Why 62% of Lenders Still Haven't Adopted AI — And What the Early Movers Got Right

The headlines about AI in mortgage lending are relentless. Agentic workflows. Conversational underwriting. Loans processed in under 60 minutes. AI voice agents handling borrower calls at 2am.

And yet, a November 2025 STRATMOR Group survey found that only 38% of mortgage lenders are using any form of AI or machine learning. That means 62% — nearly two thirds of the industry — haven't started.

That's not ignorance. These are sophisticated operators running complex, regulated businesses. So why aren't they moving?

---

The Four Structural Barriers Nobody Talks About Honestly

A Navitas Capital analysis published in April 2026 identified the structural forces that make mortgage lending uniquely resistant to AI adoption. They're worth understanding — because they explain not just the hesitation, but also what separates the lenders who are successfully implementing AI from those who aren't.

1. Market Heterogeneity

Brokers, national banks, credit unions, homebuilders with captive mortgage operations — each operates under different regulatory regimes, business models, and technology stacks. An AI tool that works for a $50B national bank rarely ports cleanly to a 12-person independent mortgage lender.

The result: AI vendors over-promise broad applicability, then struggle to deliver across the full market. Lenders who've watched this cycle play out before are understandably skeptical of the next pitch.

2. The Boom-Bust Trap

Mortgage lending is volume-driven and rate-sensitive. In boom cycles, lenders are too busy processing loans to implement new systems. In slow cycles, they're cutting costs and preserving cash for the next upswing.

The window for meaningful technology investment is, as Navitas put it, "vanishingly small." And the transaction-based revenue model makes it nearly impossible for AI vendors to build the recurring relationships that drive successful implementation.

3. A Culture Built on Risk Minimization

Mortgage lending is, at its core, a documentation exercise. Every file is assembled to withstand scrutiny — from the GSEs, from regulators, from secondary market buyers. This produces a deeply risk-averse decision-making culture that extends to technology adoption.

One lender CEO told Navitas Capital directly during a diligence conversation: "If we have to switch loan origination systems, I will not be the CEO anymore."

That's not stubbornness. That's a rational response to the consequences of getting it wrong.

4. Decentralized, Relationship-Driven Operations

Most mortgage lenders grow by recruiting loan officer teams — not by building technology infrastructure. Loan officers control volume through personal relationships with realtors, wealth managers, and real estate investors. Getting individual LOs to change a workflow they've used for 15 years is a different challenge entirely from deploying software to a centralized team.

---

The Mortgagetech Graveyard

The first wave of AI-enabled mortgage startups burned through billions of dollars in venture capital. Better and Blend — the most prominent survivors — both trade well below the capital they raised. Dozens of well-funded companies never achieved escape velocity or were quietly absorbed by incumbents.

Meanwhile, industry productivity has only fallen despite the investment. The technology was real. The problem-solving intention was genuine. But the structural barriers above weren't solved — they were ignored.

The lesson for 2026 is clear: AI tools that don't integrate into existing LOS systems, respect the regulatory environment, and deliver measurable ROI within the boom-bust cycle will meet the same fate.

---

What the Early Movers Got Right

The 38% who are successfully using AI aren't doing it by ripping and replacing. They're doing it by integrating — carefully, incrementally, with a clear eye on compliance and ROI.

The pattern that works looks like this:

Start with the workflow, not the technology. The successful deployments — ICE's Aurora agentic AI embedded inside Encompass, Ocrolus's document processing layered on top of existing LOS systems, Friday Harbor's pre-underwriting integrated into Calyx Path — all work with the existing stack, not against it.

Exception-based, not autonomous. ICE made a deliberate architectural choice with Aurora: human-authorized for all sensitive actions. No AI final decisions on approvals, pricing, disclosures, or cash movement. AI handles the routine; humans handle the consequential. This isn't a limitation — it's the reason lenders trust it.

Solve the compliance question before deployment. The GSE AI governance mandates that landed in 2026 — Freddie Mac's requirements effective March 2026, Fannie Mae's LL-2026-04 following immediately after — require lenders to inventory, document, and produce audit-ready disclosure of every AI tool touching a loan. The lenders moving fastest on AI are the ones who built compliance into the architecture from day one, not the ones retrofitting governance after the fact.

Verify before you automate. Every AI system is only as reliable as the data it consumes. Automating a workflow that ingests fabricated documents or synthetic identities doesn't eliminate fraud — it accelerates it. The lenders who are getting AI right have established deterministic verification at the data layer before deploying probabilistic automation on top of it.

---

The Opportunity in the Gap

62% of lenders haven't adopted AI. That's not a failure — it's a market in formation.

The second wave of mortgage AI will look different from the first. It will be integration-first, compliance-native, exception-based, and built on verified data. The lenders who understand those requirements now — and build their technology stack accordingly — will have a significant operational advantage when origination volume recovers fully.

The tools are ready. The regulatory framework is, for the first time, clearly defined. The question is whether lenders will move before the next rate cycle forces the decision.

---

Stephen Schrump is the CEO of PitchPoint Solutions, a data verification platform serving 2,500+ mortgage lenders across North America. PitchPoint is SOC 2 Type II certified with zero exceptions over 13 consecutive months of continuous audit.

Connect with Stephen on LinkedIn or visit pitchpointsolutions.com

---

Sources

• STRATMOR Group — "Artificial Intelligence in Mortgage Lending" (November 2025)
• Thesis Driven / Navitas Capital — "Why AI Has Struggled to Break Into Mortgage Lending" (April 2026)
• HousingWire — "The AI-native mortgage" — ICE Aurora (April 2026)
• PR Newswire — Ocrolus condition lifecycle management (March 2026)
• PR Newswire — Friday Harbor / Calyx Path integration (April 2026)
• Fannie Mae Lender Letter LL-2026-04 — AI Governance Requirements (2026)
• Freddie Mac AI Governance Seller/Servicer Requirements (March 2026)