The LOOP Framework: A Closed-Loop Machine Learning Approach To Customer Engagement in Fashion & Apparel Retail

Abstract

In retail customer analytics, machine learning applications have largely focused on customer segmentation, predicting customer engagement, and customer sales as separate, disconnected, and parallel components of a system rather than as components of a dynamic system. This study proposes and empirically validates a new, closed-loop machine learning framework for customer engagement in fashion and apparel retail, extending the grocery-sector RAISE framework, to a domain that it does not cater. The study comprises 3,291 customers and 10,000 transactions for a two-year period, with the segmentation of customers run using the K-means customer-state segmentation, followed by an XGBoost engagement-response model, a Causal Forest heterogeneous treatment effect analysis, and a formal chi-square test of segment-transition structure. The study uses K-means as the customer-state segmentation, an XGBoost engagement-response model, a Causal Forest heterogeneous treatment effect analysis and a formal chi-square test of segment-transition structure. Results indicate that the transitions between the customer engagement states are non-random over time (χ² = 616.74, p < .001, Cramér's V = 0.340); that the engagement effects of discounts are heterogeneous and patterned by customer age, but not by value segment (ATE = 0.028, 95% CI [0.026, 0.029]); and that the pattern of engagement effects is partially moderated by the return-proneness of products in each product category. The study fills a gap in the literature in retailing and introduces an empirical foundation for an age-differentiated, not uniformly applied, engagement strategy for retailers.