Went through the Airbnb DS interview loop in February 2026 for a mid-level DS role on the trust and safety side. The loop is three technical rounds plus a behavioral, and each one tests something distinct. Breaking it down.
Round 1: SQL This is a real SQL round, not a warmup. You'll write queries in a shared environment. My questions: Window functions: rank guests by booking value within a region over the last 90 days A multi-table join scenario involving hosts, bookings, and listing attributes One question that required a self-join I did not see coming
Biggest tip: write it out step by step, explain what you're doing. They grade the thinking as much as the final query. Don't just silently type.
Round 2: Product/case analytics You're given a business scenario and need to diagnose and recommend. Mine was: host acceptance rates dropped 8% in one region over 4 weeks. What do you investigate, what metrics do you pull, what do you recommend?
This is where knowing Airbnb's product deeply pays off. The better your intuition about supply-demand dynamics, host psychology, and marketplace mechanics, the more credible your diagnosis sounds. Generic "check for seasonality, check for bugs" is fine but doesn't impress.
Round 3: Stats / experimental design This is the round people underprepare for. Topics that came up in mine: Power calculation for an A/B test on a low-frequency event (they pushed on how you'd handle low sample size) Difference-in-differences setup for a policy change that couldn't be A/B tested Bayesian vs. frequentist framing question, conversational, not a proof
I'd say the stats round is harder than the SQL round if you haven't done much causal inference or experimental design in your day job.
Behavioral: Standard Airbnb values-based questions, same as described elsewhere on this forum. Have impact stories ready that involve influencing stakeholders on data findings, not just individual analysis work.
Offer: I got an L4 equivalent offer (Airbnb doesn't always use L numbers for DS). Total comp was around $210k for a Seattle-remote arrangement. I was expecting SF-anchored, so the number came in slightly under what I targeted but still solid.