Is technology changing the economics?

Each installment of this series has opened with the 30-year mortgage rate higher than the one before it. This week was no exception.

Freddie Mac reported that the 30-year fixed-rate mortgage averaged 7.03% as of September 24, up from 6.95% the previous week and 6.30% a year earlier. Our first article, published four weeks ago, quoted 6.66%.

In its mid-September forecast, the Mortgage Bankers Association trimmed its projection for single-family mortgage origination volume as Treasury yields rose and inflation stayed elevated.

We’ll keep this last article before the HW Mortgage Summit on October 1 in Dallas data rich.  Because data and agentic AI will drive cost downward over the next 2-3 years.  The question for our fifth installment: Is technology changing the economics?

Hell, yes! Just ask the tech providers. “Sun West puts its cost to originate a loan below $150…using AngelAI.”¹ A major LOS provider states its system results in an average of $1,056 increase in gross profit per loan.² A POS states its system increases profit by $914 per loan.³

These numbers remind us of automobile EPA gas-mileage estimates. “Your results may vary.”

Across the industry, the cost to originate remains stubbornly around $11,000 per loan. Whatever the benefits at individual lenders, technology spending has yet to produce a clear, sustained reduction in overall industry costs.

In the latest six quarters of MBA Quarterly Performance Report data we analyzed, compensation accounts for roughly two-thirds of direct cost to originate. Technology accounts for about 4%.

The longer view tells much the same story: the industry’s cost structure has barely moved. Fourteen consecutive quarters report the compensation cost share of direct cost to originate at more than 66%.

Cost distributions alone do not prove that technology has failed to improve productivity. But they give CEOs good reason to ask where the promised savings are showing up.

Technology alone is not the answer

“Mortgage lending is the most undisciplined business process I’ve ever seen.” That observation came from a senior partner at a Big Four consultancy. The variation in performance across functions and individual employees gives the observation some weight.

A business process needs to be designed to produce specific results, with cost standards, service levels and employee performance measures. Technology does not benefit an undisciplined process. Nor does a disciplined process ignore 75% of available automation in the tech stack that lenders are currently paying for.

Management must hold people accountable to following a specific business process. Without discipline to a process, employees follow their own script of how to perform their jobs. They ignore or work around the technology. That’s the main reason technology has not meaningfully reduced cost to originate. The variation below shows how much room there is to improve.

Customers ultimately bear much of the cost of that variation. An average cost per loan conceals what happens on each file. Every loan has its own cost to originate.  Process discipline was the recommended response to elevated origination costs in 2010, in 2015, through the refinance volume of 2020, and again last year. The recommendation was sound each time it was made, and the record shows how rarely it was implemented: fourteen consecutive quarters with compensation above 66% of direct cost. Discipline has been available to every lender for fifteen years. Adoption, not awareness, has been the constraint.

A path to $8,000+ to close a loan

Identifying how to get to $8,000 per loan or less requires decomposing the number. A retail lender’s roughly $12,200 cost per closed loan divides into approximately $7,300 of sales expense, dominated by loan officer compensation at roughly 100 basis points, and approximately $4,900 of fulfillment, production support and corporate expense: processing, underwriting, disclosures, closing, post-closing, quality control, and the allocation of executive and administrative cost. Process discipline reduces that operating expense at the margin. It does not restructure it.

AI restructures it. Platforms now in production, SnapDocs and Copperlane. AI among them, are not workflow tools that route documents to staff more quickly. They read the file, balance the Closing Disclosure, clear conditions, review the full loan population in quality control, and route only the exceptions to staff. These results are not vendor estimates. In a top 10 lender’s deployment of SnapDocs, quality control review time fell 71%, closing team throughput rose 65%, from 659 to 1,092 loans per month, and the four to five additional employees that growth would otherwise have required were not hired. Prior to automation, balancing a single Closing Disclosure required 118 minutes of a closer’s time on every loan.

The second chart below is very detailed.  We share it because it illustrates the process and discipline will not be a quick fix.  It illustrates that process change and discipline are required across the entire organization.   The chart models the effect of extending results of that kind across the full expense base, under two scenarios. 

The first applies only gains already measured in production deployments: the cost per closed loan falls to $9,675. The second reflects an AI-native operating model, in which files move straight through the system and staff handle exceptions and judgment: $8,090, with the expense below the loan officer falling to $2,140 per loan, a 56% reduction. 

One line increases in both scenarios. Technology expense rises about $300 consistent with Freddie Mac’s finding that technology’s share of origination cost has doubled and continues to rise.  But technology spend reduces cost to originate by a factor of 7 to 10x.

