Went through Figma's DS interview loop earlier this year for a data scientist role on their growth team. The job description mentioned "product analytics" and "experimentation" and the interview reflected that pretty literally.
Here's the structure I got:
SQL round (60 min). Two to three queries escalating in complexity. Started with a basic aggregation. Ended with a problem involving window functions and a subtle join that had an intentional trap (many-to-many relationship that causes row multiplication if you're not careful). They're clearly testing whether you understand what your query is actually doing, not just whether you can produce an output. I had to explain my reasoning at each step.
Stats / experiment design (45 min). This was the most important round. They gave me a scenario: Figma is testing a new feature in their editor. Walk me through how you'd design and evaluate the experiment. Questions covered: choosing the right metric, unit of randomization (user vs. session vs. team), minimum detectable effect, how to handle network effects (since Figma is a collaborative product, treating users as independent is questionable). The network effects piece is specific to collaborative tools and they clearly cared about it.
Case / product analytics (45 min). Given a messy metric: daily active collaborators dropped 8% over the past two weeks. Diagnose it. This is the classic metric investigation case. Be structured. I used a top-down breakdown: is it global or segment-specific, is it a data artifact or real behavior, what events correlate, etc. They added info as I asked. Nobody expects you to diagnose correctly in a vacuum.
Behavioral (30 min). Standard. Impact, collaboration, working with ambiguity. Same themes as the SWE loop.
Stack they mentioned: Snowflake, dbt, Looker, some Python for modeling. Knowing dbt basics will help you sound like you'll actually be useful on day one.
Level I was interviewing for was mid-senior. TC offer came in around $210-230k for SF. Base was $155k, rest equity.