Remove the work not the people
Why Bill Mitchell believes the best case for automated underwriting is not replacing underwriters, but removing the repetitive work that creates time, cost and inconsistency around the credit decision.
Bill Mitchell has a simple test for mortgage technology: Does it take complexity out of the process without taking control away from the lender? That principle becomes especially important in automated underwriting, where the promise of efficiency can easily be confused with the idea of handing the credit decision to a machine.
Mitchell rejects that framing. After nearly three decades across mortgage operations and lending technology, he sees the real target as the work around the decision - the manual guideline research, repeated calculations, overlays and rework that consume underwriter time before expertise can be applied where it matters most.
His argument for LoanPASS is therefore less about automation as spectacle and more about operating discipline. Make the rules predictable. Make the decision explainable. Run pricing and credit together. Use AI to assist judgment rather than become the judgment. Then measure success by how much unnecessary work disappears from the file.
Make the rules predictable. Make the decision explainable. Run pricing and credit together.
- Bill Mitchell
Find the niches
Mitchell began his mortgage industry career in 1999, gaining experience across help desk support, originations, servicing, and technology before spending nearly 20 years with Datatrac, OpenClose, Computershare, and eOriginal. Throughout his career, he progressed from sales leadership to the C-suite, helping transform startups into established enterprise mortgage technology companies. Each role reinforced a guiding business philosophy that continues to shape his approach to growth and innovation: "The riches are in the niches."
He has tried to align his career with what he calls leading-edge rather than bleeding-edge technology firms willing to innovate but still grounded in real operating needs. LoanPASS fit that pattern when Mike Lewis, Derek Long and Bill Roy described a plan to build automated underwriting for portfolio and non-agency lending.
Mitchell says the company has since grown 5x and that clients are running more than $60 billion in annual originations through LoanPASS' suite of products. The larger point he makes is about maturity: the platform is no longer just a concept being described to the market.
A market with too much manual work
Mitchell believes the timing is right because non-agency lending has expanded while the operating infrastructure has lagged. He cited 2026 market estimates totaling roughly $400 billion across non-QM, home equity, residential transition lending and jumbo. He uses those estimates to illustrate the scale of lending activity that sits outside the agency workflow.
He is equally clear that the objective is not to standardize non-QM itself. The differentiated guidelines are often the lender's secret sauce. A business-purpose lender, a DSCR lender and a home-equity lender may all want very different rules, and Mitchell sees that variation as something the technology should preserve rather than flatten.
The opportunity is to standardize the way those rules are executed: consistently, transparently and in a form that can move through the lending process without relying on a person to rediscover the policy on every file.
Differentiated guidelines are often the lender's secret sauce.
- Bill Mitchell
Time, cost and inconsistency
Mitchell reduces the underwriting problem to three recurring costs: time, expense and inconsistency. Before an underwriter can make a credit judgment, someone may need to research the applicable guidelines, perform calculations, apply overlays and reconcile the details that determine eligibility.
He says lenders have told LoanPASS that non-QM files can average about 1.5 loans underwritten per day per underwriter, compared with much higher throughput on the agency side. That productivity comparison is based on the company's research and client discussions, but it captures the operational gap Mitchell is trying to address.
His point is not that non-QM should be forced into an agency process. It is that experienced people should not have to spend their day on tasks that a deterministic engine can perform consistently before the file reaches the part that actually requires human analysis.
The more routine work the system can complete, the more the underwriter can concentrate on exceptions, borrower context and the pieces of the credit decision that do not reduce cleanly to a calculation or rule.
Experienced people should not have to spend their day on tasks that a deterministic engine can perform.
- Bill Mitchell
Do not automate the judgment
Mitchell focuses on one idea: LoanPASS is not trying to eliminate underwriters. The objective is to eliminate unnecessary underwriting work.
That distinction changes the implementation question. Instead of asking how close the lender can get to 100 percent automation, Mitchell asks how much work can be removed without compromising the credit decision. The boundary is not defined by a technology percentage; it is defined by where judgment begins to matter.
A rules engine can apply defined policy, calculate eligibility, return conditions and keep the logic consistent. The underwriter still owns the work that requires interpretation, accountability and experience. Automation becomes valuable because it protects judgment from being buried under repetitive tasks.
LoanPASS is not trying to eliminate underwriters. The objective is to eliminate unnecessary underwriting work.
- Bill Mitchell
AI should assist judgment
Mitchell is enthusiastic about AI but draws a firm line at the final credit decision. His preferred formulation is that AI should assist the judgment, not become the judgment.
