Humanizing Answers for Compatibility Questions in E-commerce using Large Language Models

Customer experience is a critical aspect of online purchase decisions. The service, the attendant's response, and how the customer is treated contribute to customer satisfaction. This article investigates using large language models for humanizing customer support in e-commerce. In particular, we address compatibility questions. Leveraging the infrastructure and dataset from an AI Brazilian startup, we compare the effectiveness of three different models to generate natural language answers in Portuguese. We generate human-like answers and evaluate them based on compatibility correctness, number of tokens, legibility, human likeness, and effect on the purchase. Our results highlight the effectiveness and drawbacks of the explored models in different temperature settings. This study improves customer experiences and provides guidance for e-commerce platforms in implementing humanized responses.