Airbnb Boosts Personalization with Privacy-Compliant Proximity Features.

Personalization is crucial for two-sided marketplaces, but platforms like Airbnb face a significant challenge with "cold-start" users—those who lack sufficient historical data for effective recommendations. This issue is particularly acute for new or logged-out users, especially on paid marketing landing pages, where traditional user-level features are unavailable. Furthermore, evolving privacy regulations and restrictions on third-party cookies increasingly limit identifier-based tracking. Airbnb has developed "Proximity Features" to overcome these hurdles. This innovative system creates privacy-compliant aggregated user-level signals by grouping users based on geographic proximity, utilizing geo-IP data and an adaptive clustering algorithm. Each cluster comprises approximately 1,000 nearby users, and the system operates without requiring persistent individual identifiers at inference time, ensuring privacy by design. The entire pipeline processes only consented, aggregated data within strict privacy controls. Currently deployed across multiple Airbnb surfaces, including marketing landing pages and destination recommendations, with email integration underway, Proximity Features have demonstrated significant success. Online A/B experiments showed statistically significant increases in bookings, with the most substantial gains observed among users who previously had no or stale historical data. This solution effectively addresses a long-standing personalization challenge while upholding stringent privacy standards.