Just wrapped my Meta DS loop last month. Role was on the Ads measurement team, IC4 level. Sharing the full breakdown because the posts I found were 2-3 years old and some things have shifted.
What's the same: SQL is still a big deal. I got two dedicated SQL rounds. One focused on window functions and the kind of aggregation you'd need for funnel analysis. The second was more of a product-metrics scenario where I had to write the query AND explain what the numbers would mean.
What's changed (or at least what I saw): The ML round had more focus on causal inference than I expected. They asked how I'd design an experiment when I can't randomize (observational data scenario). That's not standard "explain gradient boosting" territory. Brush up on DiD, propensity scoring, synthetic control.
They also asked a question about metric selection. Specifically: "if you could only track one metric for this product, what would it be and why." Sounds soft but it's a trap for people who list ten metrics instead of thinking hard about one.
Product sense: One round that was basically PM product sense but from a data angle. "DAU dropped 8% last Tuesday. Walk me through how you'd diagnose it." The structure matters a lot here: segment first, don't jump to conclusions.
Behavioral: Two behavioral rounds, both heavily focused on cross-functional conflict and data-driven decision making under ambiguity. Standard STAR structure works fine.
Timeline: Applied through referral. Recruiter screen to offer was 6 weeks.
Overall the loop is longer than I expected (5 rounds plus a recruiter screen) but each round felt purposeful. The SQL and causal inference prep is non-negotiable.