FinRegLab Research
FinRegLab is a nonprofit research organization that conducts independent, peer-reviewed studies on financial technology and inclusion. Their research on cash-flow data for credit underwriting is the most rigorous available and is used by regulators including the CFPB.
Key Studies
1. Empirical Research Findings (2019)
Independent analysis conducted by Charles River Associates using loan-level performance data from six non-bank lenders: Accion, Brigit, Kabbage, LendUp, Oportun, and Petal.
"FinRegLab retained Charles River Associates to help design and conduct an independent analysis of the predictiveness of the participants' cash-flow variables and scores based on actual loan performance."
Source: FinRegLab Empirical Research Findings
2. Small Business Spotlight (June 2025)
Analysis of over 38,000 small business loans from two fintech lenders between February 2015 and January 2024.
"The empirical analysis examined over 38,000 small business loans from two fintech lenders between February 2015 and January 2024."
Source: PRNewswire - Small Business Lending Study
3. Machine Learning & Cash Flow Study (July 2025)
Comparison of machine learning models with cash flow data versus traditional logistic regression.
Source: FinRegLab ML & Cash Flow Study
Most Predictive Cash Flow Variables
Deposits and Balances
"Account deposits and balances were generally the most powerful variables."
Source: PRNewswire - Small Business Lending Study
Additional Predictive Variables
"Factors such as withdrawals, balance volatility, and low/negative balance incidents were also predictive."
Source: PRNewswire - Small Business Lending Study
Typical Model Features
"Models typically use 10 to 20 features covering areas such as low or negative balance events, stability of cash inflows, balance trend over time, number and amount of recent cash outflows, and debt-to-income proxies."
Source: FinRegLab Research Summary
Specific Metrics Used
Liquidity Indicators:
"The current ratio (balance divided by median monthly outflow) and expense-to-income ratio (median outflow divided by median inflow) quantified cash flow adequacy."
Behavioral Features:
"Transactional activity was quantified through recency, frequency, and ticket size, while affluence was inferred from the number of large purchases."
Variable Categories
Model Performance Findings
Cash Flow vs. Traditional Credit Scores
"The predictiveness of the cash-flow scores and attributes was generally at least as strong as the traditional credit scores and credit bureau attributes studied."
Combined Models
"Models that added cash-flow variables to personal credit scores and other traditional inputs predicted default risk more accurately across all borrower segments analyzed."
Source: PRNewswire - Small Business Lending Study
Machine Learning + Cash Flow (Best Performer)
"Across all models built for the study, the machine model that combined credit bureau data with cash flow data was the most predictive overall and across all subgroups. It also had the highest approval rates overall and for most subgroups at most risk thresholds."
Source: PRNewswire - ML & Cash Flow Study
ML Outperforms Traditional Methods
"Machine learning models substantially outperformed traditional logistic regression methods across all types of data (cash flow data only, credit bureau data only, or both sources combined)."
Source: PRNewswire - ML & Cash Flow Study
Approval Rate Impact
Quantified Improvement
"Machine learning models increased credit approvals by approximately 4% compared to traditional logistic regression at risk cutoffs used by mainstream lenders."
Source: PRNewswire - ML & Cash Flow Study
Real-World Translation
"This translates to roughly 2 million additional credit card accounts and 152,000 extra mortgages annually (based on 2023 origination volumes)."
Source: PRNewswire - ML & Cash Flow Study
No Increase in Default Risk
The study found that these approval increases came "without increasing lenders' default risk" and with "relatively low false positive rates (approvals of consumers who went on to default)."
Source: PRNewswire - ML & Cash Flow Study
Where Cash Flow Data Helps Most
Target Populations
"Gains were particularly large for low-score entrepreneurs whose businesses are less than five years old."
Source: PRNewswire - Small Business Lending Study
Specific Beneficiaries
"The effects are larger for low-score owners whose businesses are young (less than five years old). The results suggest that adopting cash-flow data could be particularly important in increasing lenders' confidence when underwriting financially constrained business owners that have historically been considered high risk."
Source: FinRegLab Sharpening the Focus
Subpopulation Analysis
"When divided into subgroups based on likely race, ethnicity, and gender, the degree to which the cash-flow data predicted credit risk appeared to be relatively consistent across subpopulations."
Source: FinRegLab Empirical Research Findings
Not a Proxy for Demographics
"The cash-flow data appeared to provide independent predictive value across all groups rather than acting as proxies for demographic group."
Source: FinRegLab Empirical Research Findings
Market Context
Credit Gap
"An estimated 45 to 60 million consumers lack sufficient credit history to generate reliable credit scores."
Source: FinRegLab Empirical Research Findings
Bank Account Penetration
"More than 96 percent of American households have bank or prepaid accounts."
Traditional Lender Populations
"Among participating lenders where data was available, the percentage of borrowers with traditional credit scores below about 650 was approximately 45 to 50 percent."
Small Business Impact
"Small businesses created two out of every three net new jobs over fifteen years, yet racial minorities, recent immigrants, and women face particular challenges in accessing credit."
Source: FinRegLab Small Business Fact Sheet
Data Sources for Small Business Underwriting
"Use of electronic cash-flow data was originally pioneered by fintechs that used electronic bank account records, accounting software feeds, and payment records to facilitate faster underwriting of smaller loans to small business owners."
Source: FinRegLab Small Business Spotlight
Specific Data Types
Bank account records
Deposit/withdrawal history, balances
Accounting software feeds
QuickBooks, Xero, etc.
E-commerce transaction data
Shopify, Amazon, eBay sales
Payment processor data
Stripe, Square, PayPal transactions
Source: FinRegLab Small Business Fact Sheet
Application to Shopify/Stripe/Square Data
Based on FinRegLab's validated findings, the following metrics from e-commerce platforms map to proven predictive variables:
Total revenue
Deposit/inflow levels
"Most powerful"
Revenue trend
Balance trend over time
Predictive
Transaction count
Transaction frequency
Predictive
Average order value
Ticket size
Predictive
Revenue variance
Balance volatility
Predictive
Refund rate
Outflow patterns
Predictive
Success rate
Cash flow stability
Predictive
Time on platform
Account tenure
Predictive
MRR (Stripe)
Income stability
Predictive
Key Takeaways
Cash flow data is at least as predictive as traditional credit scores for default risk.
Combined models (cash flow + traditional) outperform either alone across all borrower segments.
Machine learning further improves performance when combined with cash flow data.
The biggest gains are for underserved populations: low credit scores, young businesses, thin-file borrowers.
No fair lending concerns: cash flow data provides independent predictive value, not demographic proxies.
E-commerce data (Shopify, Stripe, Square) maps directly to the validated cash flow variables.
References
Next: Cashflow Risk Model | Risk Scoring Overview | Social Trust Scoring
Last updated