An omnichannel strategy development framework leveraging customer opinion divergence using large language models and explainable AI

Abstract

Successful omnichannel retail management requires integrated strategies because customers evaluate product aspects differently across online and offline channels. However, prior research has relied on macro-level or qualitative analyses that cannot quantify which specific aspects are associated with customer satisfaction in each channel, limiting efforts to mitigate strategic gaps across channels. We propose an aspect-level omnichannel strategy development framework. First, we use a novel LLM (large language model)-based method to extract and normalize product aspects from customer reviews, thereby identifying the aspects that customers consider important in each channel while overcoming the limitations of predefined taxonomies and external tools. Next, we quantify each aspect’s contribution to customer satisfaction using SHAP (Shapley additive explanations)-based XAI (explainable AI). Finally, strategies are developed to mitigate the satisfaction gaps based on the quantified aspect contributions. We validated the framework on 6.8 million customer reviews from South Korea’s largest health and beauty omnichannel retailer. The proposed normalization method achieved a 24.70 percentage point improvement in accuracy over existing state-of-the-art LLM-based methods for e-commerce product review data that do not use predefined taxonomies or external tools. Furthermore, the strategies derived from our framework align with those employed by omnichannel retailers, demonstrating the framework’s effectiveness as a data-driven decision-support system for optimizing omnichannel management.