Learning-to-Rank in B2B e-commerce catalogs: A digital exper-iment and conversion analysis

Amid intensifying competition in the B2B e-commerce sector, particularly within the Do-It-Yourself (DIY) segment, traditional static search architectures increasingly suffer from limited adaptability and declining retrieval relevance. This study examines the limitations of rule-based ranking approaches and proposes a dynamic product ranking framework based on the Learning-to-Rank paradigm implemented with LightGBM. The primary objective of the research is to quantitatively evaluate the economic return on investment (ROI) associated with the deployment of personalized ranking algorithms. A simulation-based digital experiment was conducted using a synthetic user clickstream model to approximate real-world interaction behavior. The results indicate that the proposed dynamic ranking model yields significant improvements in search effectiveness, as measured by the metric, while simultaneously generating quantifiable gains in key business performance indicators. Specifically, the implementation resulted in a 2.1 percentage point increase in the conversion rate and a 14.5% uplift in incremental revenue. These observed effects achieved statistical significance. These findings provide empirical evidence supporting the economic viability of transitioning from static search systems to intelligent ranking architectures, highlighting their strategic importance for scalable and competitive B2B e-commerce platforms.