User Clustering for New Sales Push Notifications

At Choose, the New Sales Push Notification project aimed to increase engagement by moving away from sending the same "most-clicked" upcoming sale to all users. Instead, I sought to segment users based on their category click preferences on the home page. The initial objective was to build data-driven segments so each group could receive a push notification promoting the most-clicked sale in their preferred category, rather than a single generic sale sent to everyone.

I initially attempted clustering using Machine Learning techniques such as K-Means and DBSCAN. However, the dataset contained 30 category features, leading to a high-dimensional space where clusters were neither compact nor well-separated. As a result, the algorithmic clustering did not produce actionable segments. The challenge was to create meaningful and stable clusters that could be directly applied for targeting, while avoiding excessive complexity.

I adopted a manual segmentation strategy guided by business-driven thresholds on user engagement metrics. We defined 22 distinct user segments based on dominant category preferences. For example, a "Kids" segment included users who frequently clicked on Kids-related sales. Each segment then received a push notification promoting the most-clicked sale within its preferred category from the Upcoming Sales section. This ensured relevance while maintaining operational simplicity.