The business problem
The merchandising team at a major sportswear retailer wants to move from one-size-fits-all marketing to segment-driven campaigns. They have purchase history, demographics, and engagement data on millions of customers. Build segments that are actionable (the marketing team can describe each one in a sentence), stable (a customer doesn't bounce between segments week to week), and distinct (different segments behave differently in response to campaigns).
This case applies Decision 4 (Clustering) with operational deployment.
The hardest part isn't the algorithm
It's choosing which features to cluster on, deciding how many segments are useful for the business (not just statistically optimal), and naming them so the marketing team will actually use them.
Case kit
Everything you need
- Nike Customer DatasetCustomer-level features for segmentation.
- Segmentation NotebookReference implementation: feature selection, scaling, K-means + hierarchical, profile.
- SegmentPulse Excel ScorerScore new customers into the segments without re-running the clustering.
Topics you'll be applying
- Decision 1 — Data Prep & EDA — heavy feature engineering on transaction history
- Decision 4 — Clustering — your core technique