Went through the Datadog DS loop a few months back for an analytics-focused role (not ML engineering, more product analytics DS). Posting specifically because I had to piece this together from almost nothing and I know people are searching for it.
Loop structure: Recruiter call, technical phone screen, then a 4-round final day.
Technical phone screen: 45 minutes. Entirely SQL. Two problems. First was medium complexity: joining multiple tables, window functions, filtering on conditions. Second was harder: something involving calculating a rolling average with a specific window and handling nulls correctly. No Python, no stats in this round. Know your SQL cold.
Final round breakdown: SQL round 2 (yes, another SQL round). More advanced. I got a question that was essentially building a funnel analysis query from raw event data. Counting unique users at each step, handling out-of-order events, computing drop-off rates. This is the kind of SQL you write in real analyst jobs, not LeetCode SQL. Stats and probability round. Mostly applied: A/B testing setup, how you'd handle novelty effect, what p-value means without jargon, confidence intervals. They gave me a scenario where two variants showed a statistically significant difference but one had high variance, asked what I'd do. No trick math, just conceptual clarity. Case study / product sense. Given a hypothetical Datadog metric (something like 'DAU on our dashboards dropped 15% week over week') walk me through how you'd investigate. Classic diagnostic framework, but they wanted specifics about what data you'd pull first and why. Behavioral. Standard. Cross-team stuff, a time you influenced a decision with data, a time data told you something unexpected.
Comp for a mid-level DS role (NYC): my offer was in the 160-185k total comp range. Likely higher for senior or ML-engineering roles.
Bottom line: SQL is their litmus test. If you're weak on SQL, you won't clear the first two rounds regardless of how strong your ML background is.