Geofenced Assortment Rules

The logical boundary that triggers a catalog change. Perimeters are defined using geofencing SDKs (iOS/Android) or IP geolocation for web sessions.

• Geofence Radius: Typically 50m–500m for store parking lots
• Polygon Geofences: Custom shapes for complex locations like malls or airports
• Dwell Time Thresholds: Rules fire only after a user remains inside the fence for a configurable duration (e.g., 30 seconds)
• Exit Triggers: Assortment reverts when the device leaves the zone

The business rules engine that mutates the product catalog when a geofence event fires. Transformations are deterministic and pre-configured by merchandising teams.

• Visibility Toggle: Show or hide specific SKUs, categories, or brands
• Rank Boosting: Artificially elevate local inventory items in search results
• Price Override: Apply store-specific promotions (e.g., 'In-Store Pickup Discount')
• Content Swap: Replace hero banners and recommendations with localized creatives

The low-latency pipeline that ingests geolocation pings and executes assortment rules. Built on streaming architectures to handle millions of concurrent devices.

• Event Ingestion: Kafka or Kinesis streams consume raw location telemetry
• Stateful Sessionization: User location history is windowed to detect fence entry/exit events
• Rule Evaluation: A decision engine matches the user's current fence ID against the rule repository
• Catalog API Call: The resolved assortment delta is pushed to the storefront rendering layer in under 100ms

Geofenced rules are cross-referenced with real-time inventory positions before execution. A rule to boost a product is suppressed if that store has zero stock.

• Stockout Gating: Rules are automatically disabled when available_to_promise = 0
• Perishable Acceleration: Overstocked items with imminent expiry dates receive amplified boosting
• Shelf Capacity Limits: Rules respect physical planogram constraints to prevent promoting items that cannot be restocked

Every assortment change is logged with a deterministic fingerprint to measure incremental lift. This separates the impact of geofencing from other concurrent campaigns.

• Fence ID Tagging: All impressions and clicks are tagged with the active geofence identifier
• Holdout Groups: A percentage of users inside the fence see the default catalog to establish a control baseline
• Lift Metrics: Incremental add-to-cart rate, in-store pickup conversion, and same-day sales attribution

Geofencing requires foreground location permission on mobile or explicit opt-in for web. The system must degrade gracefully when consent is absent.

• Permission States: authorized, denied, restricted, not_determined
• Graceful Degradation: Falls back to IP-based coarse geolocation or default catalog
• Data Minimization: Raw GPS coordinates are ephemeral; only fence entry/exit events are persisted
• Compliance: Aligned with GDPR, CCPA, and platform-specific policies (Apple App Tracking Transparency)

A ranking signal that adjusts product search scores based on real-time inventory position within a geofenced location. When a user enters a store's virtual perimeter, this mechanism ensures items with zero stock are demoted or hidden, while in-stock alternatives are boosted.

• Prevents the frustration of clicking on unavailable items
• Integrates directly with geofenced assortment rules to enforce local visibility constraints
• Typically implemented as a multiplicative weight on the base relevance score

A predictive model that calculates the likelihood of an item becoming unavailable at a specific location within a defined time window. Geofenced rules consume this score to proactively suppress products before a stockout occurs.

• Uses features like current inventory velocity, lead time, and seasonal demand
• Enables preemptive catalog adjustment rather than reactive hiding
• Critical for high-traffic geofences like store parking lots where immediate availability is expected

An unsupervised machine learning method that groups geographic regions by similar purchasing patterns. These clusters form the foundation for defining which geofences receive which assortment rules.

• Uses algorithms like DBSCAN or HDBSCAN on transaction density data
• Identifies micro-merchandising zones without manual zone creation
• Allows a single geofence rule to apply intelligently across an entire cluster of stores with shared demand profiles

A predictive framework that estimates which alternative product a customer will purchase if their first choice is out of stock. When a geofenced rule suppresses an unavailable item, this model informs the optimal substitute to surface.

• Prevents lost sales by guiding intelligent substitution logic
• Trained on historical co-purchase and session-level behavior data
• Ensures the geofenced catalog remains both honest about availability and commercially optimized

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. Geofenced rules leverage these scores to personalize recommendations inside the perimeter.

• Captures regional taste variations that global models miss
• Example: A store in Minnesota may show different 'frequently bought together' items than one in Florida
• Feeds directly into product affinity graphs used by the assortment engine

A dense vector representation of a product that encodes not only its static attributes but also its real-time stock status. Retrieval models use these embeddings to natively filter out unavailable items within a geofence without post-processing rules.

• Concatenates inventory features into the product vector space
• Allows approximate nearest neighbor search to return only available items
• Eliminates the latency overhead of separate inventory checks during retrieval