Business Context

ShopSphere, an e-commerce marketplace with 2.5M active customers, wants to group customers into actionable segments for lifecycle marketing, merchandising, and CRM targeting. There is no labeled target, so the task is to build an unsupervised clustering solution and explain when clustering is appropriate versus supervised learning.

Dataset

You are given a customer-level feature table built from the last 12 months of activity.

• Size: 420K customers, 30 features
• Target: None — discover natural customer segments
• Data types: Numerical and categorical features with different scales
• Missing data: ~8% missing in engagement features, ~3% missing in acquisition metadata

Success Criteria

A good solution should produce stable, interpretable segments that marketing can use in campaigns. Aim for clusters with silhouette score >= 0.20, low drift month over month, and clear business meaning (for example, high-value loyalists, discount-driven browsers, seasonal shoppers).

Constraints

• Segments must be explainable to non-technical stakeholders
• Batch scoring should complete in under 30 minutes weekly
• The approach should support periodic re-training as customer behavior changes
• Avoid methods that are too sensitive to feature scaling or outliers without mitigation

Deliverables

1. Build a clustering pipeline for customer segmentation
2. Justify when clustering is the right choice for this problem
3. Compare at least two clustering approaches and select one
4. Evaluate cluster quality using quantitative and business-facing metrics
5. Provide a plan for assigning new customers to segments in production