Fannie Mae’s new AI-powered crime detection unit: The Pulte/Palantir project

Fannie Mae has launched an AI-powered Crime Detection Unit in partnership with Palantir Technologies to achieve what FHFA/Fannie/Freddie chief Bill Pulte has called “something big on mortgage fraud.” Fannie Mae we all know is the grande dame of American Housing.

Less known is Palantir Technologies. It describes itself as “bringing the right data to the people who need it, allowing them to make data-driven decisions.” Palantir clients include some of the most secretive US intelligence agencies. They do a lot with data, and beyond that, I am unable to articulate without revealing my lack of understanding.

So with equal parts excitement and foreboding, this is what I see Palantir bringing to the mortgage fraud fight.

The right data

Palantir brings clients the “Right Data.” So what is that? Simply put, the “Right Data” for Palantir is everything. Everything means all the granular mortgage loan data, from a borrower’s private data to the very public property data. Then there’s macro industry data, from origination through to securitization. Anything short of that will hobble their ability to spot or prevent fraud.

Right data – Occupancy misrepresentation:

Let’s apply Palantir’s Right Data to the problem of mortgage fraud. The lion’s share of fraud that Palantir and Pulte find will be occupancy fraud, so let’s look at what the Right Data means in this context. 

Occupancy fraud is simple and pervasive. Borrowers falsely claim they will live in the home as a primary residence in order to qualify for lower interest rates and down payments.

Palantir will scrape public and private data sources for the myriad of clues left when someone lies about occupancy. For example, they will check rental sites like Airbnb to see if the property is listed there.

Did the borrower register a car, boat, or even a plane at a different address? Palantir will find the addresses for professional licenses, and voter registrations. It will scrub social media to find where our borrower eats, banks, gets dry-cleaning picked up. Which ATMs does the borrower frequent? Palantir will see if the borrower owns other properties and will check their value. It will compare square footage and then search for rental ads for any of the borrower’s investment properties. Ideally, Palantir will find and report on electricity usage, trash pick-up, plus the distance to golf courses or beaches. It should run a report on the commute distance between the borrower and the employer. 

Right now, Fannie’s fraud investigators are doing most of this by hand. It’s not that Palantir will be any smarter, but it will be more thorough and quite a bit faster. The CEO of Fannie Mae, Priscilla Almodovar, was quoted in HousingWire on May 28, saying that Palantir found fraud in “10 seconds” (possible) but “it took our really talented investigators 60 days” (no). Her statement is forgivable car-salesman puffery, but her point is taken: Palantir will be fast.

Right data – Asset misrepresentation:

Money launderers use real estate to clean dirty money. Therefore, the quality of asset documentation should be a primary concern of Palantir. There are commercially available tools now to verify the legitimacy of assets.

But Palantir’s data should do much more than just verify that a document is valid. It should analyze account balances against borrower profiles, check seasoning of funds and geographical red flags, verify instantly that grantors of gifts are related to the borrower (as required by gift letters), and bump the names of grantors up against the SDN list. 

Finally, with an AI super-tool like this, every recipient of cash-out can be similarly screened: after all, we don’t want our money to be disbursed to criminals. There’s no reason not to analyze every recipient of cash-out.

Asset fraud is like the caramel marbling in Ben & Jerry’s Salted Caramel Core – you dig, you’ll find it. Palantir could be excellent at this.

Ramifications: 

When I was a loan originator, I had a borrower lie to me about occupancy. I gave her a cash-out refinance on house A. But she used that cash to buy house B, which she promptly moved into. My QC department caught her and then our legal department landed on her. In the end, the rate was adjusted up and the borrower had to put more down.

The borrower thought it was a minor issue, but it left all of us bobbing in the wake, trying to fix a loan that was “not of investment quality.”

Finding fraud is like that – it’s the speeding boat whose wake upsets the static equilibrium of investors, correspondents, and insurers. 

And what if your findings create political problems – like discovering that your largest producer is involved? In April 2024, a top producing loan officer in New Jersey was indicted for falsifying routine loan information, as pedestrian as HOA reserve statements. 

Finding fraud in the abstract sounds like a righteous cause. But the reality is that it can be a righteous pain in the backside. 

Conclusion: 

I checked my DNA through 23andMe. I did the cheek-swab and waited for weeks. Finally, I got the email – my results were ready. I logged in and got this surprise warning – stop! you can’t unsee what you’re about to see! 

Oh, yeah, right! Think you’re Italian? Think again! Have mom’s eyes? Adopted! Heir to the throne? More like the scullery maid.

Like an unexpected parentage reveal on Maury Povich, Palantir will uncover frauds in your loan files from which you would have preferred to remain estranged. 

Yes, it will be fast. Yes, it will be efficient. But it will also be heartless and right quick about it.

The actions by Pulte and Palantir, uncovering mortgage frauds in all their genetic variations, will demand uncomfortable actions – repurchase demands, rate recalculations, and tense investor conversations.

I’m for it – but a word of advice – don’t plug this system in without bracing yourself. You may be adopted.

Bob Simpson is the CEO of DaylightAML.

This column does not necessarily reflect the opinion of HousingWire’s editorial department and its owners.

To contact the editor responsible for this piece: zeb@hwmedia.com.