From reviews to constructs: Using LLMs to model customer satisfaction in platform-based services

This study investigates the business model of food delivery services (FDS) within the platform economy, characterized by the distributed nature of value creation and service delivery among partner restaurants, freelance couriers, and platform providers. To understand how customers perceive these interdependent service components, we apply Large Language Models (LLMs) to analyze unstructured customer reviews by generating synthetic data. Specifically, we leverage LLaMA with Chain-of-Thought (CoT) prompting to uncover latent constructs aligned with the American Customer Satisfaction Index (ACSI) framework—namely customer expectations, perceived quality, perceived value, customer satisfaction, customer complaints, and customer loyalty. Results reveal that perceived quality is the dominant driver of customer satisfaction, while perceived value plays a secondary role linked to platform-managed efficiencies. Unlike overall satisfaction, fulfillment of expectations is more narrowly focused on the cost-related aspects of the service evaluation. Customers evaluate quality holistically, integrating all service components, but attribute overall responsibility to the platform rather than to individual actors. Furthermore, satisfaction more strongly predicts complaint behavior than long-term loyalty, highlighting the transactional character of FDS platforms in low-switching-cost environments. Methodologically, the study introduces a novel AI-driven approach that transforms natural language content into structured input for Structural Equation Modeling (SEM), enabling both measuring latent constructs as well as deriving quantitative causal analysis based on customer-authored feedback. This scalable, theory-driven method for analyzing customer responses extends the applicability of consumer behavior models and offers actionable insights for marketing researchers and managers alike.