Lenders Find Creditworthy Borrowers Hiding in the Data

A credit file doesn’t capture every signal lenders can use to judge whether a borrower will repay.

Roughly 25 million U.S. adults lack enough recent credit activity to generate a usable score, the Consumer Financial Protection Bureau (CFPB) noted last year, as PYMNTS reported. PYMNTS Intelligence separately found that 35% of subprime consumers have neither a credit nor store card, compared with 12% of prime consumers and 4% of super-prime consumers.

Bank transactions, cash flow and more detailed credit histories are yielding additional information for lenders to supplement conventional measures. Recent underwriting results disclosed by providers also provide some indication of how often the extra information changes the decision, and translates into a widening pool of available credit.

Personify Financial, per Plaid, says verified bank-transaction data increased approvals by up to 8% among underserved consumer segments while maintaining credit costs. Affirm’s latest underwriting model generated 3.4% more completed purchases at comparable risk by approving eligible applicants its previous system had declined. Personify added current bank transactions. Affirm’s transformer model extracts more detail from the order and timing of events within consumers’ credit histories.

Plaid has separately tested the same broad category of information in its LendScore model. Its published results show a 73% approval rate versus 65% for a traditional benchmark at equivalent risk. At the same approval rate, Plaid reported 41% lower risk. The firm also notes that there had been up to 5x more approval for thin-file consumers at what it termed “near-prime default rates.”

Affirm is using additional information in two ways. Its transformer model looks for patterns within detailed credit histories that can disappear when borrower behavior is reduced to summary measures. The company said the incremental loans generated by the model performed better than a comparable expansion under its previous machine learning models.

PYMNTS Intelligence data illustrate why current financial information would offer a more holistic view of an applicant, and would be useful across several types of financial offerings. Data contained in the subprime consumer research identified about 44 million U.S. adults as subprime, or 17% of consumers. Fifty-five percent lived paycheck to paycheck with difficulty paying bills, while 35% held no credit or store card. Buy now, pay later (BNPL) use among subprime consumers was 19%, compared with 13% across the full sample.

A Federal Reserve review estimated that about 32 million U.S. adults are unscoreable, including 7 million with no credit history and 25 million with thin files. The Fed identified cash-flow data as one source that can provide information such as regular deposits, average balances and overdraft history. But it also cautioned that alternative-data models can face data quality and cost issues, and that many haven’t been tested through a full business cycle.

Accounting Data Enters Small Business Underwriting

For business lenders, additional information can come directly from a company’s accounting records.

A PYMNTS Intelligence report, “Keeping Score: Why Data Quality Determines Lending Decisions for the Smallest Firms,” based on 350 financial institution executives in the U.S. and U.K., found 57% cited inaccurate or incomplete business records as an underwriting obstacle. Outdated information was cited by 52%, inconsistent information by 52% and limited financial history by 48%.

The same research found microbusiness credit applications had a 77% approval rate, compared with 95% for large enterprises. About 19% of applications were rejected because they lacked enough information to verify creditworthiness.

Sage and Satago are putting structured accounting information into invoice finance underwriting for sole traders and microbusinesses. The expanded partnership announced over the summer gives lenders real-time access to invoices, transactions, ledger data and cash-flow information and uses the same data for ongoing monitoring.