Went through a Twilio DS loop a few months ago for a role on their Growth data science team. Here is what the interview actually covers, because the prep guides I found online were out of date.
Round structure (4 rounds):
SQL (45 min) Two problems. The first was a funnel analysis query: developer sign-up to first API call, broken down by acquisition channel. Standard window functions, CTEs, some date math. I'd rate it medium difficulty.
The second was messier: a schema with three tables, some nulls in unexpected places, and they asked me to calculate message deliverability rate broken down by carrier and country over a 30-day window. The null handling was the real test. If you're not comfortable with COALESCE and conditional aggregation, brush up.
Case / Product Analytics (60 min) Given a drop in message delivery rate over the past 2 weeks, walk me through how you'd diagnose it. This is a DS version of the classic product sense question. I laid out a structure: check if it's global or localized (carrier, geography, customer segment), separate the numerator (messages attempted) from denominator effects, look for changes in customer mix, then check for infrastructure or code deploys that coincide. They pushed: "say it's isolated to one carrier in Brazil, now what?"
Stats / ML (45 min) Basic A/B testing framework questions: how do you size a test, how do you handle novelty effects, what do you do when you detect an interaction effect across segments. One question on experiment design where the unit of randomization and the unit of analysis are different (network effects in a messaging platform).
Also one question on a practical modeling problem: given a feature set, how would you predict which customers are at risk of churning from a messaging API? They didn't want a full model spec, just the approach and what data you'd want.
Behavioral (30 min) Standard. Tell me about an insight that changed a product decision. Tell me about a time you had to communicate uncertainty to a non-technical stakeholder.
What I'd prioritize in prep: SQL window functions and clean funnel query writing, the diagnosis framework for product metric questions, and knowing how to talk about A/B testing trade-offs without going too academic.