For most of American financial history, the FICO score has functioned as an all-purpose proxy for creditworthiness. It is convenient, deeply embedded in underwriting workflows, and widely accepted by regulators. It is also structurally blind to roughly 45 million Americans who maintain real financial lives — paying rent on time, keeping utilities current, covering insurance premiums — but happen to lack the specific kind of credit history that FICO's algorithm requires. That exclusion is not a bug FICO's developers missed. It is a structural consequence of how the model was designed.
Understanding why FICO fails this population requires understanding what FICO actually measures — and more importantly, what it refuses to measure.
What FICO Actually Scores
FICO's five input categories are widely known: payment history (roughly 35%), amounts owed (30%), length of credit history (15%), new credit (10%), and credit mix (10%). Every single one of these depends on data that exists inside the three major credit bureaus — Equifax, Experian, and TransUnion. If a consumer has never opened a credit card, taken out a car loan, or held a mortgage, the bureaus have nothing. No data, no score. FICO returns a null, the lender declines, and a creditworthy borrower walks away empty-handed.
This matters because the bureau data ecosystem is self-reinforcing. You build a credit file by using credit. If access to credit is gated by a credit file, the population that never received credit in the first place has no ladder in. The CFPB has documented this dynamic extensively in its research on credit invisibility, noting that the segments most likely to be thin-file or credit-invisible are concentrated among younger adults, lower-income households, recent immigrants, and communities of color. This is not an edge case. It is a structural feature of how traditional credit decisioning allocates access.
The Thin-File Problem in Practice
Consider a scenario that is entirely realistic: a 34-year-old warehouse logistics coordinator in a mid-size Texas city. She has rented the same apartment for six years, paying monthly rent via bank transfer. She maintains an active checking account with a regional credit union, deposits a bi-weekly paycheck from a steady employer, pays her car insurance and phone bill automatically every month, and carries zero revolving debt because she has never applied for a credit card. Her FICO score is either non-existent or falls in the 580-620 range due to a single medical collection from 2019. A FICO-based underwriting system sees a marginal borrower. A 24-month cash-flow view sees an extremely reliable one.
That gap — between what the data says and what the FICO score reports — is where thin-file risk assessment breaks down. The score is not wrong given its inputs. The inputs are simply inadequate for this population.
Why 24 Months of Cash-Flow Velocity Changes the Analysis
Cash-flow based underwriting approaches the creditworthiness question differently. Rather than asking "what does this person's credit bureau file say," it asks "what does this person's actual financial behavior reveal over a sustained observation window?"
A 24-month lookback window is meaningful because it captures the full cycle of real financial life: seasonal income variation, expense shocks, recovery patterns after disruption. A 12-month window misses the longer-term trend signal. A 36-month window adds friction without proportional lift on most borrower segments.
The specific signals that carry predictive weight in cash-flow underwriting include:
- Income consistency: Are inflows recurring on a predictable schedule? Irregular inflows are not automatically disqualifying — gig workers and self-employed borrowers often have variable income but consistent monthly totals — but pattern stability over 24 months is highly predictive.
- Fixed-obligation coverage: Does the borrower consistently cover known recurring obligations (rent, utilities, subscriptions) with comfortable headroom? A borrower who reliably covers $1,200 in monthly fixed obligations on an average deposit of $2,400 shows a materially different risk profile than one with the same average balance but high variance in obligation coverage.
- NSF and overdraft frequency: Non-sufficient funds events and overdraft incidents are among the strongest predictive signals in cash-flow models. Not a single NSF in 24 months is a meaningful positive indicator. Frequent NSF events, even in months with otherwise adequate deposits, suggest cash-management risk that bureau data would entirely miss.
- Liquidity cushion: The average minimum balance between paydays — not the average balance, but the trough — tells you about a borrower's ability to absorb unexpected expense without defaulting on obligations.
Statistical Validation: Does Cash-Flow Data Actually Predict Default?
Skepticism here is warranted and healthy. The claim that cash-flow signals predict default better than FICO in thin-file populations needs to be substantiated, not assumed. The honest answer from the industry is that evidence is still accumulating, and the picture is more nuanced than proponents sometimes acknowledge.
What the evidence does support: for populations where FICO scores are either unavailable or cluster in ranges where the score has low discriminatory power (roughly 580-640), cash-flow signals provide meaningful lift in AUC-ROC measures. Academic studies using bank account data as underwriting inputs — including work published in the Journal of Financial Economics and by the CFPB's Office of Research — have found Gini coefficients for cash-flow models in thin-file populations that are competitive with FICO performance in scoreable populations.
We're not saying FICO is useless or that cash-flow signals are universally superior. For well-file prime borrowers, FICO remains an efficient, auditable, and well-validated model with decades of vintage data behind it. The argument is narrower: in the specific segment where FICO has no data, cash-flow signals offer a statistically meaningful alternative that lenders can validate against their own portfolio outcomes.
Regulatory Framing: CFPB Guidance on Alternative Data
The CFPB has expressed measured openness to alternative data in underwriting, particularly for credit-invisible and thin-file populations. Its 2017 request for information on alternative data and its subsequent guidance have acknowledged that transaction account data, rental payment history, and similar signals can expand access to credit without violating ECOA's prohibition on disparate impact — provided lenders conduct appropriate validation.
The critical qualifier is "appropriate validation." The CFPB and OCC both require that any model used in credit decisioning be validated for both predictive accuracy (SR 11-7 guidance from the OCC is the standard reference) and fair lending compliance. A cash-flow model that improves approval rates for thin-file borrowers but disproportionately disadvantages protected classes would not solve the problem — it would replace one form of exclusion with another. Disparate-impact testing against race, national origin, sex, and age is not optional; it is a legal requirement embedded in ECOA and its implementing regulation, Regulation B.
This is not a reason to avoid cash-flow underwriting. It is a reason to build it carefully and to require that any vendor providing cash-flow decisioning infrastructure delivers the model documentation, reason code structure, and audit capability needed for Regulation B compliance.
What Responsible Deployment Looks Like
Lenders integrating cash-flow signals into underwriting need to approach it as a supplement to existing decisioning infrastructure, not a wholesale replacement. The practical deployment path for most community lenders and fintech lenders looks something like this: use FICO when a scoreable file exists and the score carries sufficient discriminatory power; trigger the cash-flow decisioning layer when FICO returns null or falls in a range where the score's Gini coefficient drops significantly; combine outputs under a documented decisioning policy; and generate adverse action reason codes from whichever model governed the decision.
The 45 million Americans that FICO cannot see are not a monolithic group. Some are young adults with no credit history who are statistically low risk. Some are immigrants with strong foreign credit histories that domestic bureaus can't translate. Some have prior derogatory marks that are now stale but still score-suppressing. Cash-flow signals do not resolve all of these equally well — vintage analysis on the actual borrower cohort matters enormously. But for the large segment of thin-file borrowers whose exclusion is simply a data gap rather than a genuine risk signal, cash-flow underwriting offers a durable, explainable, and regulatory-compatible path to inclusion.
That path requires infrastructure that lenders can integrate without rebuilding their loan origination system from scratch, that delivers adverse action outputs in a format Regulation B accepts, and that has been built with disparate-impact monitoring rather than as an afterthought. That is the gap Lendiro is built to close.