Went through the Pinterest DS loop last quarter for a mid-level role (they call it DS II roughly equivalent to L4/L5 depending on how you map). Here's what to prep.
SQL round. It was an actual live coding SQL session, not a conceptual discussion. The prompt involved aggregating engagement data (impressions, saves, clicks) with a couple of joins and a window function. Specifically a RANK() over a partition. Know window functions cold. They asked me to optimize it after I got a working version. Talked about indexes and why a full table scan was bad on a 10B row table.
Case / analytics round. Classic: "Pinterest is seeing a 10% drop in saves on mobile. How would you diagnose this?" The standard diagnostic framework applies. I walked through: is it in the data or real, segment by platform/OS/country, check recent deploys, look at the funnel step by step. What they care about is your instinct for where to look first and whether you're thinking about the product, not just the query.
Stats / ML round. I got an A/B test interpretation question. They gave me results showing a significant lift in saves but a drop in downstream action (re-pins). Asked me what I'd recommend. This is a trick: conflicting metrics are a deliberate test of whether you optimize for the right thing. I recommended caution and suggested a longer holdout.
Behavioral. One full round on cross-functional influence. Specifically: how have you communicated analytical findings to non-technical stakeholders and changed a product decision.
Overall the DS loop at Pinterest is harder than most tech-adjacent companies and lighter than Google DS or Meta DS. The SQL is real coding, the stats is applied not theoretical, and the case is genuinely product-focused. Good loop for someone who likes the analytical side more than the ML modeling side.