Mortgage-AI

Mortgage Wasn’t Built for AI. The Foundation Under It Matters More Than Ever.

AI is already in mortgage. The question is whether the operation underneath it is ready for what comes next.

For our latest Calyx Sunlight Series, we sat down with Brian Vieaux, CMB, President of MISMO, to talk about what it will take for mortgage organizations to move beyond isolated AI tools and use the technology responsibly across the business.

The conversation went beyond AI itself.

We talked about the data underneath it, the systems it needs to work across, the governance required to manage risk and the importance of understanding where AI may already be operating inside an organization.

Because AI can make a strong operation more capable.

It can also amplify fragmented data, inconsistent processes and unclear accountability.

Brian’s practical advice gave lenders a clear place to start:

Start with the inventory.

Watch the conversation

Watch: Mortgage Wasn’t Built for AI.

Start with the full conversation, then use the recap below to share the practical takeaways with your team.

Mortgage has digitized a lot. That doesn’t mean the data is ready for AI.

During the conversation, Blake compared the evolution of mortgage technology to the shift from MapQuest to Google Maps and eventually to self-driving vehicles.

The analogy matters.

Mortgage has spent years converting paper forms, documents and loan files into digital versions. But many processes still depend on people to interpret information, reconcile inconsistencies, recognize what changed and determine what needs to happen next.

Digitizing a document is not the same as creating data that systems can reliably understand and act on.

Brian explained that when information remains locked in forms, AI may first have to extract it, categorize it and determine its context before it can do anything useful with it.

Every additional interpretation introduces complexity.

That is why structured, standardized and consistently defined data matters.

AI does not eliminate the need for a strong data foundation. It makes that foundation more important.

A single source of truth matters more when AI starts working across systems

For many mortgage organizations, important information does not live in one place.

A consumer may exist in the LOS. That same consumer may also exist in a CRM, potentially with different information, histories or context.

Then additional pieces of the loan may live across other vendor platforms and internal systems.

Brian emphasized the value of establishing a clear source of truth.

That could ultimately be the LOS, CRM or a separate data environment that brings information together.

The specific technology may differ by organization.

What matters is knowing which information AI and other systems should trust.

This becomes especially important as the use case moves closer to decisioning.

Using AI to support a marketing interaction presents a different risk than using it to interpret information that could influence underwriting or a credit decision.

Before asking AI to act across the operation, lenders should understand where their authoritative data actually lives.

AI governance starts with knowing what you are governing

Governance can sound like an enterprise policy exercise.

MISMO FRAME makes the starting point much more practical.

Brian explained that FRAME was created to give mortgage organizations a set of tools they can use to build and document an AI governance program.

The framework includes a policy and procedure template, an AI use-case inventory and a risk-assessment tool.

The inventory may be the most important first step.

Rather than simply documenting that an organization uses a particular AI product, FRAME encourages teams to identify what the technology is actually being used to do.

What information does it receive?

What does it produce?

Where does its output go?

Could that output influence a consumer interaction, internal process or lending decision?

That gives leadership, compliance, risk and technology teams something specific to evaluate.

You cannot govern AI you do not know is there.

The risk is in the use case, not simply the tool

Not every use of AI creates the same level of risk.

Brian used a simple example during the session.

The same AI provider might offer one capability that helps create templated emails and another that helps prepare a loan package for underwriting.

Those functions may exist inside the same product.

They should not automatically receive the same risk assessment.

Drafting an email sits much farther away from a lending decision than AI that is interpreting or preparing information used during underwriting.

That is why FRAME looks at risk at the use-case level, not simply at the vendor or product level.

Evaluating AI this way can also uncover duplication.

An organization may discover that several different products are performing similar functions across different departments.

Seeing the full inventory creates an opportunity to evaluate not only risk, but the technology stack itself.

The closer AI gets to decisioning, the more important it becomes to understand exactly what that use case is doing.

The AI your organization didn’t approve still counts

One of the more immediate governance challenges may already be happening inside the organization.

Employees are using AI.

Some may be using approved enterprise platforms.

Others may be using personal accounts, standalone tools or AI capabilities embedded within vendor technology.

That creates what is often described as shadow AI.

Brian discussed the idea of giving employees some level of amnesty when asking them to disclose the AI tools they are already using.

The goal is visibility, not simply enforcement.

If a large group of employees is independently gravitating toward the same type of tool to solve the same problem, leadership may have an opportunity to bring that capability into an approved enterprise environment with appropriate controls.

Before an organization can control AI use, it needs to understand where that use already exists.

Vendor AI belongs in the governance conversation, too

An AI inventory cannot stop with tools employees access directly.

Mortgage organizations consume technology through LOS platforms, CRMs, service providers and dozens of other vendor relationships.

Some of those products may already include AI-enabled capabilities.

Brian explained that vendor oversight is an important part of AI governance and discussed MISMO’s work around standardized AI governance information at the product level.

The goal is to make it easier for lenders to understand how vendors are using AI and incorporate that information into their own governance programs.

That could also reduce duplicate work for technology providers answering similar AI-governance questions from multiple customers.

If a vendor’s AI touches your data or your process, it belongs in the inventory.

Governance is becoming an operating issue, not just a future planning exercise

Brian also discussed why organizations should not wait until someone asks for governance documentation before beginning the work.

Seller-servicers already operate within AI-related expectations through their GSE relationships, he explained, and those expectations can extend through aggregators and correspondent relationships to other originators.

