The CFPB's 2022 report put a number on a problem the lending industry has always known but rarely quantified precisely: 45 million Americans are "credit invisible" — they either have no credit file at the three major bureaus or their file is too thin to generate a scoreable FICO. An additional 19 million have files that exist but can't produce a score because the data is stale or insufficient. That's 64 million adults who approach a lender and are functionally invisible to the model making the decision.
For lenders trying to grow responsibly, this is not an abstract policy problem. It is a revenue problem that compounds every quarter you don't address it.
What "thin file" and "credit invisible" actually mean
These two terms are often used interchangeably, but they describe different scoring conditions and warrant different treatment in underwriting design.
Credit invisible means the bureau has no file on this individual. No tradelines, no inquiries, no public records. The person simply does not exist in the bureau's data infrastructure. This is most common among recent immigrants, young adults who have never opened a credit card, and people who have operated entirely in cash-based economies. FICO cannot score what it cannot find.
Thin file means a file exists — there may be one or two tradelines, possibly a single credit card opened recently — but the file lacks the density to generate a reliable score. FICO's minimum scoring criteria require at least one account that is at least six months old and at least one account reported to the bureau within the past six months. A file with only a two-month-old credit card fails this gate entirely.
The practical difference for lenders: credit invisibles require you to source credit signal from outside the bureau entirely. Thin-file applicants may be scoreable with alternative data augmenting a sparse bureau file. Your model architecture needs to handle both cases, and conflating them leads to poorly designed underwriting logic.
Who the 45 million actually are
The CFPB's research and independent demographic analyses consistently show this population is disproportionately:
- Young adults (18–24) — approximately 30% of this age cohort has no scoreable bureau file. They have income, they have banking relationships, they have spending patterns. They simply haven't opened a revolving credit account yet.
- Recent immigrants — even individuals with strong credit histories in their home countries start from zero in the US bureau system. A physician who immigrated from Brazil last year has no US credit file regardless of their financial behavior.
- Lower-income populations using alternative financial services — people who rely primarily on debit cards, prepaid cards, and cash transactions accumulate no bureau data even as they may demonstrate consistent and reliable financial behavior.
- Rural communities — areas with lower penetration of traditional credit products systematically produce fewer bureau tradelines, creating geographic pockets of thin-file populations.
Notice what this list is not. It is not a list of high-risk borrowers. It is a list of people who lack credit history — which is a different thing entirely. The conflation of "no credit history" with "bad credit risk" is the foundational error that makes this problem persist.
The cost to lenders of automated invisibility
Consider a digital lender or neobank running a bureau-based credit model. An applicant submits. The model calls the bureau. It comes back unscorable — FICO returns no score, or a score below the minimum threshold to process. Under most automated decisioning configurations, the application is declined automatically. No human review. No alternative data pull. Just a decline letter and an adverse action notice citing "insufficient credit history."
What the lender doesn't know: that applicant may have been paying $1,400 in rent every month for three years without a single missed payment. They may have five recurring utility accounts that have never gone to collections. Their bank account may show twelve months of consistent net positive cash flow with no overdrafts. None of that is visible to the bureau model. The lender declined a creditworthy borrower and gave the business to whoever is willing to look at the actual financial behavior.
Federal Reserve research has estimated that false decline rates in this population segment run between 30–45% of all automatic declines based on thin-file or unscorable results — meaning a significant fraction of the applicants you're automatically rejecting would have repaid had you lent to them. That's not a small edge case. At any meaningful lending volume, that number represents material revenue sitting on the table.
Why bureau expansion hasn't solved this
The bureaus have introduced several initiatives aimed at thin-file populations: trended data (adding 24-month payment history to tradelines), rental payment reporting programs, and utility payment reporting. These help at the margin but don't resolve the core problem.
Rental payment data is only available for the subset of renters whose landlords participate in bureau reporting programs — a minority. Utility payment data flows in through specific bureau data partnerships that have inconsistent coverage. Trended data helps existing credit users but does nothing for the truly credit invisible.
The deeper issue is structural: the bureau model requires historical credit product usage to score future credit risk. For populations that haven't used bureau-reported credit products, you can't bootstrap a meaningful score from within the bureau framework. You need to go outside it.
Alternative data as the answer — but not all alternatives are equal
The industry term "alternative data" covers a wide range: rental payment history, utility and telco payments, bank transaction data, employment verification data, and in some contexts, non-financial behavioral data. Not all of these are equally useful or equally fair for credit decisioning.
The most signal-dense alternative for creditworthiness is bank transaction data. This is not because it's novel — banks have been looking at checking account behavior for decades in their own internal models. It's because it captures the actual cash flow of a person's financial life: how much comes in, how regularly, how much goes out, what obligations are being met consistently. These patterns predict repayment behavior with meaningful accuracy on populations where bureau data is absent.
We're not claiming that alternative data replaces bureau scoring entirely for the population that has bureau data — for scoreable applicants, FICO contains decades of validated predictive signal that would be foolish to discard. What we are saying is that for the 64 million unscorable adults, the only responsible path is building models that can read the financial behavior actually present in their lives, rather than penalizing them for not having participated in the particular credit product ecosystem that the bureaus track.
Regulatory framing: ECOA and fair lending obligations
One concern lenders raise is whether using alternative data creates fair lending exposure under the Equal Credit Opportunity Act and Regulation B. This is a real consideration, but it cuts in both directions.
Systematically declining credit-invisible applicants without evaluating alternative creditworthiness signals creates its own disparate impact exposure, given the demographic composition of the credit-invisible population. A policy of "decline all thin-file applicants" is not a safe harbor from ECOA — it may in fact create a disparate impact pattern that warrants scrutiny.
Well-designed alternative data models, particularly those using bank transaction data rather than demographic proxies, can be evaluated for disparate impact using the same adverse impact ratio analysis applied to any other model feature set. The CFPB has issued guidance encouraging lenders to consider alternative data as a way to expand credit access — not as a regulatory trap.
Alternative data adverse action codes are an area where regulatory clarity is still evolving. Lenders using alternative data models need to build reason code logic that is specific, accurate, and sufficient to help an applicant understand the factors in their decision — the same standard that applies to bureau-based models, but harder to implement cleanly when the feature space is novel.
What this means for your underwriting strategy
If you're running a digital lending platform, neobank, or BNPL operation, the thin-file population isn't a niche edge case you can defer to a later product roadmap. It's a substantial portion of the applicant pool arriving at your application flow today. Every decline on a creditworthy thin-file applicant is a customer you're handing to a competitor who has built the infrastructure to evaluate them.
The decision to build alternative data underwriting capability is ultimately a bet on your total addressable market. Bureau-only models are accurate for the scoreable population. They're simply silent on a large and growing segment that represents real credit demand. Building the capability to hear that signal is not a charitable concession — it's a competitive advantage in a segment where the lenders who figure this out first will acquire a defensible and underserved customer base.
At Lendiro, we built our platform precisely because we spent enough time looking at this data to see that the decline decisions being made on thin-file applicants were wrong at rates that made no economic sense. The financial behavior that predicts repayment is present in these borrowers' lives — it's just not in the bureau file.
