just finished the spotify DS loop for a mid-level (DS II) role on their listening behavior analytics team. the 'spotify data scientist interview SQL case stats' search returns a lot of vague content so here's the actual breakdown from 2026.
the structure was four rounds after the phone screen: SQL technical round (45 min) stats + probability round (45 min) product/case round (45 min) behavioral (30 min)
SQL round the questions were legitimately hard. not 'join two tables' hard. more like: window functions over time-series data, calculating retention cohorts, and one question about identifying anomalies in streaming event data. know your window functions cold (LAG, LEAD, ROWNUMBER, DENSERANK). they use a collaborative editor and expected me to run the query mentally and explain the output before running it.
one specific question: given a table of user listening sessions with start/end timestamps, write a query to find users who had overlapping sessions (possible bug in the data). that's a self-join with interval logic, medium-hard.
stats round probability basics, A/B testing setup, and one question about experiment design: 'we want to test whether a new UI change increases daily listening. how do you design the experiment and what are the risks?' power calculation came up. they asked about novelty effect explicitly. know Bayesian vs frequentist at a conceptual level but don't need to go deep.
case round this felt the most like a PM interview. a listening metric had dropped, walk them through the diagnosis. i used a structured funnel approach (acquisition, activation, engagement, retention) and they kept adding constraints. 'assume the data shows it's only US users.' then 'only on mobile.' it was more about diagnostic reasoning than quantitative execution.
overall bar higher than i expected for DS II. felt more like a senior DS bar at some other companies. SQL especially.