Personalized Retail Analytics: Using Bedrock LLMs for Demand Prediction and Product Recommendations

Abstract

The retail landscape today is pushing for the development of more advanced personalization tools in order to keep up with the changing consumer expectations and the growing competition. This study explores the use of Large Language Models (LLMs) by Amazon Bedrock to simultaneously predict demand and generate product recommendations in the retail sector. Commonly used methods consider these functions to be distinct that rely on different algorithms thus not utilizing the common customer insight patterns. Our architecture brings together Bedrock's Claude and Titan models to develop a single personalization engine that understands customer transactions, website browsing, and other factors through natural language. By testing with three different product categories—fashion, electronics, and grocery—we prove a 31% boost in demand forecasting and a 27% increase in the acceptance of recommendations over the traditional collaborative filtering methods. The system also provides narratives that explain the predictions, hence, the retailer can comprehend the recommendation logic and earn the customer’s trust. Moreover, it was found that LLM-based approaches particularly excel in cases of new products and sparse data where the traditional approaches fail. This research provides a scalable infrastructure for retail personalization that overcomes the cold-start challenge without compromising on the computational efficiency required for real-time customer interactions over both online and offline retail channels.