Discord · Primly Community

Discord data scientist interview, SQL + case + stats breakdown (2026 loop)

ds_dmitri · 5 replies

Went through Discord's DS interview loop in late 2025 and early 2026 for a role on their product analytics team. Here's exactly what the process looked like and what they tested.

Process overview: Recruiter screen, HM call, then a 4-round virtual onsite. No take-home for this DS role (apparently some teams do have one, so clarify with your recruiter).

SQL round: Hands down the most technically demanding SQL I've had in a consumer company interview. It wasn't just "write a query to find users who did X." They gave me a realistic schema with events tables, user tables, server membership data, that kind of thing, and asked me to answer a business question. One of the prompts was something like: find the top servers by 30-day retention of new users who joined in the last quarter, and then identify if there's a difference in retention by server category. That requires window functions, multiple CTEs, conditional aggregation. Not beginner stuff. Know your window functions cold.

Stats / probability round: Standard DS interview territory but they went deeper than most. Topics that came up: A/B testing and when to use one-sided vs two-sided tests, power analysis basics (they wanted to know if I understood why sample size matters before running an experiment), Simpson's paradox (they had a real example from their data framing), and one bayesian probability question that was more of a thought exercise than a computation.

Case study / product analytics round: This is the one that felt most Discord-specific. They gave me a scenario: a community feature's DAU dropped 15% week over week. Walk them through your investigation. The right move is to break it down systematically: is it real or a tracking issue, is it global or segment-specific, new users vs. retained users, any recent product changes or infrastructure incidents. I found this round more interesting than a standard case because the product surface (communities, servers, voice channels) is rich enough to reason about genuinely.

Behavioral / cross-functional round: Standard senior DS stuff: how do you communicate findings to PMs who disagree with your analysis, have you killed a project based on data, how do you work when you don't have clean data. They care about influence and stakeholder work, not just technical execution.

Comp context: For a mid-to-senior DS role (L5 equivalent internally), I've seen ranges around 200-240k TC in remote/SF depending on the team. A friend who joined last year is on the growth analytics team and mentioned her package was at the lower end of that because it was partially equity-heavy and Discord is pre-IPO.

Bottom line: If your SQL is rusty, that's where to focus first. The stats round is fair. The product case is the most fun and the most differentiated.

5 replies

analyst_ana

The SQL question about server retention sounds genuinely hard. Do they let you use any resources during the SQL round or is it fully live coding from memory?

ds_dmitri

Live coding, shared screen, no docs. They weren't cruel about it but you can't look things up. If you know your SQL you'll be fine. If you're rusty on CTEs or window functions, practice those specifically.

ml_mike

Simpson's paradox coming up is interesting. That's one of those things that's worth having a concrete example ready for. The usual one (Berkeley admissions) is fine but if you can connect it to engagement data or funnels it'll land better.

de_derek

Curious if they care at all about data engineering skills or if the DS role is purely analytical. Like do they ask about pipelines, dbt, anything like that?

ds_dmitri

Not for this role. Very analytics-side DS. They asked about tools I use day to day but it was more like conversational context setting, not a technical test on infra.