowledge into configurable, auditable decision logic.
If that works at scale, Lewis believes lenders can make decisions more consistently, manage their own product logic more confidently and expand into opportunities that are difficult to support with manual processes alone. AI can help them see where the system might improve. People can decide what should change. The rules engine can make the resulting policy repeatable.
That is the more durable promise behind automated underwriting: confidence in how a decision was made - with no guessing when it matters most.
ABOUT THE AUTHOR
Laura Kay Sheely is a two-time Amazon best-selling author with 25 years of experience researching and writing about home finance and the technology companies shaping how consumers finance, buy and own homes. Through The 52 Academy Press, she explores the people, ideas and innovations changing the mortgage industry.