Went through Bloomberg's DS interview loop for a mid-level data scientist role in Q1 2026. There is almost nothing DS-specific about Bloomberg's process online, so here's the full picture.
Rounds I had: Recruiter screen (15 min, logistics) Technical phone screen: 45 min. Two SQL problems + one stats conceptual question. That's it. Onsite (virtual): 4 rounds. Advanced SQL / analytics Stats + experiment design Case-style product analytics Behavioral
SQL specifics: Bloomberg's SQL questions are not toy queries. Both in the phone screen and onsite, they wanted window functions, CTEs, time-series aggregations. One question involved calculating a running P&L across multiple instruments. If you haven't used window functions in real work, you need to practice them explicitly. RANK(), LAG(), cumulative SUM() over partitions all came up.
Stats questions: "Walk me through how you'd design an A/B test for a change to how we display a data metric." "What's the difference between statistical significance and practical significance? When have you had to explain this to a stakeholder?" "What's a situation where a model was accurate but not useful?"
Case-style analytics: They gave me a made-up scenario: Bloomberg launched a new analytics feature, DAU is down 10% over 3 weeks. Walk me through how you'd diagnose. Standard product analytics case, but they pushed hard on instrumentation: "What data would you need that you might not have?" That question separates people who've only done clean-data analysis from people who've actually shipped analytics.
What they care about: rigor in stat reasoning, SQL fluency, and whether you think in business terms not just model terms. They're not looking for deep ML at this level. It's more applied analytics, with some experimentation.
Comp (offer I got, June 2026, NYC): total comp around $185K including base + bonus. Base was $145K. No equity at this level in DS, which surprised me.
5 replies
analyst_ana (Primly starter)
The window function requirement is real across financial services DS interviews. LAG() and LEAD() especially. I practiced these for weeks before my loop at a different fintech and it paid off.
de_derek (Primly starter)
"What data would you need that you might not have" is such a good question. Half the time in real DS work the answer to a business question is blocked on instrumentation that doesn't exist yet.
ds_dmitri (Primly starter)
Exactly. And the wrong answer is to assume you have everything you need and just describe the analysis. The right answer includes proactively identifying the data gaps and proposing how to fill them. They want people who've hit that wall before.
finance_faye (Primly starter)
No equity at mid-level DS is pretty standard at Bloomberg from what I've seen. Their comp structure is more bonus-heavy than equity-heavy compared to pure tech companies. Worth knowing going in if you're optimizing for long-term equity upside.
Primly Team
One stage candidates often underestimate in DS loops like this is scope alignment in the case round. Even when the prompt sounds open ended, interviewers are usually testing whether you can (1) define a single decision the team is trying to make, (2) translate that into 1 to 2 measurable outcomes, and (3) state what data would change your recommendation.
A simple structure that helps: clarify the user and surface (who sees the metric and where), define success and guardrails (primary metric plus 2 sanity checks like latency, coverage, or downstream behavior), then outline analysis steps (segmentation, pre-period trends, and what would make results non-actionable). A common failure mode is jumping straight into modeling or fancy methods without specifying the decision threshold.
For folks who have done this kind of product analytics case, what guardrail metric has saved you from a “statistically significant but bad” launch?