Can AI Replace Your Pricing Engine?


The Buy vs. Build Debate Is Back, But Is AI Really Ready to Run Your Pricing Operation?


Since the national MBA conference, I’ve been hearing a question from mortgage lenders: Can I just replace my SaaS pricing engine with AI?


It’s an understandable question.


With mortgage rates stubbornly remaining above 7%, lenders are looking everywhere for opportunities to reduce operating expenses, improve productivity, generate new revenue, and do more with fewer resources. At the same time, we’re being inundated with AI marketing that makes it sound like artificial intelligence is the silver bullet that can solve both sides of the P&L statement.


You can’t scroll through LinkedIn or your social media feed without seeing another post about how AI has solved the world’s problems, or another expert promising to teach you how to use AI to solve all of yours.


I’ve had a couple of prospective clients mention that they’re considering using AI to build their own mortgage pricing engine. I’ve also heard from existing clients who are interested in exploring what they could build internally.


It got me thinking: Has AI advanced enough to fundamentally change the traditional buy-versus-build equation for mortgage technology?



Building Your Own Technology Isn’t New


Building proprietary technology is nothing new in mortgage lending.


When I started in the industry back in 1998 at Flagstar Bank, its IT department had recently started building a proprietary loan origination system (LOS). At the time, there simply weren’t many commercially available SaaS solutions capable of supporting the complexity of Flagstar’s wholesale and correspondent mortgage operations.


I recall the project taking approximately three years and costing around $50 million to develop and deploy, with another 50–100 full-time employees dedicated to ongoing support, development, and maintenance.


That was a different era of technology.


With today’s advancements, including AI-assisted software development, a project of that size and scope could likely be completed with considerably less time, money, and personnel.


But here’s the distinction: building the initial technology is only part of the equation, and sometimes the easier part. You still have to support it, maintain it, integrate it with your other technology, adapt it to changing business and regulatory requirements, and continually invest in its evolution.


AI may significantly reduce the cost of building software. But, it doesn’t eliminate the responsibility of owning software.


So, does the rapid advancement of AI change the traditional buy-versus-build equation? And more specifically, is AI ready to become an enterprise mortgage pricing engine?


We decided to conduct a small experiment to find out.



A Simple Experiment: Can AI Actually Price a Mortgage Loan?


We started with something relatively straightforward.


We selected a random investor’s daily rate sheet containing approximately a dozen DSCR loan products. We also created a generic set of borrower inputs, without any personally identifiable information.


Using the same investor rate sheet and borrower scenario in the LoanPASS pricing engine, we received pricing results and approvals for at least half of the available products in approximately 1–2 seconds. Not bad.


That gave us our baseline.


Next, we took the exact same investor rate sheet in Excel format and the same borrower inputs in JSON format and provided them to one of the most capable AI models currently available.


The prompt was intentionally simple: price the borrower scenario against one specific product on the investor’s rate sheet and determine the resulting price, including all applicable adjustments.


No pricing engine. Just AI with a rate sheet and borrower data.


The results? Remarkable.


The AI model returned the exact same price as LoanPASS, including the adjustments listed on the investor’s rate sheet.


It got it right.


How long did it take?


Eleven minutes.


Since the first test was successful, we decided to raise the stakes.


We provided a second prompt asking the model to price all six approved products we got in LoanPASS with the same borrower inputs.


This time, the process took nearly 20 minutes, and all six pricing results were incorrect.


Not only were the individual prices wrong, but it was also difficult to establish exactly how the model arrived at each result.


We decided not to continue testing. For our purposes, we had learned what we needed to learn.



What Did We Actually Learn?


First, AI has advanced considerably, even within the past year.


In an earlier experiment conducted approximately a year ago, the most capable model we tested then couldn’t accurately price even a single product from an investor’s daily rate sheet using a static set of borrower inputs without user intervention and multiple prompts.


This time, it did.


That’s significant progress, and I don’t want to minimize it.


But it also highlighted two important challenges: speed and reliability.


Consider the borrower experience.


What borrower is going to sit on the phone with a loan officer for 11 minutes, perhaps discussing the weather while waiting, just to find out what interest rate they qualify for? And, you better hope it’s the product and rate the borrower wants.


How many borrowers are going to sit on a lender’s website for that long waiting for a mortgage pricing result?


Not many.


And when we introduced more complexity by asking the model to price multiple products, the results became unreliable.


In mortgage lending, an incorrect price isn’t simply an inconvenience. It can mean losing a borrower, quoting an unprofitable loan, misrepresenting an investor’s execution, or creating downstream problems that ultimately affect the lender’s ability to operate profitably.


The stakes are too high to treat pricing accuracy as an approximation.



But Couldn’t You Build Around Those Limitations?


Absolutely.


I can already hear the more technically inclined readers pointing out that there are ways to improve these results through better prompting, structured outputs, model orchestration, retrieval-augmented generation, custom tools, deterministic calculations, and other techniques.


They’re right.


And they’re actually making my point.


You can build a solution that uses AI to help address these limitations. But once you start building the necessary infrastructure, you are no longer simply asking AI to price a mortgage loan. You’re developing and maintaining a software platform.


Think about what happens when the business requirements change.

  • What happens when you need to add another borrower input or pricing variable?

  • What happens when you add another investor’s rate sheet? Or 40?

  • What happens when an investor publishes multiple rate sheets throughout the same day?

  • How do you manage historical rate sheets and ensure a loan is priced against the correct version?

  • How do you calculate relock pricing and worst-case execution?

  • How do you manage overlays, lock extensions, adjustments, exceptions, and investor-specific pricing rules?

