Weekly Real Estate News
Summary
A survey of 31 mortgage lenders and servicers found AI is broadly in production but rarely scaled enterprise-wide. Regulatory uncertainty was the most frequently cited barrier, while post-deployment governance remains uneven.
Artificial intelligence is already inside most mortgage companies surveyed in a new industry study. Scaling it across the business is another matter.
A survey conducted by Boston Consulting Group for the American Association of Residential Mortgage Regulators and the Mortgage Bankers Association found nearly all participating lenders and servicers had at least one AI use case in production. Yet only about one-quarter had fully scaled even one use case.
The study, conducted from April through July, included 31 residential mortgage lenders and servicers representing roughly 40% of the U.S. mortgage market. It examined 38 AI applications across origination, servicing, marketing and sales, capital markets and corporate functions. The survey was reported Monday by HousingWire; MBA had previously described the BCG survey process and its collaboration with AARMR while responses were being gathered.
The results show a mortgage industry that has moved beyond experimenting with AI but has not yet demonstrated enterprise-scale deployment across much of the loan lifecycle.
Production use is ahead of scaled use
Respondents had roughly 10 of the 38 surveyed use cases in production on average. About 80% expected to increase AI investment during the next 12 months.
Adoption was concentrated in corporate productivity, operations and document-heavy origination work. Secondary and capital-markets functions remained comparatively underused, as did several servicing applications.
That gap matters because the economic promise of mortgage AI has often centered on reducing manual touches and the cost of manufacturing and servicing loans. A tool can be in production without being deployed broadly enough to materially change enterprise cost or performance.
The survey found the clearest reported benefits in employee productivity and employee experience. Evidence of broad gains was less developed in cost reduction, customer experience, regulatory compliance and credit-risk management.
Regulatory uncertainty tops the barrier list
Regulatory and compliance uncertainty was cited by 59% of respondents, making it the most frequently identified obstacle to scaling AI. Unclear return on investment followed at 45%.
Data quality and availability of solutions were each cited by 24%, while 21% pointed to concerns about AI reliability and hallucinations.
The regulatory finding is especially significant because AARMR represents state agencies that supervise nonbank residential mortgage companies. The survey was designed in part to give regulators a clearer picture of where AI is—and is not—being used across mortgage operations.
The results arrive as state regulators are developing more structured approaches to AI supervision. A recent mortgage-banking analysis of the Conference of State Bank Supervisors’ AI supervisory framework noted that the framework is intended as a risk-based examination tool rather than a new set of substantive AI requirements.
Governance is strongest before deployment
Most respondents reported foundational controls. Written AI policies and standards were in place at 87% of firms, privacy controls at 84% and human-review mechanisms at 81%. Vendor controls were reported by 74%, while 68% had formal cross-functional AI governance and security controls.
Post-deployment monitoring was less consistent. Only 58% reported ongoing monitoring for issues such as model drift and accuracy, and 45% reported regular AI reporting to their boards.
The survey also surfaced unauthorized or “shadow AI” use. More than one-quarter of respondents acknowledged employees were using AI outside approved company environments at least occasionally; another 7% said they were unsure whether it was happening.
For mortgage executives, that may be the more consequential finding than raw adoption. The industry is not waiting for a settled regulatory framework before putting AI into production. But deployment is moving faster than broad scaling, measurable ROI and some elements of continuing governance.
That leaves the next phase less about whether mortgage companies will use AI and more about whether they can demonstrate that it works reliably, produces measurable economics and remains governable after deployment.
