Snowflake · Primly Community

Snowflake data scientist interview (SQL + case + stats): what actually showed up

analyst_ana · 5 replies

just finished the full Snowflake DS loop for a mid-level role on their growth analytics team. posting the breakdown because i couldn't find a recent one when i was prepping.

the process was 5 rounds over about 3 weeks: recruiter screen (30 min, mostly background + why Snowflake) take-home SQL challenge (2 hours given, took me ~1.5) technical screen with a DS: SQL live coding + one stats question virtual onsite: 4 back-to-back sessions hiring manager debrief at the end of onsite day

SQL: this was heavier than i expected. the take-home had 4 problems, two of which were genuinely hard window function stuff. think running totals partitioned by account, plus a self-join to find customers who upgraded tiers. the live coding SQL was easier but they wanted clean, readable code and asked me to talk through the query plan.

Stats/probability: one classic probability question (conditional probability with a Bayes framing), one question about A/B test design for a feature rollout. they specifically asked how i'd handle low-traffic segments and what the minimum detectable effect would be. felt like they wanted someone who actually runs experiments, not just runs the numbers afterward.

Case: the case was product-analytics flavored. something like: "our weekly active accounts metric dropped 8% last week, walk us through your investigation." they wanted a structured approach: segment by product line, by region, check for data pipeline issues first, then user behavior changes. no slides, just whiteboard.

Behavioral: two rounds of this. they lean hard on the Snowflake values. cross-functional collaboration came up a lot. they want examples of influencing without authority.

overall the bar felt genuinely high for SQL and stats. i'd say LC medium is a floor but the SQL is more about data modeling intuition than raw algorithmic cleverness. if you're coming from a pure ML background and your SQL is rusty, prep that first.

offer: L4 equivalent, San Mateo hybrid. base in the 175-185k range from what i can tell (they haven't officially leveled me against their bands yet but that's the range the recruiter floated).

5 replies

analyst_ana

the case format is really helpful to know. did they give you any data to look at during the investigation case or was it purely verbal/whiteboard?

qa_quinn

purely verbal. they described the scenario and i had to drive the structure myself. no dataset. i think they're testing the diagnostic thinking more than actual number-crunching.

de_derek

the take-home window function stuff tracks with what i heard from a friend who went through DE loop there. they're a data warehouse company so they expect you to actually know SQL deeply, not just SELECT * level.

finance_faye

any pushback on comp or was the band pretty fixed? curious if there's room to negotiate the L4 base.

consultant_cam

haven't gotten to that stage yet, will update when i do. recruiter seemed like there's a band range, not a single number, so probably some room.