  • How do you ensure the same inputs produce the same pricing result every time?

  • How do you explain and audit exactly how a price was calculated?


And those are just some of the requirements of a modern mortgage pricing engine.


The complexity isn’t necessarily in calculating one mortgage rate. It’s in managing thousands of pricing variables, hundreds of product configurations, multiple investors, constantly changing rate sheets, and the operational requirements that surround the entire mortgage pricing lifecycle.


You can absolutely use AI to help build and maintain that infrastructure. But you still need developers, architects, product experts, testing, governance, and ongoing support to keep it running. And, when it goes down, it’s up to you to get it fixed because your business depends on it.


At some point, you have to ask whether you’re building a pricing engine or building a software company.



AI and Purpose-Built Software Aren’t Mutually Exclusive


I don’t believe the future of mortgage technology is AI versus software.


I believe the future is AI working alongside purpose-built technology.


A mortgage pricing engine has a very specific responsibility: take structured borrower and loan information, apply the appropriate investor guidelines and pricing rules, and return a consistent, accurate, and explainable result.


That requires deterministic calculations, configurable business rules, reliable integrations, version control, and the ability to process transactions at the speed the borrower and loan officer expect.


AI can bring tremendous value alongside that foundation.


Imagine AI helping lenders:

  • Analyze investor rate sheets and identify pricing changes.

  • Identify opportunities to improve product pricing and profitability.

  • Analyze loan production and investor performance.

  • Identify pricing trends and product gaps.

  • Recommend new products based on borrower demand and market opportunities.

  • Assist product and pricing teams with complex configuration and analysis.


That’s where I see tremendous potential.


Rather than asking AI to replace the pricing engine, why not use AI to make the pricing engine, and the people managing it, more intelligent?


The distinction is important.


One is asking AI to perform a critical financial calculation independently. The other is using AI to improve the intelligence, efficiency, and decision-making capabilities of an established technology platform.



So, Should Lenders Buy or Build?


I don’t think there is one answer for every lender.


For smaller lenders with limited production volume, a relatively small investor network, and straightforward pricing requirements, using AI tools to assist with basic pricing scenarios might make sense, provided they understand the limitations and protect borrower information.


For larger lenders with established technology organizations, proprietary systems, and dedicated development resources, building certain capabilities internally may also make economic sense.


But for the majority of lenders in between, there’s another consideration.


Do you really want to be in the business of being a mortgage lender and a software company?


Every dollar and every employee dedicated to building and maintaining proprietary technology is a dollar and an employee that isn’t necessarily focused on originating, funding, and servicing loans.


A purpose-built SaaS mortgage pricing engine exists because pricing is a specialized business function with significant complexity, ongoing maintenance requirements, and operational consequences when something goes wrong.


The question isn’t whether you can build one with AI.


The question is whether you should.



The Bigger Question


AI is going to change how we develop software, operate businesses, and serve borrowers. On that point, I have little doubt.


But we should be careful not to confuse the ability to demonstrate that AI can perform a task with the ability to operate an enterprise business function reliably, repeatedly, and at scale.


Our little experiment demonstrated something encouraging: AI can now accurately price a mortgage product under the right conditions. It also demonstrated how quickly accuracy and speed can become challenges as complexity increases.


I’m excited about where AI is headed, and we’re actively exploring how to incorporate it into the mortgage technology ecosystem.


But I also believe there’s tremendous value in purpose-built technology that has been designed, tested, and refined to do one thing exceptionally well.


Maybe the next evolution isn’t replacing the mortgage pricing engine with AI.


Maybe it’s building a better pricing engine by putting AI to work alongside it.


I’d be interested in hearing from other mortgage executives and technology leaders: Are you seriously considering replacing proprietary or SaaS mortgage technology with AI? And if so, where do you draw the line between experimenting with AI and trusting it to run a critical part of your business?


Michael “Mike” Lewis is a growth-oriented executive and intrapreneur with more than 28 years of leadership experience across SaaS, Mortgage Technology, Fintech, and Financial services. Throughout his career, he has built and scaled technology platforms, led operational transformations, and developed high-performing teams.


As President of LoanPASS, Mike leads product strategy, innovation, and go-to-market execution for a modern pricing and decisioning platform that enables lenders to replace legacy systems with configurable, automated infrastructure. Under his leadership, the company has grown revenue 4.5x from 2023 to 2025 and launched multiple enterprise SaaS products serving a rapidly expanding national client base.


Before leading LoanPASS, Mike spent decades working within banks, lenders, consultancies, nonprofits, and enterprise software firms, driving systems consolidation, operational scale, and cultural alignment. His career reflects a consistent belief that meaningful innovation does not require starting from scratch, it requires ownership, initiative, and disciplined execution from within.


Mike is passionate about mentoring diverse teams and helping professionals develop the skills and mindset that make intrapreneurship as rewarding and impactful as entrepreneurship. Through writing, speaking, and coaching, he encourages others to cultivate strategic thinking, initiative, and innovation; skills that empower individuals to make meaningful contributions, advance their careers, and, if they choose, pursue entrepreneurial ventures with confidence.


He holds a Bachelor of Science from Brigham Young University and an MBA in E-Business from the University of Phoenix.

Keep up with industry insights and LoanPASS updates.

Subscribe for updates, insights, and other great content.

or

Keep up with industry insights and LoanPASS updates.

Subscribe for updates, insights, and other great content.

or

Keep up with industry insights and LoanPASS updates.

Subscribe for updates, insights, and other great content.

or

Keep up with industry insights and LoanPASS updates.

Subscribe for updates, insights, and other great content.

or