The AI-native scenario floors at $8,090, because $4,700 of every closed loan is sales compensation, which automation does not address. Closing the remaining gap to the top quartile near $6,900 requires restructuring sales compensation, a strategic decision rather than a technology investment. The same arithmetic explains the flat aggregate curve. STRATMOR’s Technology Insight Study shows AI adoption among lenders rising from 15% in 2023 to 37% in 2024 to 60% in 2025. Most of the industry adopted these tools within the past two years, and the addressable base is 40% of the cost per loan. The aggregate does not yet reflect the change. Individual lenders are already realizing it.

One assumption in the model deserves scrutiny before it hardens into a conclusion: that sales compensation stays untouched. We both expect that assumption to hold only through the current adoption cycle. Inbound and outbound voice AI platforms, Flair Labs among them, are beginning to absorb the sales work that precedes a licensed conversation: outbound follow-up, lead qualification, prequalification and the initiation of the application. As those systems take over the top of the funnel, corporate technology investment performs work that commission dollars pay for today. Once that is proven on purchase business, lenders will be positioned, justifiably, to compensate licensed originators for a narrower scope of work. The originator’s economics change shape rather than simply shrinking: 25 to 50 basis points less per loan, on materially higher volume per originator, with the company’s technology carrying acquisition and qualification.

The timing matters. Consumer-direct platforms absorb this first, incrementally in 2027, because their sales models are already centralized and technology-forward. Distributed retail follows in 2028 and beyond, as competitive pressure normalizes the adjusted structure. At current average loan balances near $386,000, a 25 basis point adjustment is roughly $950 per loan and a 50 basis point adjustment roughly $1,900. Applied to the AI-native model’s $8,090, the cost to close falls into the $7,000 range at the lower end of the adjustment and toward the low $6,000s at the upper end, at or below today’s top quartile. The compensation decision described above stops being theoretical at that point. It becomes a competitive requirement.

The three rules that follow remain necessary. They establish the operating discipline an AI deployment requires to produce measured results rather than another unused license. The material cost reductions, however, will come from the technology.

Technology must enable and reinforce a process designed around specific cost, service and throughput targets. Before investing another dollar, consider three rules.

Rule 1: Commit to cost and profit targets

Make cost to originate and profit targets strategic requirements. Set service-level and customer-experience standards alongside them. Compensation consumes roughly two-thirds of direct costs. Decide what productivity must look like to deliver the results you need.

Business process change is hard. Hiring people to work around process flaws is easy. Jonathan Corr, former CEO of Ellie Mae, called it “human spackle.” Keep filling the cracks with people and the underlying process never gets fixed.

Measure what each employee produces. Touches and resubmissions can help diagnose a problem, but they are not the result you are paying for. Measure completed work, quality, timeliness and cost. You’ll see why your cost to originate is where it is, and what must change to get it where you want it.

Rule 2: Hold technology partners accountable

A promised return costs the provider nothing unless there is a consequence for missing it. You sign a three- to five-year contract. The provider gets recurring revenue for its investors. You still must deliver the savings.

Put material ROI promises in the contract. Agree on the baseline, the results, the measurement period and each party’s responsibilities. If the provider does not deliver its commitments, require a price reduction or another meaningful remedy.

Seek three- or six-month initial terms, or termination rights that let you exit if the product fails to perform. The promises need teeth.

Rule 3: Tie compensation to results

Hold everyone, from senior management to the front line, accountable for defined results. Reflect those responsibilities in performance goals and compensation. The executive sponsoring a technology initiative should have as much reason to make it work as the employees expected to adopt it.

Apply the same discipline to your technology team. “We can build our own LOS. Or POS. Or CRM. We can vibe code everything.”

Maybe so. Set the deliverables, deadlines and budget before the work begins. Monitor actual performance and stop projects that cannot deliver. Building something is not the same as improving the economics.

Before you approve the next technology investment, ask who is accountable for the result, and what happens if it never arrives.

On October 1, at the HousingWire Mortgage Banking Summit in Dallas, questions go before the industry in a live survey, with the results displayed as the room responds. This week’s entry is on the ballot: Is technology changing the lender economics?

After the Summit, we’ll publish the final article in this series, “Are lenders creating strategic optionality?” The live HW Mortgage Summit will inform that final article.

This is the fifth article in a six-part HousingWire Mortgage Banking Summit series. Jim Deitch is the CEO and Founder of Teraverde. Dr. Rick Roque is the Sr. VP of Strategic Growth and M&A at NFM Lending and founder and Managing Director of Menlo Company. Originator movement data provided by RETR, the industry’s largest repository of housing and origination data.

This column does not necessarily reflect the opinion of HousingWire’s editorial department and its owners. To contact the editor responsible for this piece: [email protected].

Notes

1. https://www.techtimes.com/articles/327042/20260909/how-pavan-agarwal-built-angelai-give-same-mortgage-answer-every-time.htm

2. https://www.ice.com/publicdocs/mortgage/IMT-data-sheet-Encompass-platform-overview.pdf

3. https://info.blend.com/hubfs/PDF/Ebook/Mortgage_Suite_ROI.pdf