The point was not to create alarm.

The point was to recognize that accountability around AI is already beginning to take shape.

Brian also pointed attendees toward guidance from the Conference of State Bank Supervisors (CSBS) as another resource organizations can review when evaluating how regulators may approach AI governance.

Requirements will vary by organization, relationship and jurisdiction.

Lenders should confirm applicable expectations with their compliance teams, legal counsel, regulators and investor partners.

What we’d be checking operationally

Start with the people.

Ask teams across origination, operations, marketing, compliance, technology and leadership where AI is already being used.

Do not stop at the name of the product.

Ask what they are using it to do.

Get down to the use case.

Document what information goes into the tool, what comes out and what happens to that output next.

An AI-generated marketing email and AI-assisted loan decisioning should not be treated as the same risk.

Look at vendor AI separately.

Ask technology and service partners where AI operates inside their products, what data it touches and what documentation they can provide around governance and controls.

Identify ownership.

Someone needs responsibility for maintaining the inventory, reviewing new use cases, documenting risk decisions and updating governance as technology changes.

Look underneath the AI at the data.

Where does authoritative information live?

How consistent is it across your LOS, CRM and other systems?

If AI is expected to work across those environments, inconsistent data will not disappear simply because a smarter technology is sitting on top of it.

Then make governance proportional to the risk.

The goal is not to treat every AI-assisted task like automated underwriting.

The goal is to know the difference.

MISMO’s role is bigger than publishing a data standard

Brian also explained how work like FRAME gets developed.

MISMO itself has a relatively small staff.

Much of the work happens through participation from the broader mortgage ecosystem in communities and work groups.

Lenders, vendors, technologists, compliance professionals and other stakeholders come together around shared industry problems.

Brian specifically encouraged organizations to participate in groups such as MISMO’s Data Governance Community of Practice.

You do not have to arrive with an answer.

Sometimes the value is simply hearing how other organizations are approaching the same challenge.

Industry standards work when the industry participates in building them.

Standardized data could change compliance work, too

The conversation also turned to another MISMO initiative: the Mortgage Compliance Dataset, or MCD.

The problem behind it is familiar to multi-state lenders.

Different state regulators may request essentially the same loan information in different formats.

That can leave lenders dedicating significant time and resources to repeatedly repackaging information for different examinations and audits.

MCD was developed through collaboration among lenders, regulators, legal and compliance professionals and technologists to create a common data specification.

Brian said Iowa had already incorporated MCD into its process and that California planned to make it an available method for delivering compliance data in the first quarter of 2027.

The long-term idea is simple:

Create the information once. Use it multiple times.

As adoption grows, that could reduce repetitive compliance work and give lenders another example of what becomes possible when the industry agrees on consistent data definitions.

There’s a bigger signal underneath this

The AI conversation in mortgage is becoming a data and governance conversation.

The industry does not simply need smarter tools.

It needs information those tools can reliably understand.

It needs processes they can safely operate within.

And it needs clear accountability for what happens when AI participates in the work.

That does not mean every mortgage organization needs to rebuild its technology environment before using AI.

It does mean organizations should understand the environment they are asking AI to operate inside.

Structured data makes automation more useful.

A clear source of truth makes it more reliable.

Governance makes it more accountable.

Together, those pieces will determine whether AI reduces friction or simply scales it.

Want to go deeper

Watch the full conversation for more on MISMO FRAME, data governance, shadow AI, vendor oversight and how standardized mortgage data could support a more AI-enabled operation.

[WATCH THE FULL SUNLIGHT SERIES RECORDING]

Resources from the discussion

MISMO FRAME
A practical framework for establishing AI governance, documenting use cases and evaluating risk across the organization.

MISMO Business Glossary
Standardized mortgage terminology and data definitions designed to create greater consistency across systems and workflows.

MISMO Data Governance Community of Practice
A forum for mortgage professionals to learn from and contribute to industry conversations around data governance and management.

Mortgage Compliance Dataset (MCD)
A standardized dataset designed to create greater consistency in mortgage compliance reporting and regulatory examinations.

CSBS Artificial Intelligence Supervisory Framework
The AI governance framework from the Conference of State Bank Supervisors referenced during the conversation.

Keep the conversation going

MISMO Winter Summit
Learn more about MISMO’s upcoming industry summit and opportunities to connect around standards, technology and emerging mortgage issues.

MISMO Communities & Workgroups
Explore the workgroups helping develop and advance MISMO standards across the mortgage industry.

The biggest takeaway

AI readiness starts with visibility.

Know where AI is being used.

Know what data it touches.

Know what each use case is doing.

Know who is responsible for it.

Then build governance around the actual level of risk.

Mortgage does not need to wait for some future version of AI to begin this work. AI is already here.

The opportunity now is to make sure the data, processes and accountability underneath it are ready for what comes next.

 

About Path Software

Path Software is the industry’s most modern and flexible mortgage loan origination system, delivering a fully digital, secure and compliant workflow that is highly configurable and simultaneously accommodates retail, wholesale, correspondent and TPO channels. Path is cloud-based, scalable and device-independent, allowing loans to move seamlessly across departments with real-time updates and analytics to improve employee productivity. Its built-in point-of-sale component allows loan officers to capture borrower information that is directly integrated into the Path platform and conveniently accessible via their personalized URL and branded landing page.

Want to learn more?

Contact us today to learn how Path can help you.