Large Behavior Model: A Promptable Digital Twin of the Retail Customer
06:00 · July 9, 2026 · arXiv cs.AI RSS
Customer behavior modeling underpins recommendation, marketing, and decision support, yet existing approaches either optimize predictive accuracy without explaining decisions or simulate users without grounding them in real behavioral data. We present the Large Behavioral Model (LBM) that learns customer decision making directly from large-scale retail transactions through a unified Person-Environment formulation. Customer state is represented by a behavioral profile derived from historical purchases, while product context is incorporated through retrieval-augmented generation. The model is trained using continued pre-training on verbalized behavioral data, supervised fine-tuning for decision generation, and reinforcement learning with verifiable rewards for evidence-based calibration. We evaluate the proposed framework on purchase prediction, hard-negative discrimination, basket completion, promotion response, and cross-domain voucher redemption. The model consistently outperforms frontier general-purpose language models on in-domain retail tasks while demonstrating strong zero-shot and fine-tuned transfer across retailers and decision domains. Ablation studies show that continued pre-training is the primary driver of behavioral generalization, retrieval is most effective when applied during both training and inference, and reinforcement learning improves reliance on explicit behavioral evidence over generic language-model priors. These results demonstrate that behavioral knowledge encoded in transaction histories can be effectively learned by language models, providing a scalable foundation for customer digital twins and behavior simulation.
Summary
The Large Behavioral Model (LBM) addresses shortcomings in existing customer modeling techniques, which typically prioritize predictive accuracy at the expense of interpretability or generate synthetic user simulations detached from actual transaction histories. The approach frames customer decision making through a unified Person-Environment formulation in which a behavioral profile, derived directly from historical purchases, represents the customer state while product context is supplied via retrieval-augmented generation.
Training proceeds in three stages. Continued pre-training on verbalized behavioral sequences allows the model to internalize patterns latent in large-scale retail data. Supervised fine-tuning then aligns outputs with decision-generation objectives, after which reinforcement learning with verifiable rewards calibrates the model to favor explicit evidence drawn from the retrieved context rather than generic language-model priors.
Evaluations span purchase prediction, hard-negative discrimination, basket completion, promotion response, and cross-domain voucher redemption. Across these tasks the LBM outperforms frontier general-purpose language models on in-domain retail benchmarks and exhibits robust zero-shot and fine-tuned transfer to new retailers and decision domains. Ablation results indicate that continued pre-training contributes most to behavioral generalization, retrieval yields the largest gains when applied at both training and inference time, and reinforcement learning strengthens reliance on observable transaction evidence.
Why it matters
This research is highly relevant for Dutch AI researchers and practitioners in the robust local retail and e-commerce sectors (e.g., Bol.com, Ahold Delhaize). The methodology offers an actionable, transparent approach to customer modeling that aligns with the EU's demand for explainable and evidence-based AI systems.
