Leveraging Machine Learning for Customer Segmentation and Personalized Recommendation in Online Retail

Abstract

This study explores how artificial intelligence (AI) can support the full online sales process, not just a single feature. Using a public transactional dataset from a UK retailer, we build a practical framework that moves from simple customer profiling to clear, actionable recommendations. We first create customer features and review their relationships, finding a strong “purchase intensity” pattern: frequent buyers tend to spend more, buy a wider range of products, and show larger month-to-month variability, while long gaps since the last purchase align with lower activity. We then group customers into segments and, within each segment, list the best-selling items. Each shopper receives suggestions for top products they have not yet bought. The pipeline includes basic data cleaning and outlier handling to keep the signals stable. Although the method is simple, it is transparent, easy to deploy, and works well when individual histories are thin. We outline practical checks for success—time-based splits for offline tests and A/B experiments online—and note limits, such as popularity bias and modest personalization. We also describe straightforward upgrades: giving more weight to recent activity, considering revenue or margin, adding variety to the list, and re-ranking with lightweight personal signals. Overall, the work offers a clear, reproducible path that links customer understanding to day-to-day recommendation decisions, supporting both business action and future research.