Reinforcement Learning for Dynamic Pricing: Consumer Welfare Implications and the Limits of Algorithmic Optimisation

Reinforcement learning-based dynamic pricing systems have achieved significant revenue gains for early adopters, yet their welfare implications remain poorly understood. Using agent-based simulations calibrated against publicly available pricing data from three major e-commerce platforms, we demonstrate that RL pricing agents consistently extract surplus from price-insensitive consumer segments while creating volatility that disadvantages lower-income shoppers. Our analysis calls for sector-specific algorithmic auditing requirements.