Large Behavior Model: A Promptable Digital Twin of the Retail Customer

Short summary

Researchers introduce the Large Behavioral Model (LBM), a language-model-based digital twin trained on large-scale retail transactions using continued pre-training, supervised fine-tuning, and reinforcement learning. LBM outperforms frontier general-purpose LLMs on in-domain retail tasks including purchase prediction, basket completion, and promotion response. Ablation studies show continued pre-training drives behavioral generalization, while retrieval-augmented generation and RL with verifiable rewards further improve evidence-based calibration.

• •LBM learns customer decision-making from retail transaction data via a unified Person-Environment formulation
• •Outperforms frontier LLMs on purchase prediction, basket completion, and promotion response
• •Continued pre-training is the primary driver of behavioral generalization

Generated with AI, which can make mistakes.