Citi · Primly Community

Citi data scientist interview (SQL + case + stats): what the loop actually looks like in 2026

ds_dmitri · 6 replies

went through the full citi DS loop for their risk analytics team in NYC, early 2026. sharing because i couldn't find good signal online before my interviews.

the process for me was: recruiter screen, HireVue, then a 4-round virtual onsite.

HireVue was two video-response questions. pure behavioral. nothing technical. you get like 2 minutes per answer. i treated it as the bar for clearing, not the bar for the job.

round 1: SQL + data manipulation. not leetcode SQL. this was more like analyst-level query writing: window functions, cohort queries, identifying data quality issues in a given schema. the questions assumed you know how a real relational database works, not that you've memorized SQL syntax trivia. they gave me a scenario about customer transaction data and asked me to find anomalies. one subquery was genuinely tricky. study GROUP BY edge cases, rolling averages, and lag/lead window functions. that covers 90% of what you'll see.

round 2: stats and probability. regression assumptions, A/B testing design, type 1 vs type 2 error tradeoffs. nothing from first principles probability theory. more applied: "we ran an experiment, here's the result, what concerns do you have?" i got a question about sample size estimation that i fumbled a bit. know the formula and be able to explain it to a non-statistician.

round 3: case study. they sent a dataset ahead of time (24 hours before), asked me to prep a 15-minute presentation on my findings. i chose to focus on one clean insight rather than cover everything. i think that was right. they asked follow-up questions about how i'd operationalize the model in a production environment. that felt more like a practical ML question than a DS one. know something about model monitoring and drift.

round 4: behavioral. leadership principles style. STAR format. questions were around stakeholder management, ambiguity, and influencing without authority. classic stuff but they dug in on specifics.

leveling: i was interviewing for a VP-equivalent DS role (their internal titles are a bit unusual). comp felt competitive for the street, roughly where you'd expect for NYC fintech. the hiring process moved slowly, about 6 weeks from application to offer.

feel free to ask anything.

6 replies

analyst_ana

this is really helpful, thank you. did the SQL round have a time limit? like were you writing queries live with someone watching, or was it async?

ds_dmitri

live, with an interviewer watching on zoom. not in a shared IDE, just screen share. no strict time limit per question but i could feel the pacing. i had about 45 minutes for two problems. don't overthink out loud too much, just talk through what you're doing as you write.

sec_sasha

the case study 24h takehome is interesting. did they specify a tool or was it open format? ppt, jupyter, whatever?

ds_dmitri

open format. i did a jupyter notebook exported to slides basically. saw other candidates mention they used ppt. they didn't seem to care. content over format.

newgrad_neil

do you know if this loop is the same for a senior analyst vs VP level? i'm applying for a mid-level DS role (3 YOE) and trying to calibrate expectations

ds_dmitri

i think the structure is similar but the depth of the case study and the behavioral bar shifts. at lower levels they're probably less focused on the 'influencing without authority' angle and more on raw technical skills. but i genuinely don't know for sure.