Key facts
- Nearly all mortgage lenders surveyed have at least one AI use case in production.
- Only about 25% of mortgage lenders have fully scaled at least one AI use case.
- Employee productivity tools are in production at 87% of respondents.
- Data extraction from documents is in production at 61% of respondents.
- Regulatory and compliance uncertainty is the largest barrier to scaling AI (59%).
- Unclear return on investment is the second-largest barrier to scaling AI (45%).
Mortgage lenders and servicers are increasingly deploying artificial intelligence, but adoption remains concentrated in productivity and document-heavy tasks, with regulatory uncertainty and limited measurable business results slowing broader implementation. This is according to a joint survey released Monday by the American Association of Residential Mortgage Regulators (AARMR), the Mortgage Bankers Association (MBA) and Boston Consulting Group. The survey, conducted from April through July 2026, included 31 residential mortgage lenders and servicers representing about 40% of the U.S. mortgage market. It assessed 38 AI use cases across various functions. Findings indicate that mortgage companies are in early stages of AI adoption compared to other financial services sectors. Nearly all respondents had at least one AI use case in production, but only about one-quarter had fully scaled at least one use case. Respondents had about 10 of the 38 use cases in production on average, and roughly 80% expected to increase AI investment over the next 12 months. Adoption is concentrated in corporate functions, operations, and origination document workflows, while secondary and capital markets and parts of servicing remain largely untapped. Most common applications focus on helping employees perform existing tasks rather than fundamentally changing processes. Employee productivity tools for writing and summarization were in production at 87% of respondents, followed by code generation and developer support (65%), data extraction from documents (61%), agent assistance and knowledge search (54%), and investor guideline and eligibility extraction (52%). Document classification and summarization was also in production at 52% of respondents. Within origination, data extraction from documents and investor guideline or eligibility extraction were among the most advanced use cases, with 42% of respondents reporting scaled production. Document classification and summarization reached scaled production at 35% of respondents. More complex applications, including underwriting decision support, fraud detection, and credit risk analytics, had significantly lower adoption levels. Adoption in secondary and capital markets was even lower, with bond and investor information chatbots, portfolio analytics, and capital markets pricing insights in production at relatively small shares of respondents. Sales and distribution analytics had not reached the production stage. Servicing also presents significant room for expansion, particularly in early default warnings, collections prioritization, and loss-mitigation decision support. Measurable benefits from early AI adoption are limited. Employee productivity and experience were identified as benefits, but gains were less developed in cost reduction, customer experience, regulatory compliance, and credit risk management. Regulatory and compliance uncertainty was the largest barrier to scaling AI (59%), followed by unclear return on investment (45%). Data quality and solution availability were cited by 24%, while concerns about AI reliability and hallucinations were cited by 21%. Mortgage companies generally have basic AI controls, but governance often weakens after deployment. Written AI policies and standards were reported by 87% of respondents, privacy controls by 84%, and human review by 81%. Third-party and vendor controls were in place at 74%, while 68% reported formal cross-functional AI governance and security controls. Only 58% reported ongoing monitoring for issues like model drift and accuracy, and 45% reported regular AI reporting to their boards. More than one-quarter of respondents acknowledged employees using AI outside approved environments.