Went through the Vercel DS/analytics loop last quarter. The role was data scientist on the growth team. This was a bit unusual because Vercel is primarily known as an infra/dev-tools company so I wasn't sure what to expect from their data practice.
Stage 1: SQL screen. A 45-minute SQL exercise over a shared notebook. The schema they gave me was clearly modeling a SaaS deployment product (projects, deploys, users, usage events). Queries ranged from joining across tables to compute retention cohorts, to a trickier one involving window functions to find the first successful deploy per user. Nothing exotic but it required clean thinking about what the data actually meant in a product context.
Stage 2: Case interview. Given a product scenario: "Our 30-day deploy success rate dropped 5 percentage points month over month. Walk me through how you'd investigate." Classic analytics case format. They wanted decomposition (is it a segment issue? A new cohort? A geography? A framework change?), hypothesis prioritization, and what data/queries you'd pull. They did NOT want a polished deck, just structured reasoning out loud.
Stage 3: Stats round. Shorter than I expected, about 30 minutes. Covered A/B testing setup, minimum detectable effect, sample size, and one question about a result that was directionally positive but not statistically significant. What do you do with it? How do you communicate to stakeholders who want to ship anyway? That last part was as much about communication as statistics.
Stage 4: Behavioral. Same themes as the rest of the Vercel loop. Autonomy, async communication, how you've influenced without authority. The DS-specific angle was: tell me about an analysis that changed a product decision. Have a real story here with the actual decision and actual outcome.
Overall. They're not a data-heavy company yet in the sense that Airbnb or Meta is. The team is small and the problems are more product analytics than ML-heavy. If you're coming from a team that ran complex models in production, recalibrate your expectations. The core of this role is helping a relatively small team make better decisions faster, not building ML pipelines.