Optimizing Merchandising with SageMaker

Situation

A leading footwear brand wanted to better understand customer buying patterns across digital and retail store formats. Despite having rich transactional data, insights into product affinities were fragmented, inconsistent across geographies, and slow to generate. Manual analysis consumed days of effort, often missing deeper patterns across categories, seasons, and customer segments -limiting the impact of bundling, promotions, and inventory decisions.

Problem

• Lack of a unified framework to analyze millions of transactions across regions and store types.
• Fragmented data silos made it difficult to detect consistent product affinities.
• Limited visibility into how affinities vary by customer segment, season, or store format.
• No automated way to refresh and surface insights in business dashboards.
• Heavy manual analysis led to delays and overlooked bundling opportunities.

Solution

Mu Sigma built a scalable Market Basket Analysis (MBA) framework powered by AWS SageMaker to uncover product affinities and deliver actionable cross-sell recommendations.

Key Components:

• Data Layer: Snowflake served as the central data source for transactional and product master data, with SageMaker notebooks preparing and cleansing multi-year transaction histories.
• Processing & ML Layer: SageMaker executed Apriori and FP-Growth algorithms to mine frequent itemsets and generate association rules with Support, Confidence, and Lift. Metadata-driven threshold tables enabled dynamic tuning without code changes.
• Segmentation & Contextualization: Rules were enriched with geography, store format, and seasonality-based metadata for targeted recommendations.
• Forecast & Inventory Integration: Product affinity outputs were cross-referenced with forecast and stock availability to ensure campaign feasibility.
• Application & Delivery Layer: Final recommendation datasets were published back into Snowflake and consumed via Tableau dashboards for cross-selling, bundling, and merchandising decisions.
• Continuous Learning: AWS SageMaker Pipelines enabled automated weekly retraining to incorporate evolving buying patterns.

Impact

• 35% increase in cross-sell identification through discovery of high-value affinity pairs.
• ~70% faster time-to-insight, reducing 5–6 days of manual analysis to under 1.5 days.
• ~12% sales uplift in top-performing bundles through improved shelf placement and bundle design.
• Identification of gateway products boosting accessory attach rates by ~8% during pilot campaigns
• Scalable framework capable of processing 50M+ transactions in under 3 hours.
• Real-time integration of affinity recommendations into business workflows.