SalesSim: Benchmarking and Aligning Multimodal Language Models as Retail User Simulators

SalesSim is a framework and testbed designed to evaluate Multimodal Large Language Models (MLLMs) as realistic, persona-driven customer simulators in online retail conversations. It models retail interaction as an agentic process, benchmarking state-of-the-art models and identifying behavioral gaps in decision alignment and conversational quality.