Back to Insights
PressPrudent Perspectives

August 20, 2026

Not All AI Is the Same: What "AI-Powered" Actually Means in Mortgage

By Suha Beidas Zehl, CMB®, AMSP™

Cover image for the article Not All AI Is the Same: What "AI-Powered" Actually Means in Mortgage

Welcome to Prudent AI’s monthly newsletter. Every edition, we bring you real stories, industry data, and practical insights from the front lines of mortgage lending — so you can make smarter decisions faster.

— Suha Beidas Zehl, CMB®, AMSP™ , Brand Ambassador, Prudent AI

Fannie Mae’s Lender Letter LL-2026-04 took effect on August 6. Freddie Mac’s Guide Section 1302.8 has been in force since March 3. On request, you now have to disclose the types of AI you’re using, the purpose and manner of that use, and the safeguards behind it. And guess what? That responsibility sits with you, not the vendor who sold you the model.

Harder to answer than it sounds, because "AI-powered" covers two very different things.

Ask any vendor what makes their platform "AI-powered" and you’ll hear something like this: it reads documents faster, it answers borrower questions, it flags exceptions for review. Yes, each of those is useful. But none of them is a decision.

Let’s start with where technology actually sits. STRATMOR’s 2025 Technology Insight Study, released in June, found that the two leading AI use cases among lenders are document classification and indexing at 68%, and document reading at 59%. STRATMOR calls both back-office functions.

That work is worth doing. But let’s call a spade a spade, ok? Those tools move files faster through a process whose judgment calls are made somewhere else. Intelligence, in the sense that actually changes an outcome, sits at the decision point, where income gets interpreted; a file gets structured, and the loan either moves forward or stalls. Both are worth buying. But they do not carry the same risk, and they should not share a label.

This industry has blurred that difference before. Anyone who was originating when automated underwriting arrived remembers what Desktop Underwriter and Loan Prospector, now Loan Product Advisor, actually delivered: manual review standardized and accelerated at a scale nobody had matched. They applied credit policy consistently and scored risks against it. That was enormously valuable. But it was not judgment, and the language of the day often suggested otherwise.

The vendor conversation needs to change. STRATMOR found 53% of AI-using lenders partner with third parties for purpose-built solutions, while native LOS capabilities account for just 6% of current usage. For most of us, the AI in question is someone else’s model running inside someone else’s product. So the question isn’t whether a vendor uses AI. It’s where their AI sits in your process.

Faster OCR and a borrower-facing chatbot? That’s automation wearing an AI costume. Useful automation, but a costume all the same. A demo that shows you speed is showing you the costume. A demo that shows you a changed outcome is showing you the decision. Speed without judgment is just faster mistakes. So the sharper question isn’t whether a system can move faster. It’s whether it knows when to hand off to a human.

Non-QM is where that line shows up most clearly. QM is standardized enough that AI applied to it is largely rule automation. Non-QM requires weighing layered, conflicting signals the way a seasoned underwriter does.

Here’s the other reason the vendor conversation has to change: explainability is the new currency. In short, a model that’s right 95% of the time is still indefensible to a regulator or an investor if no one can explain why it decided what it decided. Nearly half the lenders in The Mortgage Collaborative’s June survey already reported fair lending concerns tied to AI decisioning, and those concerns don’t resolve themselves at the vendor’s end.

If your AI doesn’t sit at the decision point, it may still be worth buying. But buy with your eyes wide open. Automation makes the process faster. Not necessarily better.

So, what should "AI-powered" mean in mortgage? Not a checkbox. Not a trust signal. A designation. And like every other designation in this business, it has to be earned at the decision point.

eRESI Mortgage, LLC 's non-delegated correspondent sellers now run income analysis on Prudent AI’s Upfront Income platform — bank statement, self-employed, wage-earner, and rental income. The evaluation happens at submission, not deep in underwriting.

"Income analysis is one of the biggest friction points in Non-QM lending," said Lisa Schreiber , SVP at eRESI. "This partnership with Prudent AI helps them move faster, with more confidence, from the moment a file is submitted."

Because income is validated consistently and documented at submission, files are better positioned for investor review, downstream. This brings up fewer surprises for the seller, and cleaner sales at purchase.

Read the full announcement on PROGRESS in Lending Association : https://mymortgagemindset.com/eresi-partners-with-prudent-ai-to-bring-fast-upfront-income-analysis-to-non-qm-sellers/

Delfino Aguilar built Kind Lending’s Wholesale Production division from the ground up. Five years in, it’s one of the top five wholesale lenders in the country. He’s Chief Production Officer now, running Wholesale, Retail, and Third-Party Origination — known for strong broker relationships and for a leadership style that measures

On this episode of #TheUpfrontRadio, Delfino talks with Suha about where AI stops and judgment starts at Kind Lending | NMLS #3925 . AI takes on the noise around a file — validating income documentation, identifying file types but not the risk decision itself. Once the data’s calculated, a person still clears it: speed doesn’t make a wrong number right.

That discipline shows up in the numbers. Kind Lending estimates a 20 to 30 percent production lift, driven by catching problems before they compound rather than by processing files faster. And the redesign behind it wasn’t handed to executives or AI but with the staff who actually touch the files in the room making the calls.

Watch the full episode: https://www.youtube.com/watch?v=O6PTIEbkvFQ

Subscribe to stay ahead of what’s next in mortgage technology, lending operations, and the AI developments reshaping how loans get closed.

This is such a timely topic ... thank you for publishing my article. Commenting to share with my network #CFR

Putting This Into Practice

This is the work I do with lenders, banks, and credit unions. If your organization is working through these questions, these pages explain how I help.