Assessing the Spillover Effects of Marketing Promotions on Credit Risk in Consumer Finance: An Empirical Study Based on AB Testing and Causal Inference

While marketing campaigns enhance customer acquisition in consumer finance, they may also induce unintended credit risk spillovers. This study develops a computer-aided causal inference framework to quantitatively evaluate the trade-off between conversion gains and risk exposure under various promotional strategies. Leveraging real-world A/B testing data from over two million users, the system integrates doubly robust estimation, causal forest modeling, and offline policy evaluation techniques into a unified computational pipeline. The modeling components are implemented using high-performance computing tools, including gradient-boosted decision trees and cross-fitted treatment effect learners, to capture nonlinear covariate interactions and estimate treatment heterogeneity. Furthermore, a counterfactual simulation engine is designed to evaluate policy effects under safety constraints based on Conditional Value at Risk (CVaR), allowing scalable strategy testing without online deployment. Experimental results show that promotions increase loan conversion rates by 12.3% but raise default rates by 1.8 percentage points on average. By deploying a CATE-based selective targeting strategy through the computational framework, the optimized policy reduces default rates by 9.7% while maintaining acquisition efficiency. The findings demonstrate the value of combining machine learning and causal inference with computer-based simulation platforms to support risk-sensitive decision-making in financial marketing. This research contributes a replicable computational approach for credit risk-aware campaign optimization, offering both methodological innovation and practical relevance for consumer finance institutions.