Coinbase · Primly Community

Coinbase data scientist interview: SQL, case study, and stats questions i actually got

ds_dmitri · 4 replies

did the coinbase DS loop targeting a mid-senior role (their L4/L5 range). wanted to write up the technical breakdown because it's different from pure-SWE loops and there's almost nothing on it.

format: recruiter screen, a technical phone screen (SQL focus), then onsite which is 4 rounds: advanced SQL, stats/probability, product case, and behavioral.

technical phone screen: 45 minutes, one SQL problem. not a simple aggregation. mine involved a window function over a time series. they gave me a schema for a hypothetical transactions table and asked me to identify suspicious activity patterns. you need to be comfortable with LAG, LEAD, PARTITION BY, and writing subqueries. if you've only done basic GROUP BY SQL in practice, you'll struggle.

SQL onsite round: deeper version. two problems. one was finding users who hit a specific threshold within a rolling 7-day window. the second was a classic retention cohort analysis -- given a login events table, find 30-day and 90-day retention by signup cohort. both are solvable with window functions. they asked me to also explain how i'd validate my results, which tripped up a lot of people i talked to who prepped only for writing the query and not for QA-ing it.

stats and probability: this was the most interesting round. questions ranged from basic (explain p-values to a PM) to applied (how would you design an A/B test for a feature that affects transaction frequency, where the baseline rate is rare events). the rare events A/B testing question is classic fintech stats -- Poisson distribution or bootstrap-based approaches, not just your standard t-test. know your stuff here.

product case round: they gave me a metric that had dropped and asked me to diagnose it. coinbase-specific metrics -- something like daily active traders dropping 15% week over week. standard diagnostic framework works: segment by surface, check for data pipeline issues first, then external factors, then product changes. they cared a lot about how i prioritized my hypotheses.

what i'd prep: window functions are non-negotiable. practice retention cohorts and rolling windows until they're automatic. brush up on experimental design for rare events and for situations where you can't randomize cleanly. have a clean mental model for product metric diagnostics that you can walk through out loud. crypto context helps for the case round but they don't require deep domain knowledge. understanding fintech transaction dynamics (seasonality, external market effects) is more useful than knowing what ethereum L2s are.

4 replies

analyst_ana

the retention cohort question is showing up everywhere in DS interviews right now. good sanity check: can you write it from scratch in under 10 minutes? if not, that's where to spend your prep time.

de_derek

the 'explain p-values to a PM' question -- did they actually care about technical precision or were they just checking that you can translate stats to business language? i've been asked this and honestly have no idea which way interviewers grade it.

ds_dmitri

both. you need to be technically correct but not robotically so. something like 'a p-value tells you how surprising your result would be if there were actually no effect -- a small p-value means the result would be very surprising by chance, but it doesn't tell you the effect is large or meaningful' hits both. what tanks you is either a textbook wrong definition or one that's so simplified it falls apart under follow-up.

ml_mike

the rare events A/B test design question is underrated as a prep area. most DS interview guides focus on standard t-test setups. but fintech is full of rare event problems -- fraud detection, conversion on specific flows, etc. if you're going into coinbase DS, know your Poisson and negative binomial distributions and know when to reach for permutation tests instead.