Went through the Mercury DS loop a couple months ago. Here's the breakdown by round because that's how I'd want to read it.
SQL round (45 min): Live SQL, shared doc or their own environment. Problems ranged from medium to hard. One was a window functions problem: calculate rolling 7-day transaction volume per account. Another involved a multi-table join with some tricky conditions around transaction statuses. They care about query performance, not just correctness. I mentioned adding an index and the interviewer followed up with "why that column and not another." Good question.
Stats + experiment design (45 min): This was the round I was least prepared for. One question: Mercury is rolling out a new feature for business accounts. Design an A/B test. That sounds simple until they start poking: how do you handle network effects between business accounts in the same industry? What's your minimum detectable effect? How long do you run it before you can call it? I stumbled on the network effects piece initially and they helped me think through it, but I think I lost points there.
Case / product analytics (45 min): They gave me a scenario: transaction failure rate has gone up 3% week over week. Walk me through your investigation. This was my strongest round. I broke it down: was it a specific payment rail? A merchant category? A time-of-day pattern? New customers or existing? They added information as I asked for it, which is a good sign that the round is meant to be collaborative.
Behavioral (30-45 min): Standard STAR, Mercury values. One question was specifically about presenting analysis to a non-technical stakeholder. Have a story ready.
Overall: the DS loop at Mercury is heavier on SQL than most. Stats matters but it's secondary. The case question is really testing product thinking as much as data skills. Prepare for all three.