interviewing at Snap for a data scientist role on the ads measurement team in three weeks. i've done some research but the public info is pretty thin for DS specifically (most threads are SWE).
if anyone has gone through a DS or applied ML loop at Snap recently, i'd really value knowing: how heavy is the SQL component vs stats vs ML theory did they give a take-home or is it all live rounds how much do they test on experimentation / causal inference specifically
i know ads measurement at Snap would lean heavily on experiment design and attribution, so i'm prepping for that, but any data points would help calibrate. doesn't have to be recent, even 2024 info would help.
4 replies
ml_mike
i did a loop for an ML platform role (not ads measurement) about eight months ago. heavy on system design for ML pipelines, light on coding, almost no SQL. but ads measurement is a different beast. that team lives in experiment design so i'd expect a lot of causal inference questions: A/B test validity, novelty effects, holdout group sizing. brush up on CUPED if you haven't.
de_derek
my partner did a take-home for a data role there about 18 months ago. was a Jupyter notebook exercise, realistic dataset, about 3 hours of actual work plus writeup. they said recent loops had moved toward fewer take-homes because of candidate feedback. so it might be all live now, no way to know until you ask the recruiter.
ds_dmitri
good to know. i'll ask. i actually prefer live rounds, take-homes have this miserable ambiguity about how much polish they actually want
analyst_ana
not DS but i went through a BA loop at Snap for a strategy analytics role. they asked a lot about defining metrics for vague product goals. like 'how would you measure Snap Map success' and then 'ok now how do you know if the metric you chose is actually good.' felt like they wanted to see how you handle being pushed on first instincts.