Wrapped a DS loop at Wells Fargo targeting their consumer analytics team. Level was mid-senior (5 YOE). Here's every section of the interview with what actually got asked.
SQL round (45 min): This was the hardest SQL I've seen outside of a pure analytics engineering role. They do window functions, partitioning, running totals, lag/lead. The specific problem was something like: given a transaction table with customerid, amount, transactiondate, and transactiontype, find customers who had a transaction decline followed by a successful transaction within 72 hours. They want to see you think through the join logic and handle edge cases like multiple declines.
They use Snowflake internally. Syntax matters. Know that DATEDIFF and DATEDIFF behave differently in Snowflake vs SQL Server.
Stats / probability round (30 min): Not a stats theory test. They asked about a real scenario: we're seeing fraud rates spike on mobile transactions. Walk me through how you'd diagnose whether this is a true trend or a sampling artifact. Classic A/B testing fundamentals, power, significance, but framed as a detective problem. They pushed on: how would you control for the fact that mobile transactions skew toward younger customers who might just have different fraud patterns anyway?
Case / business round (45 min): My case: a new credit card product has launched and early activation rates (customers who get the card and actually use it within 30 days) are below forecast. What might explain this and how would you investigate? I structured it as: funnel analysis, cohort comparison, external factors, and proposed a decision tree of what data I'd pull first. They liked that I started with diagnostic framing before jumping to hypotheses.
Behavioral (30 min): One question that surprised me: tell me about a time your analysis led to a decision that turned out to be wrong. What did you do? WF culture stuff again. Have a real answer for this.
Comp offer I received: $148k base + $20k sign-on + 8% target bonus. Charlotte, hybrid 3 days in-office. Accepted.