How we built LendScore: Turning cash flow data into credit risk insights
Blog post from Plaid
LendScore is an innovative credit risk assessment tool developed by Plaid, leveraging cash flow data to complement traditional credit bureau information for a more comprehensive understanding of financial behaviors. By transforming unstructured transaction data into structured signals using Plaid's AI-powered categorization, LendScore distills these insights into a single score from 1 to 99, where higher scores indicate lower risk. The model, built on the XGBoost algorithm, incorporates both cash flow attributes and Plaid's unique Network Insights, which account for 81% and 19% of predictive power, respectively. LendScore is designed to address the challenges of compliance and fairness by employing monotonic constraints and independent fairness audits, ensuring equitable and transparent credit decisions. Additionally, it provides adverse action reason codes based on SHAP values to maintain regulatory compliance and explainability. This tool aims to enhance access to credit, particularly for near-prime and credit-invisible populations, while supporting lenders with enriched insights and performance across various loan types.