Can LLMs simulate realistic e-commerce shoppers?
PAARS: Persona Aligned Agentic Retail Shoppers
April 1, 2025
https://arxiv.org/pdf/2503.24228This paper introduces PAARS, a framework for creating realistic simulated shoppers powered by Large Language Models (LLMs). These "agentic shoppers" use assigned personas derived from real shopping data and interact with a simulated retail environment. The key points relevant to LLM-based multi-agent systems are the persona-driven approach, which improves the fidelity of the simulated shoppers, and the introduction of a group-level alignment suite to measure how well the overall simulated population reflects real human shopper behavior, rather than just individual shopper mimicry. This group-level analysis is particularly relevant for applications like A/B testing and market research.