Data-Driven Decision Making in Logistics: An Integrated Predictive–Inferential Framework for Operator Selection

Background: This paper examines customer satisfaction in postal and logistics systems through an integrated predictive–inferential approach. The main objective is to extend an existing comparative machine learning framework for operator selection by adding an inferential layer to examine differences among customer groups. Methods: The study uses customer-profile data comprising demographic, geodemographic, shipment-related and preference-related characteristics, with customer satisfaction as the dependent variable. In the baseline framework, Linear Regression, Random Forest, and Gradient Boosting were trained and evaluated independently to predict customer satisfaction and support operator selection. The inferential extension applies one-way ANOVA, Welch’s ANOVA where variance homogeneity is violated, Holm adjustment for multiple testing, Kruskal–Wallis robustness analysis, and appropriate post hoc comparisons. Results: Statistically significant differences in mean satisfaction were identified for two of the nine examined attributes: customer category and expected service delivery time. The robustness analysis additionally identified differences across respondent groups defined by gender and organizational status; however, this finding was not supported by the Holm-adjusted primary ANOVA and was treated as a secondary result. Conclusions: The proposed extension complements individual-level satisfaction prediction and predictor interpretation with formal group-level comparisons, providing a basis for customer segmentation, service-level management, and data-driven operator decision-making.