Before Mortgage Can Be AI-Ready, We Need to Be Data-Ready

I spent a few hours Saturday morning doing something I probably do more often than I should: scrolling LinkedIn and reading what smart people across different industries are talking about.

My rabbit hole this week was Data Governance.

The more I read, the more I kept coming back to a problem that I believe exists across a meaningful portion of the mortgage industry.

We have spent the last couple of years talking about AI, automation, machine learning and the next generation of mortgage technology, but I am not convinced that enough mortgage companies have established the data foundation required to take full advantage of any of it.

For many mid-sized and smaller mortgage companies, there really isn’t an enterprise data model. Instead, there are systems. The CRM has data about prospects, leads and past clients. The LOS has data about borrowers currently in process and borrowers whose loans closed years ago. Marketing platforms have another set of data. Servicing systems may have another. Individual loan officers sometimes maintain their own databases completely outside the enterprise technology stack. The same human being can exist in three, four or five different places. Worse, the same data element may be named differently in several of those systems.

Which brings me to a deceptively simple question that David Pacific, Chief Data & Technology Officer at SQA Group, likes to ask: “How do you know if the data is right?” Think about that question inside your own company. If two dashboards produce different answers to the same question, which answer is correct? If your CRM and LOS disagree about a customer, which system wins? Who owns the definition of that data? Who is responsible for fixing it? Can you trace where the information originated and how it changed along the way? Those questions have always mattered. AI is about to make them matter a whole lot more.

AI Doesn’t Fix a Weak Data Foundation

Puneet Sharmaa, a technology and digital transformation leader whose work I came across during my Saturday morning reading, recently wrote something that stopped me: “The AI conversation is really a data conversation.” I think he is exactly right. Puneet went further, arguing that without proper data and Data Governance, an AI governance policy or committee becomes largely irrelevant.

That is an important point for the mortgage industry right now. We are appropriately spending significant time talking about AI Governance. At MISMO, we developed FRAME, the Framework for Responsible AI in the Mortgage Ecosystem, specifically to help mortgage companies establish policies, inventory their AI use cases, assess risk and prepare for increasing scrutiny from regulators, investors and the GSEs.

I talked about this extensively on a recent MIC’d Up episode, where I turned the microphone around and had my ChatGPT assistant, Freeda, interview me. One point I made during that conversation was that no mortgage company is too small to understand where AI is being used in its business and how that risk is being managed. But there is another layer underneath all of that.

What data is the AI using?

If the organization doesn’t know which source is authoritative, doesn’t have consistent definitions, doesn’t understand its data lineage and doesn’t have clear ownership of the data, adding AI doesn’t make those problems disappear. It potentially makes them bigger and faster.

A recent Forbes Technology Council article shared by SQA Group made this point particularly well. AI requires high-quality data, and inconsistencies that may have previously created reporting headaches can become significantly more consequential when AI begins using that data to generate recommendations or influence decisions. In mortgage, where decisions increasingly need to be explainable, reproducible and auditable, that matters. A lot.

What Does “Getting Your Data House in Order” Actually Mean?

This can sound like a project reserved for the largest banks with Chief Data Officers, large technology teams and enormous budgets. It isn’t. I recently reviewed the Data Governance framework implemented by a mid-sized independent mortgage banker. I won’t identify the company because that isn’t the point. The point is that they did it.

Their framework establishes named Data Owners and Data Stewards across important mortgage data domains. It identifies authoritative systems of record. It establishes Critical Data Elements. It documents lineage. It defines responsibilities for data quality, access, classification and third-party use. It establishes a governance committee and an operating cadence for keeping the work alive.

One element jumped out at me because it speaks directly to the problem I described earlier. Their Data Glossary requires one approved definition per term and identifies related terms and synonyms used in other systems. It also identifies the authoritative system of record, Data Owner and Data Steward for that term. Think about the practical impact of that.

Instead of three departments debating what a field means, the organization decides once. Instead of wondering which system contains the correct version, the authoritative source has been established. Instead of everyone owning the data, which usually means nobody really owns it, accountability has a name attached to it. That is Data Governance becoming operational.

The Goal Is Trust

Another perspective I found Saturday came from Malcolm Hyman , MBA, who made an observation that gets to the business outcome of all this. One of the biggest signs that a Data Governance program is creating value, he wrote, is when people stop asking: 

“Can I trust this data?”

That may be the best measure of success. The objective isn’t to create another committee, another binder full of policies or another compliance exercise that employees have to navigate. The objective is confidence.

Executives should be able to make decisions without first reconciling three reports. Loan officers should be able to trust the information presented to them. Compliance teams should understand where information originated. Technology teams should know which definition governs. Eventually, AI applications should be able to operate against data that the organization understands, owns and trusts. When that happens, data stops being something a company merely possesses and starts becoming an enterprise asset.

Mortgage Already Has a Head Start

Here is where I think our industry has an advantage that we may not be using enough. We don’t have to start with a blank sheet of paper. The MISMO Business Glossary contains more than 8,000 terms and definitions representing common mortgage business terms, processes, events, calculations, documents, forms and other industry concepts.

The content has been built from MISMO standards, the Logical Data Dictionary, MBA Business Glossary, eMortgage Glossary, GSE extensions, the Data Governance & Management Community of Practice and other industry sources. The industry has already spent years doing a lot of the hard work of agreeing on what things mean. Why would every mortgage company independently reinvent that vocabulary? That doesn’t mean every company’s internal data model will be identical. It won’t be. But if the mortgage industry already has a common language, it seems like a pretty logical place to begin.

This Is Work the Industry Can Do Together

MISMO also has a Data Governance Community of Practice specifically focused on helping organizations leverage, adopt and implement Data Governance and management best practices alongside MISMO industry standards. The group has developed resources around Data Governance frameworks, Fit-for-Purpose concepts, governance policies, roadmaps, metrics and maturity models, and there is more work to be done. This is where I want to make a very specific invitation.

Anyone in the mortgage industry can participate in a MISMO development workgroup or Community of Practice. You do not have to be a MISMO member to participate.

If you are a data leader, technologist, compliance professional, lender, servicer, vendor or simply someone inside your organization wrestling with these issues, we need you in the room.

One of the things I talked about on MISMO MIC’d Up is what happens when industry professionals enter a MISMO workgroup. Competitors sit beside each other, take off their corporate hats and effectively put on a Team Mortgage hat. They bring their experience and expertise together to solve problems that are bigger than any one company. Data Governance feels like exactly that kind of opportunity.

Puneet Sharmaa made another point that has stayed with me from my Saturday reading. He observed that the organizations moving fastest with AI aren’t necessarily the organizations with the fewest rules. They are the organizations whose data house is in order and whose rules are clear. I think there is an important lesson in there for mortgage. We have spent a lot of time asking how quickly we can adopt AI and what AI will allow us to do next.

I would add another question to that conversation.

Before we ask whether our mortgage company is AI-ready, are we certain that our data is ready?

Because the foundation comes first.

#VieauxPoint