Segment and Predict Retail Customers

Business Context

ShopSphere, an online retail marketplace with 1.2M customers, wants to improve lifecycle marketing. The growth team needs both customer segments for campaign design and a purchase propensity model to predict whether a customer will buy in the next 30 days.

This question is designed to test whether you understand the practical difference between unsupervised learning (finding structure without labels) and supervised learning (predicting a known target), and how to use both on the same dataset.

Dataset

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

• Size: 240K customers, 25 input features
• Target: Binary label indicating whether the customer made a purchase in the next 30 days
• Class balance: 18% positive, 82% negative
• Missing data: 12% missing in csat_score, 6% missing in avg_session_duration, 3% missing in demographic fields

Success Criteria

A strong solution should:

• Produce meaningful customer segments for marketing use
• Train a supervised model with ROC-AUC >= 0.82 and PR-AUC >= 0.45 on the held-out test set
• Clearly explain when to use clustering vs classification and the tradeoffs of each

Constraints

• Marketing needs segments that are explainable to non-technical stakeholders
• Batch scoring must finish in under 10 minutes for 240K customers
• Retraining should be feasible monthly with standard Python tooling

Deliverables

1. Build an unsupervised learning pipeline to segment customers
2. Build a supervised learning pipeline to predict purchased_next_30d
3. Compare the two approaches and explain the difference in objective, inputs, and evaluation
4. Recommend how both outputs should be used together in production
5. Provide evaluation metrics, feature importance, and segment summaries