Airbnb Experiment Roadmap: A/B Tests to Lift Guest Bookings Across the Funnel

You are a product manager working on the guest side of Airbnb. The goal is to raise the booking rate among guests, and you are asked which A/B tests you would run to get there. Explain how you would find where the opportunities are, which experiments you would propose and in what order, and how you would set up and read the one you would run first. Assume ordinary travel demand: the question was posed before the pandemic, so pandemic travel restrictions are out of scope. ```hint Follow a guest to a confirmed stay List the steps between a guest's first search and a confirmed reservation, including the host's side of the transaction. Where would a guest who wanted to travel give up, and what change at that step could you test? ``` ### Clarifying Questions - Does "booking rate" mean the share of searching guests who end up booking, or how often each guest books over a period such as a year? - Which guests and surfaces are in scope: first-time visitors or returning guests, the mobile app or the website? - Does a reservation request that the host has not yet accepted count as a booking, or only a confirmed one? - Is the target the number of bookings alone, or must the tests also hold up nights booked, booking value and host earnings? ### What a Strong Answer Covers - A precise definition of the booking rate and a guest funnel used to find the largest drop-offs before any test is proposed - Test hypotheses tied to specific stages (search, listing page, price, host response, checkout), prioritized by impact, confidence and cost - A complete design for the first test: unit of randomization, primary metric and booking window, sample size and duration - How the two-sided marketplace can bias results, and guardrails such as cancellations, host acceptance, reviews and booking value - How results are read and acted on, including novelty effects, seasonality and differences between segments ### Follow-up Questions - A new search ranking lifts bookings in treatment, but treatment guests may simply be taking the best listings before control guests can. How would you detect this, and how would you redesign the test? - A change raises the booking rate but also raises cancellations in the week after booking. Do you ship it? - Guests often research a trip over several weeks and on more than one device. How does that change your randomization unit and measurement window?