The Thin-File Problem
In the United States, an estimated 50 million adults have no traditional credit score—either because their credit history is too thin or because they have opted out of the credit system entirely. Globally, the number exceeds 3 billion. These are not necessarily high-risk borrowers; they are simply invisible to conventional underwriting models.
Alternative credit scoring uses data sources that lie outside the traditional credit bureau ecosystem: bank transaction records, rent payments, utility bills, mobile phone top-ups, and even psychometric assessments. By analyzing cash-flow patterns rather than credit history, these models can identify creditworthy borrowers that FICO would reject.
Cash-Flow Underwriting
The most powerful alternative data source is bank transaction history. By analyzing income stability, spending patterns, and savings behavior, cash-flow underwriting models can assess credit risk with surprising accuracy.
Key signals include:
- Income regularity: Is the borrower’s income stable and predictable?
- Expense management: Does the borrower maintain a positive cash-flow margin?
- Savings behavior: Does the borrower consistently set aside funds?
- NSF frequency: How often does the borrower overdraft or bounce payments?
Cash-flow-based underwriting has been shown to predict default risk with accuracy comparable to traditional credit scores, while approving 27% more borrowers.
Regulatory and Fair-Lending Considerations
The use of alternative data in credit decisions raises important fair-lending questions. Machine learning models can inadvertently encode bias if training data reflects historical discrimination. Regulators are increasingly demanding explainability—the ability to articulate why a specific credit decision was made—which is challenging for complex ML models.
The emerging best practice is to use alternative data within a transparent, rules-based framework that can be audited for disparate impact. This combines the predictive power of alternative signals with the accountability that fair-lending regulations demand.
