Business Context

ShopSphere, an omnichannel retail platform with 1.2M active customers, wants to improve marketing efficiency. The growth team needs both customer segments for campaign design and a purchase propensity model to predict which users will buy in the next 30 days.

This problem is designed to test whether the candidate understands the practical difference between unsupervised learning (discovering structure without labels) and supervised learning (predicting a known target using labeled data), and can implement both on the same dataset.

Dataset

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

• Size: 240K customers, 29 input features + 1 binary target
• Target: purchased_next_30d = 1 if the customer makes at least one purchase in the next 30 days
• Class balance: 18% positive, 82% negative
• Missing data: 12% missing in demographics, 6% missing in engagement for newly acquired users

Success Criteria

A strong solution should:

• Produce actionable customer clusters with clear business interpretation
• Train a supervised model with ROC-AUC >= 0.82 and F1 >= 0.58 on the holdout set
• Explain when clustering is useful vs when prediction is required

Constraints

• Marketing needs interpretable outputs for campaign planning
• Batch scoring must complete in under 10 minutes for 1.2M customers
• Retraining can happen weekly; no real-time serving required

Deliverables

1. Build an unsupervised workflow to segment customers and justify the number of clusters.
2. Build a supervised model to predict purchased_next_30d.
3. Compare the goals, inputs, outputs, and evaluation methods of both approaches.
4. Describe how cluster assignments could be used as features in the supervised model.
5. Provide model evaluation, feature importance, and production recommendations.