What the Latest Mortgage Industry Data Tells Us About AI Investment Priorities

A look at some key 2025–2026 mortgage industry data and the implications to your tech decisions.

The mortgage industry has always run on data. Rate sheets. Pipeline reports. Pull-through ratios. But there is a set of numbers circulating right now, from Fannie Mae, the Mortgage Bankers Association, Stratmor, and others, that some lenders are treating as market color when it may be more beneficial to treat them as a structural mandate.

The question is not what the data says. It is what you decide to build in response to it.

Data at the Surface Vs. a Deeper Look

The headline numbers are familiar by now. The MBA’s mortgage origination forecast for 2025 landed near $2.0 trillion, with 2026 projected to reach $2.2 trillion; a meaningful recovery from the contraction years, but still well below the $4.4 trillion peak of 2021. Purchase originations are forecast to grow 7.7% to $1.46 trillion in 2026, while refinance activity remains structurally constrained as long as the rate lock-in effect persists.

The surface reading of these numbers is: it’s a tough market, margins are thin, hold on.

But there is another way to look at this data. The data is not just describing a cycle. It is describing a potentially drastic restructuring of the economics of mortgage origination and servicing, and the organizations that read it that way are building differently. That’s because the story of the economics needs to be paired with the story of AI integration, adoption, and mortgage business adaptability to that change.

“Origination costs are still elevated and the pull-through of loan closings to applications has declined over the past four years. Many lenders are exploring ways to reduce origination costs and increase productivity through technology advances and process improvement.” — Marina Walsh, CMB, Vice President of Industry Analysis, Mortgage Bankers Association — MBA Annual Convention, October 2025

— Marina Walsh, CMB, Vice President of Industry Analysis, Mortgage Bankers Association — MBA Annual Convention, October 2025

The Deeper Signal in the Numbers

Let’s explore some of these data points:

Cost per loan remains structurally broken. According to MBA’s Quarterly Performance Report, the average cost to originate a mortgage reached approximately $11,600–$12,000 per loan through 2025 — up from roughly $7,000 five years ago. That number does not move meaningfully without changing how work flows through a loan file, not just who does it.

Cycle times are a competitive liability. STRATMOR’s research consistently identifies time-to-close as a primary driver of borrower satisfaction and lender differentiation. Borrowers no longer compare their mortgage experience only to other lenders, they compare it to Amazon, their bank’s mobile app, and any digital experience that works. Meeting those expectations with legacy processes is becoming increasingly difficult.

Based on Indecomm’s own surveys and data, the challenge isn’t necessarily the front-end borrower experience. It’s what happens behind the scenes to produce a high-quality loan. The number of processes, the people, the hand-off. These are all points of friction in the borrower experience that no point-of-sale can fix. But, there are solutions designed to address those challenges and that’s what Indecomm provides: well thought-out mortgage SaaS solutions, designed to drive a more efficient middle- and back-office, powered by AI.

The borrower experience gap is widening. NAR’s 2025 Profile of Home Buyers and Sellers showed that first-time buyers represent a growing share of purchase transactions ; a demographic with high digital expectations and low tolerance for opaque, document-heavy processes. Lenders serving this cohort with 2015-era workflows are losing them before the application is complete.

“This shift is not about replacing people. It is about reallocating human effort away from repetitive, transactional work and toward activities that require human judgment, empathy, and expertise.”

— STRATMOR Group, “Prepare for Takeoff: AI and the Fight for Mortgage Lending Efficiency,” January 2026

Fewer defects. Shorter cycle times. Lower cost per loan. A borrower experience that does not fall apart at the document stage. None of these outcomes come from buying a platform. They come from rebuilding how work gets done.

The Human Dimension the Data Doesn’t Capture

Defect rates have not improved at scale. Fannie Mae’s Loan Quality Connect data and FHFA oversight reports continue to show that repurchase demands are concentrated in loans with income calculation errors, document classification failures, and data inconsistency across the file. These tend to be process failures that well-defined AI is specifically designed to address. But, keep in mind, while AI-powered solutions help address these challenges, they will also need to keep up with the pace of new risks presented by misuse of AI.

