At Airbnb, understanding host availability is essential for product development, campaigns, and operations. Airbnb uses segmentation to segment availability for millions of listings daily. By enriching availability data with novel features and applying machine learning techniques, they developed a practical approach.
Traditional segmentation methodologies, such as “RFM” (Recency, Frequency, Monetary), are not focused on calendar dynamics and cannot handle calendar data and daily inference for millions of listings. Airbnb's segmentation approach is designed to capture nuances of how a listing is available throughout the month.
Airbnb uses several features to differentiate between hosts' availability patterns, including nights available, streakiness, quarters with at least one night of availability feature, and maximum consecutive months.
Airbnb identified eight clusters, including Always On, Short Seasonal, Long Seasonal, Weekend Warriors, Event Motivated, Occasional Summer Hosts, Gap Fillers, and Urban Extended Stays.
Airbnb uses a decision tree algorithm to classify listings into clusters with if-else rules, which is then translated into a SQL query by converting the decision tree's conditions.
Various teams at Airbnb leverage these segments, including product, marketing, and UX research teams.
Airbnb's segmentation methodology has also been adapted to host activity data, differentiating occasional Settings Tinkerers from frequent Settings Optimizers.
Airbnb's segmentation approach can be adapted to other industries where understanding temporal engagement is essential, such as distinguishing between occasional and frequent users of a service.
This clustering approach gives Airbnb insights into host behavior patterns and preferences, helping the company improve its platform and provide better services for its customers.
By understanding host behavior, Airbnb has introduced features that support hosts with different styles, such as seasonal pricing tips, onboarding guides, and tools for managing multiple reservations.