MBA Premier Member Editorial: The State of AI Governance in Mortgage Servicing
Sapan Bafna is Founder & CEO of Outamation, a mortgage operations automation company serving lenders and servicers. He believes the industry’s AI challenge is not adoption but operationalization: fix the workflow first, measure what goes wrong as closely as what goes right, and keep people where judgment and accountability matter.
For the last two years, mortgage executives have asked me the same question in different forms: How do we use AI without creating a bigger problem than the one we are trying to solve?

It is the right question.
AI is no longer a side conversation in mortgage. It is entering real operational workflows, from document processing and quality control to customer operations and decision support. But using AI and governing AI are not the same thing. The gap between the two will determine which companies can scale safely.
I have spent nearly two decades in mortgage operations and technology, at CoreLogic, First American and now Outamation. I have seen companies automate individual tasks while leaving the underlying process unchanged. In mission-critical, SLA-driven work, technology that fails when you need it is stressful. We have had our own moments. They taught me that reliability, accountability and visibility have to be designed into the process. They cannot be added after something goes wrong.
Governance Is No Longer Optional
The signals are now formal.
Freddie Mac’s AI and machine learning governance framework took effect March 3, 2026. Fannie Mae followed with Lender Letter LL-2026-04, effective August 6, 2026. Both put formal expectations around AI governance for seller/servicers, including policies, risk management and oversight of AI used by vendors and subcontractors. Fannie Mae also requires seller/servicers to be prepared to disclose, upon request, the AI they use, how they use it and the safeguards they have implemented.
The industry is building practical tools around those expectations, too. In June, MISMO released FRAME, a mortgage-specific AI governance toolkit that includes a governance policy template, an AI system inventory and a risk assessment.
A joint survey by the Mortgage Bankers Association, the American Association of Residential Mortgage Regulators and Boston Consulting Group found that 59% of respondents identified regulatory and compliance uncertainty as their biggest barrier to scaling AI. Meanwhile, 87% reported having written AI policies or standards, but only 58% reported ongoing monitoring for issues such as accuracy and model drift.
Read those numbers together. Most companies have a policy. Far fewer have the operating discipline behind it.
The industry does not have an AI adoption problem. It has an AI operationalization problem.
Asking ChatGPT, Gemini or Claude a question is easy. Operationalizing a workflow is the real work. That means every check and balance, the variability of daily volume, and complete observability so an auditor or operator can see what happened and why.
The Four Pillars of Operational AI Governance
Writing an AI policy is easy. Running governance inside daily operations is hard. In practice, real governance lives in four places.
- People: Named owners, clear accountability and defined reviewer roles.
- Process: Confidence thresholds, rules for when automation stops and clear exception routing.
- Technology: Drift monitoring, version control, regression testing and reliable fallbacks.
- Data: Lineage to the source document, field-level confidence and durable audit logs.
ISO/IEC 42001 treats AI governance as a management system that is established, maintained and continually improved. We learned the value of that discipline the hard way. At Outamation, we review compliance, information security, infrastructure and code reliability every month. We would rather find gaps on our own schedule than have a crisis reveal them for us.
Governance has to live inside the operating model, not beside it.
The Vendor Problem Is the Servicer’s Problem
Most servicers will license AI from vendors rather than build every model themselves. Buying the technology does not transfer the accountability.
Fannie Mae’s requirements make that clear. Seller/servicers must manage the risks and governance of vendor and subcontractor use of AI/ML in a way that is “no less protective” of Fannie Mae’s requirements.
That changes procurement.
The old question was: “Does your AI work?”
The new question is: “What happens when it doesn’t?”
Servicers need to understand how a vendor monitors its models, how changes are controlled, where people remain in the loop and what records can be produced for an examination. Vendor diligence cannot stop at a product demonstration or an accuracy statistic.
The Human in the Loop Is a Control
Some people describe human review as a sign that AI is not good enough. I disagree.
In regulated mortgage operations, the goal is not to remove human judgment. It is to let machines do what they do well and put people exactly where judgment and accountability matter.
A good AI workflow extracts the data, applies the rules, flags the exceptions and sends higher-risk items to a person. Reviewing the small share of files that cross a defined risk threshold is not inefficiency. It is controlled automation.
Good governance is not a brake on AI. It is what makes scale possible.
The servicers that win will not be the ones with the most AI. They will be the ones that can put AI into production, monitor it, explain it and trust it.
(Views expressed in this article do not necessarily reflect policies of the Mortgage Bankers Association, nor do they connote an MBA endorsement of a specific company, product or service. MBA NewsLink welcomes submissions from member firms. Inquiries can be sent to Editor Michael Tucker or Editorial Manager Anneliese Mahoney.)
