Ask most people where credit invisibility is a serious problem and they'll point somewhere overseas. That instinct is incomplete. The Consumer Financial Protection Bureau put out a technical correction in June 2025 that quietly reframed this issue for the US, and it's the kind of update that deserves more attention than it got.
Here's what happened. Back in 2015, the CFPB estimated that 26 million Americans were credit invisible, meaning they had no credit file at all with any of the major bureaus. That number got cited everywhere for a decade. Then the Bureau went back and found a real error in its own data. Records tied to deferred student loans, old collections, and closed accounts had been left out of the original analysis, which inflated the invisible count substantially. Once corrected, the actual 2020 figure came in at 2.7 percent of adults, about 7 million people.
What grew instead is the group that matters more. The CFPB now says 9.8 percent of American adults, roughly 25.3 million people, have a credit file that exists but can't produce a usable score.
We now have a clearer view of how many consumers are actually unscorable.Jennifer Tescher, Financial Health Network
These people aren't invisible to the system. They're sitting right there in the data, unscoreable, and lenders treat them functionally the same way they'd treat someone with no file at all.
Where my own research fits, and where it doesn't yet
I read that update and thought about my own paper on securitized auto loans, which set out to answer a version of this exact question. Strip a lender's model of the borrower's credit score entirely, and does the model fall apart? Inside my sample, it didn't. Loan to value ratio, vehicle age, contract term, and payment burden all kept meaningful predictive power on their own, and removing score dropped the model's fit only modestly, from a pseudo R squared of 0.087 down to 0.059.
I want to be precise about what that finding actually covers, because the honest boundary matters here. My sample consists of loans that were already originated and securitized, meaning every borrower in it had already cleared some underwriting screen. It's a genuinely open, untested question whether the same collateral based approach would work as well extending credit to someone who's currently unscored and has never been approved for anything at all. What I can say is that the mechanism held up among borrowers with varying degrees of score visibility. Whether it extends cleanly to the fully unscored population is a claim worth testing directly, not one I've already proven.
There's already a live legislative signal that Washington is thinking about this problem too, through a different lever. Senator Tim Scott, who chairs the Senate Banking Committee, introduced the Credit Access and Inclusion Act, which would let the credit bureaus incorporate rent, phone, and utility payment data into scoring models. That's a different mechanism than collateral based underwriting, adding new inputs to a score rather than substituting for one, but it's aimed at the same 25 million people, and it confirms this is a live policy conversation, not a fringe concern.
The two approaches aren't competing. Rent and utility reporting depends on payment data getting collected and reported consistently, which takes time to build. Collateral based underwriting doesn't require any new infrastructure, the loan to value ratio and the amortization schedule are already sitting inside every auto loan a lender originates today. If my research's mechanism does extend to the unscored population the way I suspect it might, lenders wouldn't need to wait for a new reporting pipeline to responsibly serve borrowers who are creditworthy right now and simply lack the file to prove it. That's the test I'd want to run next, on a sample that actually includes declined and unscored applicants, not just approved ones.