Decoding customer voices using large language models: assessing customer satisfaction in platform-based food delivery services

Building on recent advances in AI-driven text mining, this study applies a theory-guided text analytics approach based on the Large Language Model LLaMA to examine customer satisfaction in food delivery platform ecosystems. Using Uber EATS customer reviews, the study employs Chain-of-Thought (CoT) prompting to derive constructs grounded in the European Customer Satisfaction Index (ECSI) and interprets them through a Service-Dominant Logic lens. The research proceeds in two stages: measurement validation and structural modelling of relationships among constructs. Study 1 indicates that perceived quality is most strongly associated with overall satisfaction, with service quality contributing more strongly than food quality. This supports SDL’s emphasis on value-in-use emerging from integrated service performance. While overall satisfaction reflects a holistic evaluation of the platform ecosystem, confirmation of expectations is more closely related to transactional fairness. Study 2 indicates that platform image functions primarily as a loyalty-shaping resource. Moreover, overall satisfaction is more strongly associated with customer trust and loyalty than with reductions in complaint behaviour, suggesting that dissatisfaction may not always translate into observable complaint behaviour and may instead reflect unexpressed customer disengagement in low-switching-cost food delivery markets. Overall, the findings support the applicability of the ECSI framework within platform ecosystems and indicate that overall satisfaction is shaped by coordinated value co-creation among platforms, restaurants, and independent delivery providers. This study extends customer satisfaction research to platform-based services and demonstrates how AI-enabled text analytics can support theory-consistent empirical analysis of unstructured customer feedback.