Every loan officer knows the speech.
“I looked at everything. I ran it every way I could. I’m sorry, the numbers just don’t work.”
I was ready to give it. Then I asked AI for a second opinion, and it found something in the lender’s guidelines that I had skated right past.
Not a loophole. Not a hack. A provision that had been sitting there the whole time, waiting for someone to actually read it.
Which raises a question far more useful than anything in this week’s AI headlines: What happens when another loan officer finds the yes you walked away from?
The income was there. I just didn’t count it
I was working through a borrower’s income calculation and had AI check my numbers against the lender’s guidelines. It came back with qualifying income I would have left on the table.
That extra income improved the debt-to-income ratio. In this case, it was enough to get the loan approved.
Let’s be clear about what did not happen. The borrower didn’t get a raise. AI didn’t approve anything, invent income or bend a rule. The income was already there, and the guidelines already allowed it to be counted. AI just read the fine print more patiently than a human juggling a full pipeline.
Without that second look, I would have delivered the speech. The buyer would have walked. And nobody would ever have known the loan was approvable.
That last part is what bugs me. Declines don’t come with receipts.
Your competition isn’t the robot
I’m not worried about a chatbot taking my job. I’m worried about the loan officer who learned to use one while I was busy ignoring it.
Experience matters. Relationships matter. Neither one can search hundreds of pages of guidelines on demand at 9 p.m. on a Tuesday.
And this goes way beyond writing prettier emails. Fannie Mae’s Ask Poli already helps mortgage pros pull answers from its Selling and Servicing Guides, lender letters and eligibility matrix.
The industry knows where this is heading. In Fannie Mae’s 2023 Mortgage Lender Sentiment Survey, 73 percent of lenders already using AI or machine learning named operational efficiency as their primary goal, up from 42 percent in 2018.
Worth noting: That’s what lenders are aiming for, not proof it’s working. The real question is what you do with the time you get back. If the answer is “the same thing, faster,” congratulations, you bought a faster treadmill. Spend it on the better borrower conversation, the second option, the clearer answer for the agent.
Agents, stop asking the wrong question
“Does your lender use AI?” is a yes-or-no question, and you’ll get a yes. Everybody uses AI now, the same way everybody “does social media.”
Ask this instead: “When the numbers don’t work, what do you check before you tell my buyer ‘no’?”
Then listen for a process, not a pep talk. A good answer names the guideline, spots what’s missing and shows how the income was calculated. A weak answer sounds like, “Trust me, I’ve been doing this 20 years.”
I’ve been doing this 20-plus years myself. I still missed one.
Not every tough file can be saved. But some of those no’s are really “Not yet. Let me look again.” If you’re the agent, that’s your buyer and your commission riding on which one it is.
How to use AI without getting burned
None of this means tossing a file into a chatbot and believing whatever comes back. Here’s how I’d start:
- Start with a hypothetical. Build one made-up scenario and run it in a company-approved tool.
- Keep borrower data out of unapproved systems. Non-negotiable. No exceptions for “just this once.”
- Make it cite its sources. Ask it to compare the scenario against specific guidelines, cite the provisions and flag missing information. No citation, no credit.
- Check its homework. Verify the source, the program requirements and the math. Even Fannie Mae tells Ask Poli users that its official guides and updates win when the tool conflicts with them.
- Make it argue against you. Ask what could make the loan ineligible. You want a sparring partner, not a hype man.
So what about the LO sitting this out?
Not using AI is a choice. That’s fine. Just make it on purpose, not out of habit.
I’m not asking anyone to buy the hype. I’m asking you to test these tools against the work you actually do.
For me, that test turned a no into an approval. Somewhere in your pipeline, there’s probably a file that deserves the same second look.
Go find it before the loan officer down the street does.
This October, Inman turns its focus to AI and its rapid rise in real estate. During Artificial Intelligence Month, we’re digging into the startups shaking things up and the established players folding AI into their offerings.
Paul Apostolakis is a Michigan mortgage professional with more than 20 years in mortgage origination, sales leadership and mortgage technology. NMLS #1467435. Connect with him at his website or Bolt Home Loans.
