Before AI Can Deliver, the Foundation Has to Work

A conversation on last Friday’s Last Word show stayed with me. We spend a lot of time talking about what artificial intelligence can do, and for good reason. The technology can analyze information, automate routine work, identify patterns, and help companies operate more efficiently. Still, the more immediate lesson may be what AI reveals about the organization using it.

AI has a way of exposing the quality of a company’s underlying data very quickly. If information is inconsistent, fragmented, duplicated, poorly defined, or stored across disconnected systems, those weaknesses become harder to ignore once AI is introduced. The technology depends on the information it is given, along with the rules and processes surrounding that information. When those inputs are unreliable, the output will be unreliable as well.

Mortgage companies have learned to operate around imperfect data for years. Employees know which report requires a second look, which field is rarely updated, and which spreadsheet contains the information everyone actually trusts. That institutional knowledge may allow people to navigate a flawed process, but an AI tool does not automatically inherit that understanding. Unless those exceptions and definitions are intentionally addressed, the tool will treat the available information as accurate.

This is why the AI conversation cannot be separated from data quality and data governance. A company may have accumulated an enormous amount of information, but volume alone does not make that information valuable. The real questions are whether the data is accurate, consistently defined, accessible to the right people, protected from the wrong people, and useful for the decisions the organization wants to make.

Poor data has always created costs. Employees spend time reconciling reports, correcting errors, entering the same information into multiple systems, and searching for documents that should be easy to locate. AI raises the stakes because it can carry those weaknesses into more workflows and more decisions. A broken process that once affected one employee or one department can suddenly influence activity across the company.

In other words, AI may not create the underlying problem. It will often make an existing problem much easier to see.

That reality should influence how companies approach AI governance. Too often, governance is viewed as a compliance exercise or a technology policy owned by a small group of specialists. Those responsibilities are certainly part of it, but effective AI governance is ultimately about how the business operates. Companies need to understand where AI is being used, what information it can access, who is responsible for reviewing the output, and what happens when the technology produces an incorrect or questionable result.

Third-party technology deserves the same level of attention. Most mortgage companies will not build their own artificial intelligence systems, but nearly every lender will use vendors that are adding AI capabilities to their products. The lender may outsource the technology, but it cannot outsource responsibility for how that technology affects borrowers, employees, counterparties, or business decisions.

If a tool influences borrower communications, fraud detection, quality control, underwriting, property valuation, compliance monitoring, or servicing activity, someone inside the company should understand how the output is being used. Leadership does not need to master every technical detail, but it does need enough visibility to ask practical questions. What business purpose does the tool serve? What data does it use? How was the tool evaluated? Who reviews the results? Can the company explain an outcome if a borrower, regulator, investor, or auditor asks about it?

Those questions also connect directly to the value of a mortgage company. Historically, we have looked at production, people, customer relationships, servicing assets, operational capabilities, and market presence when evaluating a lender’s franchise. The quality and usefulness of the company’s data should increasingly become part of that equation.

Two lenders may have originated the same number of loans over the same period, yet they may not possess equally valuable information. One lender may have clean, well-defined data that can support analysis, automation, customer engagement, compliance, and future technology. The other may have years of information scattered across platforms, spreadsheets, inboxes, and employee-controlled files. One organization can put its data to work. The other must spend time and money repairing it first.

There is another part of this conversation that deserves more attention. AI governance and data quality cannot remain solely in the boardroom, the executive suite, the compliance department, or the technology team. These disciplines need to reach every individual contributor in the company, including the loan originator.

Loan originators make decisions about data throughout the day. They enter borrower information, maintain CRM records, document conversations, collect sensitive documents, send communications, and decide where information will be stored. Each of those actions affects the quality and security of the company’s data.

An incomplete CRM record is not simply an administrative inconvenience. It becomes a data-quality issue that can affect marketing, customer retention, reporting, compliance, and future AI applications. A borrower document saved in an unapproved location creates governance and security concerns. Consumer information entered into a publicly available AI tool can create privacy, compliance, and reputational risks. A personal spreadsheet maintained outside company systems may feel efficient to the employee using it, but it can create problems for continuity, oversight, and data protection.

Most loan originators would not describe these activities as data governance, yet they participate in data governance every day. The same is true for processors, underwriters, closers, servicing professionals, sales managers, and executives. A company’s governance program is shaped by thousands of routine decisions made by people across the organization.

Publishing another policy will not solve this by itself. Employees need to understand why the rules matter and how those rules apply to their actual work. Training should use real situations that employees encounter, including which AI tools are approved, what information may be entered, how results should be reviewed, and who remains accountable for the final decision.

AI literacy is becoming part of professional competency. For a loan originator, responsible AI use may soon be as fundamental as understanding fair lending, advertising requirements, information security, and proper loan documentation. The technology will continue to become easier to access, which makes individual judgment even more important.

The encouraging news is that progress does not require every lender to begin with a massive technology transformation. A company can start by selecting one important workflow and examining how information moves through it. Where does the data originate? Who enters or changes it? How many systems contain the same information? Which steps require manual reconciliation? Where have employees created workarounds because the established process no longer meets their needs?

That exercise will usually uncover opportunities to improve the business before a new AI tool is purchased. A field may need a consistent definition. A report may need a reliable source. An approval process may need to move out of email. A recurring manual task may reveal that two systems are not exchanging information properly.

None of those improvements will generate the same excitement as announcing a new AI initiative. They may, however, save time, reduce errors, improve compliance, and give the company better information for future decisions. In a mortgage market where lenders are fighting for every basis point of profitability, small operational improvements matter, especially when they are repeated across hundreds or thousands of loans.

Companies also have industry resources available to help with this work. MISMO’s Framework for Responsible AI in the Mortgage Ecosystem, or FRAME, provides a practical starting point for evaluating and governing AI use. MISMO’s standards, Business Glossary, and Data Governance and Management work can help organizations create greater consistency around the information flowing through their businesses. Lenders can use these resources to move forward without inventing every definition, process, and control on their own.

AI may be receiving most of the attention, but many lenders have a more immediate opportunity in front of them. They can use this moment to examine how their businesses actually operate, identify where data and processes break down, and address those weaknesses before introducing more technology.

Before asking what AI can do for the company, leaders should ask whether the company is prepared for what AI will reveal. That work begins with leadership, but it cannot end there. It must reach every employee who enters, changes, stores, shares, or relies on the company’s information.

#VieauxPoint