Dynamic (AI-Powered) Customer Segmentation in eCommerce   

Project Overview

This project leveraged unsupervised machine learning to segment customers dynamically based on purchase behavior, engagement metrics, and wallet usage patterns. Implemented on Shopee’s big data infrastructure, the solution created actionable buyer personas, directly feeding personalized marketing campaigns via CRM integrations. This enabled the marketing team to pivot from one-size-fits-all campaigns to targeted, revenue-optimized messaging.  

Context / Problem Statement:

At Shopee, customer acquisition was strong, but retention and repeat purchase rates varied significantly by category and region. Traditional RFM (Recency-Frequency-Monetary) models failed to capture behavioral shifts during flash sales, regional campaigns, or digital wallet promotions—leading to generic outreach and missed revenue opportunities.

AI-Augmented Solution:

To enable hyper-targeted engagement, we implemented an unsupervised machine learning pipeline for dynamic customer segmentation:

• Built a feature-rich dataset combining behavioral, transactional, and engagement metrics (product views, coupon usage, wallet usage frequency).

• Applied K-means and DBSCAN clustering techniques to create adaptive customer clusters—“Bargain Hunters”, “Loyal Buyers”, “One-Time High Spenders”, etc.

• Integrated clusters into the CRM system, triggering personalized campaigns via app notifications and emails.

• Introduced predictive uplift modeling to prioritize segments based on likelihood to convert during major campaigns.

AI Stack / Tools Used:

• Python (scikit-learn, pandas, seaborn)

• BigQuery for data prep

• Looker Studio for visualization

• Airflow to automate daily feature pipelines

• Campaign APIs integrated with CRM for real-time sync

Outcome & Business Impact:

• 📈 Campaign conversion rate increased by 21%

• 🛍️ Repeat purchases improved by 17% in “Dormant Spenders” segment

• 💬 3x uplift in campaign engagement for “Flash-Sale-Responsive” cohort

• 📊 Provided marketing leadership with actionable persona heatmaps by region and channel

My Role:

I led the end-to-end segmentation initiative—from data engineering and feature crafting to model selection, stakeholder training, and campaign design strategy. Worked closely with CRM and product marketing teams to operationalize insights into live campaigns.