I just came across a really thoughtful piece from HousingWire that gets at something we all bump up against in this business. The article explores what happens when mortgage lending meets artificial intelligence, and it's less about the technology working and more about a fundamental challenge the industry still hasn't cracked.
Here's the reality they're describing. When we're processing a loan, a borrower's income information might show up differently depending on where we're looking. Payroll records say one thing, the loan origination system shows something else, tax transcripts paint another picture, and bank statements tell part of the story but not all of it. AI can absolutely read all of this information, compare it, and make educated guesses about what probably happened. But mortgage lending isn't about probability. We need to know whether the available evidence actually supports what we're claiming to be true.
According to HousingWire, the industry has evolved from struggling to get data in the first place to now struggling to figure out what that data actually means when legitimate sources disagree. Processing teams still investigate discrepancies, underwriters still recalculate, quality control still checks the work. All of that friction tells us the real bottleneck has shifted from access to interpretation.
The article suggests we need something called "evidence adjudication" – essentially a governed system that decides whether the available evidence sufficiently supports a fact, separate from the decision about whether someone qualifies for a loan. This isn't underwriting. It's the foundation that underwriting rests on. And here's where it gets interesting. Sometimes the correct answer is to say the evidence is unresolved. That doesn't mean declined or fraudulent or ineligible. It means we can't yet justify relying on that fact with enough confidence.
The piece makes the point that we shouldn't just keep asking how many manual steps technology can eliminate. Instead, we should ask how well a system can identify when automation should stop because the evidence won't support moving forward. A system that forces every file through to completion, even by making unsupported assumptions, might create more problems than it solves.
What I'm seeing locally in the Bay Area and Fremont is that as lending continues to digitize and AI plays a bigger role, we're going to need to be really clear about what we actually know versus what we're assuming. The clients I work with want accuracy and honesty about their financial picture, and lenders need to be confident they're making decisions on solid ground. Getting rigorous about evidence and staying comfortable with uncertainty when it's warranted is going to matter more, not less, as these tools become more powerful.
