Roughly 53 million U.S. adults are credit invisible, meaning no credit history exists at all, or credit thin-file, meaning their record is too sparse or stale to score reliably. About 28 million fall into the first group, another 25 million into the second. A separate study of auto loan defaults, using explainable machine learning, found that a model built on income level and employment type predicted defaults with 85% accuracy, well beyond what a simple days-past-due bucket would catch.
Put those two facts together and a clear picture emerges: a lot of auto loan risk sits outside what a credit score or a delinquency bucket measures. Here's what the research shows about where that risk lives and how predictive scoring picks it up.
Why Doesn't a Credit Score Tell the Whole Story for Auto Borrowers?
A credit score is a summary of past credit behavior, not a forecast of what a specific borrower can do this month. For the 53 million credit-invisible or thin-file Americans, that summary barely exists at all, which is part of why lenders have started looking at alternative data such as utility payments, rent history, telecom bills, and bank transaction data to fill the gap.
The results have been notable. FICO XD, one such alternative score, has been able to score roughly 9 million previously unscorable consumers at 620 or higher. One widely cited study found that lending models built on alternative data approved 43.4% more Black borrowers and 46% more Hispanic borrowers than traditional credit scoring, with roughly 25% lower APRs for approved borrowers.
What Predicts Whether an Auto Borrower Will Default or Cure?
Income level and employment type are consistently the strongest predictors in recent auto default research, stronger than most rules-based segmentation gives them credit for. One explainable-ML study found default risk peaking in a specific income band (around $13,500 in that dataset) and rising for borrowers in service or commercial-sector jobs, or those who were retired. Education and household structure mattered too, but at a smaller scale.
Days past due is still a useful signal. It just isn't the only one. Two borrowers at 30 days past due can have very different capacity to pay and very different odds of curing on their own, and a segmentation built only on delinquency stage can't tell them apart. A model that also weighs income stability and employment type can.
What Makes a Score Explainable?
An explainable model is one where you can see why it produced the score it did. The explainable-ML research used SHAP analysis, a method that shows exactly which factors drove a given score and by how much, so the model's reasoning could be inspected feature by feature rather than treated as a black box.
That visibility is useful for the people running a collections program day to day. When a team can see that an account scored high-risk because of a recent change in payment pattern, rather than just seeing a number, it's easier to trust the score and easier to decide what kind of outreach fits. EQ Engine's predictive scoring, payment projection, and collection probability models are explainable by design, so the reasoning behind each score is visible alongside the output.
How Predictive Scoring Changes Over the Life of an Auto Loan
A score set at loan origination and left alone doesn't reflect an account six months later. Borrowers' circumstances shift: a new job, a reduced schedule, a string of on-time payments after a rough start. Scoring that refreshes as new payment and account behavior comes in captures those changes.
EQ Engine's models update as new account behavior arrives, so a borrower flagged high-risk at the start but making consistent payments since doesn't stay stuck in that early bucket, and one who started strong but has since missed payments shows up earlier in the process.
Better scoring doesn't eliminate defaults. What it does is give a collections team a clearer view of which accounts are likely to resolve on their own and which ones a conversation could change.
Auto Loan Scoring FAQ
What factors best predict auto loan default? Recent explainable-ML research points to income level and employment type as the strongest predictors, more so than simple days-past-due status. Borrowers in service or commercial-sector jobs, or those who are retired, showed higher default risk in that research, with education and household structure contributing at a smaller scale.
What does credit invisible or thin-file mean? Credit invisible means a person has no credit history on file. Thin-file means there is a record, but it's too sparse or out of date to produce a reliable score. Together, the two groups include an estimated 53 million U.S. adults.
What is alternative data in auto lending? Alternative data is information outside a traditional credit report, such as utility, rent, telecom, and bank transaction history. Lenders use it to understand borrowers who have little or no traditional credit history.
What is explainable AI in auto loan scoring? Explainable AI, often implemented with techniques like SHAP analysis, shows which specific factors drove a given score rather than producing an output with no visible reasoning behind it.