He wants the production engine to remain deterministic because credit decisions need to be predictable, explainable, auditable and controlled. A lender should be able to understand why the rule fired and why the loan received the result it did.
AI can still be embedded in operations, help users work more efficiently and support the system around the decision. Mitchell objects to the idea of simply placing a probabilistic model on top of the problem and asking it to figure out the policy on its own.
Credit decisions need to be predictable, explainable, auditable and controlled.
- Bill Mitchell
Automation has to stay flexible
For Mitchell, the hardest product tradeoff is balancing automation with flexibility. Portfolio and non-agency lenders often compete through the details of their guidelines, so a rigid AUS can solve the wrong problem by making the process efficient only after the lender gives up what makes the product distinct.
A deterministic rules engine gives the lender another option. The policy can remain customized while the execution becomes repeatable. That is how Mitchell thinks LoanPASS can reduce underwriting drift without turning every client into the same credit model.
The company's integrations and investor relationships also matter because the decision has to live inside a larger workflow. Flexibility is only useful if the result can move back to the LOS, stay aligned with pricing and fit into the systems the lender already uses.
Pricing and credit on the same file
Mitchell sees the combination of pricing and automated underwriting as one of LoanPASS's most important differentiators. The system can run pricing while it runs the AUS file, allowing the economics and the credit decision to be evaluated in the same process.
Traditional stacks often separate those moments. A loan can qualify one way in underwriting while pricing logic is being managed elsewhere, creating another opportunity for data or assumptions to fall out of sync.
By tying the two together, the lender can see not only whether a loan fits the credit box but also how that fit relates to the product terms being offered. Mitchell argues that the combination changes the conversation from a point solution to a broader operating model around the credit box.
The system can run pricing while it runs the AUS file, allowing the economics and the credit decision to be evaluated in the same process.
- Bill Mitchell
Control belongs with the lender
The larger governance question is credit policy. Chief Credit Officers need confidence that the institution's policy is being applied consistently. CEOs and CFOs want to know whether volume can increase without costs rising at the same rate. Operations teams want fewer touches and less rework.
Mitchell believes a useful underwriting platform has to answer all three constituencies without taking policy ownership away from them. The software can execute the rules, but the lender still decides what the rules are and how much risk the organization is willing to accept.
That is why predictability and auditability matter as much as speed. An efficient black box would create a different set of risks. Mitchell's version of automation is valuable because the lender can explain the decision and retain control over the credit strategy behind it.
The common thread is control. Efficiency is valuable only if the institution can still explain what happened, identify which policy was applied and decide where the technology should stop. That is the standard Mitchell uses when he talks about reducing touches without weakening the credit decision.
Efficiency without surrender
The strongest idea Mitchell reminds lenders of is that efficiency should not require surrender. The lender should not have to give up a unique credit policy, the underwriter should not have to give up judgment, and the chief credit officer should not have to give up explainability just to move faster.
A rules engine changes the work by applying what the institution already decided. It can remove research, repeated calculations and inconsistent interpretations. It can run pricing alongside credit. It can feed conditions back into the workflow. But it does not decide what the institution believes about risk.
That separation is what allows automation to scale responsibly. Technology owns the repeatable work. People own the policy, the exceptions and the accountability.
Mitchell's view is practical because it avoids the false choice between a manual process and an autonomous machine. There is a large middle ground where software can remove friction while human judgment remains intact.
People own the policy, the exceptions and the accountability.
- Bill Mitchell
Remove the work, keep the decision
Mitchell has spent his career close to both lenders and lending systems, and he has seen how easily technology projects become focused on features instead of operating outcomes. Automated underwriting, in his telling, matters only if it changes what the people inside the organization have to do and does so without weakening the accountability attached to the credit decision.
The meaningful metric is not how automated the system sounds. It is how much unnecessary work disappears: fewer guideline searches, fewer repeated calculations, fewer inconsistent interpretations, fewer touches that add time without adding credit insight. If those tasks are absorbed by a deterministic engine, the underwriter can spend more of the day on the work that actually benefits from experience and judgment.
AI can help make that process smarter. Pricing can be aligned with the AUS. Integrations can move conditions through the stack. But the lender's credit policy still needs to belong to the lender, and the underwriter's judgment still needs a place in the process. The technology creates leverage by making the repeatable work truly repeatable.
That is the promise Mitchell wants the market to evaluate: remove the work that does not need a person, preserve the decision that does.
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.