JPMorgan Chase · Primly Community

JPMorgan Chase data scientist interview (SQL + case + stats): what to actually prepare

ds_dmitri · 5 replies

Did a full DS loop at JPMC (Consumer & Community Banking team, NYC, mid-senior level) a few months ago. Going to break this down by component because the 'data scientist interview' label covers a lot of ground and JPMC's version is different from your typical tech company DS process.

SQL round (45 min) This was harder than I expected. They don't ask 'find the second highest salary' type questions. The questions were genuinely analytical: multi-table joins across transactional data, window functions (specifically LAG/LEAD for trend analysis), and one question that required building a cohort analysis query. I'd say the SQL difficulty was intermediate to advanced. If you only know basic SELECT/WHERE/GROUP BY you will struggle.

Case / product sense round (45 min) Framed as: 'we're seeing a 15% drop in mobile check deposits this quarter in a specific region. Walk me through how you'd investigate it.' This is classic case-style problem solving but with a data lens. They wanted: hypothesis generation, what data I'd pull first, what analyses I'd run, and how I'd differentiate between product issues, external factors, and seasonality. I've done a lot of these; this was well-run.

Stats / ML round (60 min) Honestly the most varied round. Topics covered: A/B test design (minimum detectable effect, sample size, when to stop early), basic probability (conditional probability problems involving fraud rates), and a conceptual ML question about model selection for a credit risk scoring problem. They didn't ask me to code a model, just reason about tradeoffs between logistic regression vs. gradient boosting in a regulated environment. Key word: regulated. They specifically asked about model explainability because of fair lending laws. Be ready for that.

Behavioral (part of a combined round) Similar to what others have described: tell me about a time you influenced a decision using data, how do you work with non-technical stakeholders, and one question about handling pressure.

Comp for DS at JPMC in NYC is competitive but structured differently from tech companies. Less equity, more cash and bonus. My final offer was base around $165k with a target bonus of 20-25%. For comparison, a similar role at a fintech startup might have been $155k base with equity that may or may not be worth something.

5 replies

analyst_ana

The cohort analysis SQL question sounds tough. Was it written in a specific dialect? BigQuery, Snowflake, standard SQL? I always get tripped up when syntax varies.

ds_dmitri

They said to use whichever SQL syntax I was most comfortable with. I wrote standard SQL with some PostgreSQL-ish window function syntax and nobody pushed back. Focus on the logic, the syntax is secondary in their grading.

finance_faye

The model explainability angle for credit risk is a regulatory requirement under fair lending law, specifically ECOA. Any ML model used in credit decisioning has to be explainable to regulators and to applicants who get denied. That's why they'd choose logistic regression over gradient boosting in production even if gradient boosting has better accuracy. Good question from them.

de_derek

The comp structure you described (higher base/bonus, lower equity) is typical for financial services. It's also more predictable, which some people prefer. The startup equity dream is mostly that: a dream. Cash pays rent.

ml_mike

Appreciated the breakdown on the stats round. 'Minimum detectable effect' is a filter question at a lot of companies because it tells them whether you actually understand experiment design or just know how to run a t-test. What did they want you to compute it for?