Indecomm tackles data integrity issues in early stages of origination and underwriting with its IDXGenius | ai solution and further driving higher quality loans with its DecisionGenius solution. As we improve the quality of document-to-data accuracy and decisions, we reduce downstream issues in the loan file.

Here is the uncomfortable truth: the biggest constraint on AI adoption in mortgage is not the technology.
It is the identity shift required of the people operating it.

That transition does not appear in the MBA’s cost-per-loan data. It does not show up in Fannie Mae’s defect reports. But it is the reason most AI implementations underperform their projections. The technology works. The adoption curve is where the investment either compounds or stalls.

The organizations getting measurable returns from AI are treating the change management as a first-class project, not a training module appended to a software rollout.

The Paradox Leaders Face Right Now

There is a genuine tension in the current data environment that deserves naming directly.

The case for inaction has surface logic: volumes are uncertain, capital is constrained, and committing to a technology investment in a margin-compressed market feels like poor timing. Some leaders are watching peers make expensive AI bets and waiting to see the outcomes before moving.

The case for blind adoption is equally dangerous. Point solutions deployed without integration into loan workflow, AI models trained on data sets that do not reflect current product mix or regulatory environment, and automation that speeds up bad processes rather than correcting them — these are real failure modes, and the industry has examples of each.

AI in mortgage operations is neither a guaranteed advantage nor an optional experiment. It is a capability that must be built into the structure of how you operate — designed for the loan file, embedded in the process, and governed with the same rigor you apply to credit decisions. It cannot be bolted on.

What the Regulatory Environment Is Telling You

The data story does not end with origination economics. The regulatory backdrop is tightening the stakes.

On April 8, 2026, Fannie Mae issued Lender Letter LL-2026-04, establishing a binding AI/ML governance framework for all seller/servicers. Freddie Mac enforces similar requirements under Bulletin 2025-16. Both GSEs now require documented policies for AI system development, operation, risk control, and vendor oversight. The April 2026 OCC Bulletin 2026-13 extended interagency model risk management guidance in the same direction.

“Artificial intelligence and/or machine learning is rapidly reshaping the mortgage landscape, introducing new opportunities to enhance efficiency, strengthen risk management, and deliver more personalized customer experiences. At the same time, the pace of innovation brings heightened responsibility.”

— Fannie Mae Lender Letter LL-2026-04, April 8, 2026

The documentation of how a decision was made is becoming as important as the decision itself. Lenders will need audit trails, explainability, and data integrity at a level that manual processes cannot reliably produce.

This is not a compliance burden. It is the infrastructure requirement for operating in the market that is actually coming.

What the Best Organizations Are Doing

The lenders and servicers generating measurable returns from technology investment in this environment share three observable behaviors.

They are starting with the data layer – ensuring loan data is clean, consistently classified, and traceable before layering decisioning or automation on top of it. Document intelligence is treated as a foundational capability, not a feature. Check out Indecomm’s IDXGenius | ai.

They are measuring the right outcomes – not technology adoption rates or implementation milestones, but cost per loan, defect rates, cycle time, and repurchase exposure. The technology earns its seat at the table by moving those numbers. Request an ROI evaluation.

They are investing in the operational transition as seriously as the technology itself — redefining roles, not eliminating them, and building processors and underwriters who can supervise AI-assisted workflows.

The Real Test

The numbers circulating through MBA research, Fannie Mae guidance, and STRATMOR benchmarking studies are not market commentary. They are a description of a structural shift that is already underway.

The demolition is in progress. The question was never whether AI would change the economics of mortgage operations. The question is what you decide to build in its place, and whether you have the organizational courage to build it deliberately, with the unglamorous foundational work that actually determines whether any of these compound into advantage.

The data tells you the direction. The rest is a choice.

Indecomm’s Genius AI Suite — including IDXGenius, BotGenius, DecisionGenius, AuditGenius, and DocGenius — is designed to embed AI-powered SaaS into the operating model of mortgage lenders and servicers, not sit beside it. Learn more at indecomm.com

Sources: MBA Annual Convention & Expo (October 2025) • Fannie Mae Lender Letter LL-2026-04 (April 2026) • STRATMOR Group Technology Insight® Study & InFocus Reports (2025–2026) • NAR 2025 Profile of Home Buyers and Sellers • FHFA / Freddie Mac Bulletin 2025-16