Airbnb faces a cold-start problem with new or logged-out users who lack individual history. To address this, they developed Proximity Features, which leverage aggregate user behavior within geographic clusters. The core hypothesis is that users in the same location share similar travel preferences, regardless of their individual data. These features help personalize the experience without relying on persistent user identifiers, aligning with privacy regulations like GDPR.The system derives a coarse geographic location from a user's IP address and groups nearby users into adaptive proximity buckets. Each bucket represents approximately 1,000 users, identified by a proximity key encoding location tiles and, in dense areas, IP hash buckets. This key acts as an aggregation point, allowing machine learning models to use collective signals for personalization. Features are computed daily and include short-term engagement, long-term booking patterns, and bucket metadata.An adaptive clustering algorithm ensures buckets are appropriately sized across varying geographic densities. For dense urban areas, it uses finer subdivisions; for sparse rural locations, it employs coarser groupings. This creates a zoom-adaptive map for effective clustering globally. Proximity keys are stable over time, and features are fetched in real-time from a distributed key-value store, with a soft dependency to avoid blocking requests.Privacy is a foundational element, with features reflecting group behavior rather than individuals. Input data excludes unconsented users, and geo-IP coordinates are anonymized. Proximity Features have demonstrably improved personalization on marketing landing pages and homepage AutoSuggest, showing lifts among never-booked and dormant users. The approach is designed to be reusable, extending personalization capabilities to cold-start users across various Airbnb surfaces and potentially for engagement emails.