just finished a microsoft data scientist loop, targeting a role in the experimentation/analytics org. 5 rounds over two days (virtual). sharing the full breakdown because the ds interview content for microsoft specifically is thin compared to google/meta/amazon.
round 1: SQL not optional, not easy. they gave me two problems. the first was a join + aggregation question, medium difficulty. the second involved window functions (lag/lead + rank) and required me to compute rolling retention rates. i had about 30 minutes per problem. coderpad environment. no autocomplete.
if you're going into a microsoft DS role assuming sql will be light: don't. the experimentation org cares about whether you can actually query data.
round 2: stats + experimentation this is the one that separates people. questions i got: 'an A/B test shows a significant result but the sample sizes were different across treatment and control. what do you do?' 'how would you handle a metric where you expect a strong network effect to contaminate your control group?' 'what's the difference between type I and type II error in the context of shipping a product feature?'
the network effect / interference question is classic experimentation at scale. they want SUTVA, not just a textbook answer.
round 3: case / product analytics given a scenario: 'bing search sessions have dropped 8% month over month in a specific geography. walk me through how you'd diagnose this.' very standard product analytics case. expected structure: external factors, product changes, data quality issues, segment breakdown. i did all of that plus proposed a specific analysis plan with the metrics i'd pull.
round 4: ML concepts not super deep for this role (experimentation focused). asked about bias-variance tradeoff, how i'd evaluate a recommendation model offline vs. online, and when i'd use a simpler model over a complex one. no coding here.
round 5: behavioral standard growth mindset framing. one question was specifically about how i communicated a data-driven recommendation to a non-technical stakeholder who pushed back.
the whole process was respectful of my time. they were direct about the timeline. offer came 6 days after my final round.