Predicting Behaviors with Large Language Model (Llm)-Powered Digital Twins of Customers

Place : Room 3.216 - Grand Paris Campus & online
Time : 12.00-01.30 PM

Speaker : Xin (Shane) Wang, Professor of Marketing – Virginia Tech.

Digital twins of customers (DToC) have emerged as a promising approach to simulate consumer thinking, feeling, and decision-making in marketing contexts. This research proposes and empirically tests a methodological framework that combines fine-tuning and retrieval-augmented generation (RAG) to construct LLM-based customer digital twins. Fine-tuning on user-generated content allows the model to internalize individual traits, preferences, and behaviors, while RAG equips the twin with real-time access to contextual product information. We demonstrate the framework using Amazon e-commerce data, constructing 306 personified digital twins and evaluating their performance in predicting both purchase decisions and review contents. The resulting digital twins achieve high accuracy in predicting future purchases (83%) and generate product reviews with strong semantic alignment to actual customer content (cosine similarity above 0.94). This method opens new possibilities for personalized marketing, pre-deployment campaign testing, and privacy-compliant consumer modeling. The findings contribute to emerging literature on generative AI and synthetic agents in marketing, advancing the conceptual and technical foundation for predictive, interactive, and individualized customer simulation.
 

For further information, please contact Professor Margherita Pagani: margherita.pagani@skema.edu