Assortment Breadth Optimization

The fundamental tension between carrying a wide variety of categories (breadth) and a deep selection of variants within a single category (depth). A grocery chain might optimize for breadth in a small urban store by stocking 50 distinct cheese types but only 1-2 brands per type, while a suburban hypermarket optimizes for depth with 15 cheddar brands. The optimal frontier is calculated by modeling the marginal revenue per SKU against the holding cost per unit. Over-breadthing leads to fragmented inventory; over-deepening risks cannibalization and stockouts in adjacent categories.

A predictive framework that estimates which alternative product a customer will purchase if their first choice is unavailable. This prevents revenue leakage when breadth is constrained. Key components include:

• Substitution probability matrices built from co-purchase graphs
• Attribute-based similarity scoring (brand, size, flavor, price tier)
• No-purchase probability estimation when no acceptable substitute exists A well-tuned transference model allows a retailer to confidently remove a low-velocity SKU, knowing 85% of its demand will transfer to a higher-margin alternative rather than being lost entirely.

A collaborative filtering technique that calculates product similarity based on the purchasing behavior of users within the same geographic cluster rather than a global user base. A sunscreen and surfboard wax might show low affinity globally but extremely high affinity in coastal micro-markets. This geospatial constraint ensures that breadth decisions reflect heterogeneous local tastes. The model segments users into micro-merchandising zones—often at the store or neighborhood level—and computes item-to-item similarity matrices independently for each zone, preventing urban preferences from distorting rural assortments.

An analytical method that identifies when the introduction or promotion of one product reduces the sales of another similar item in the same catalog. Without this guardrail, adding breadth can become a zero-sum game. Detection relies on:

• Incremental lift measurement via controlled A/B tests
• Causal inference models isolating the new SKU's impact from seasonality
• Cross-elasticity matrices quantifying substitution effects For example, introducing a store-brand organic pasta may cannibalize 40% of sales from the premium imported brand while only growing the category by 5%, signaling a net-negative breadth expansion.

A dense vector representation of a product that encodes not only its static attributes (category, brand, price) but also its real-time stock status. This allows retrieval and ranking models to natively filter out unavailable items without post-hoc rules. The embedding concatenates:

• Static features: product description, image, categorical taxonomy
• Dynamic features: current stock level, days-of-supply remaining, restock ETA
• Contextual features: local demand velocity, sell-through rate When a product's stock drops below a critical threshold, its vector shifts in latent space, naturally reducing its cosine similarity to user query embeddings and suppressing its visibility.

The computational process of identifying missing product categories or attributes in a local catalog by comparing current offerings against predicted unmet consumer demand. The analysis pipeline:

• Ingests search query logs with zero results or low click-through rates
• Clusters failed queries into latent product attributes using NLP
• Cross-references against competitor catalogs and market share data
• Scores gaps by estimated revenue opportunity minus incremental carrying cost A gap analysis might reveal that a store in a predominantly vegan neighborhood has zero plant-based protein options, quantifying a $12K monthly revenue leakage that justifies adding 8 new SKUs.