Practice real data science interviews tuned by the interviewers who actually run them.

Facebook Groups wants to test a new feature: AI-generated discussion prompts posted automatically in groups with declining activity. The goal is to re-engage dormant groups. How would you design this experiment?

You're a data scientist at a food delivery company. The operations team ran a test where they gave Dashers a guaranteed minimum per delivery in a specific market. Results are in — but they're mixed.

Facebook is integrating Reels into the main Newsfeed experience. This is a major surface change that could boost short-form video consumption but risks cannibalizing existing content types. You need to design the experiment and measurement plan.

The Messenger team wants to improve the video call experience. They have several ideas — better connection quality, AR filters, screen sharing, and background blur. You need to design an experiment to test improvements and define what success looks like.

Amazon's search team developed a new ranking algorithm that prioritizes seller quality (reviews, fulfillment speed, return rate) more heavily. Design the experiment and evaluate the results.

Uber is replacing discrete surge tiers (1.0x, 1.2x, 1.5x, 2.0x) with dynamic continuous surge (any multiplier from 1.0x to 3.0x, updated every minute). Design the experiment and evaluate the results.

Reddit is redesigning new-user onboarding. Current: shows popular subreddits. Proposed: personalized recommendations based on stated interests. Design the experiment and evaluate the results.

DoorDash is launching in Austin (sprawling, car-dependent) and Portland (compact, bike-friendly). The current dispatch algorithm optimizes solely for distance-to-restaurant and was trained on suburban markets. It systematically under-assigns bike Dashers in dense areas. Design an experiment to test city-optimized dispatch.

DoorDash launched Peak Alerts — push notifications urging offline Dashers to go online during high-demand periods. Headline metrics look great: 28% more Dashers during alerts, 62% fewer unfulfilled orders. But deeper analysis may reveal that Peak Alerts cannibalize off-peak supply instead of creating net new supply.

The subscription team wants to test a $2/month price increase for DashPass (from $9.99 to $11.99). You're the data scientist tasked with designing this experiment. Pricing experiments have unique challenges — long-term churn effects aren't visible in short tests, and existing vs. new subscribers react very differently.

The marketing team wants to test a 5% off delivery fee coupon to drive consumer orders. But you know that many orders would have happened anyway. How do you measure the true incremental impact when marketplace dynamics create spillover effects?

You're testing a new real-time batching algorithm that assigns 2 orders to a single Dasher trip. This affects supply, demand, and timing across the entire local marketplace. Standard A/B testing breaks down because of interference effects — you need a switchback design.

DoorDash sends push notifications to offline Dashers during high-demand periods, asking them to go online. The operations team wants to know if this is actually effective at increasing supply, or if Dashers would have come online anyway.

DoorDash is testing a new restaurant ranking algorithm that prioritizes delivery speed and proximity over ratings and popularity. Early data looks promising for delivery times, but there are concerns about fairness to merchants and consumer discovery.

The hypothesis is that late deliveries cause customer churn. But you can't randomly assign late deliveries — you need to find a clever identification strategy to prove or disprove the causal relationship and size the business impact.

Netflix historically offered a 30-day free trial to new users, but recently eliminated it in many markets. Leadership wants to understand: was the free trial actually effective, and was removing it the right call? You need to define how you'd measure trial effectiveness, detect trial abuse, compare trial lengths, and evaluate the removal decision retroactively.

Netflix is testing a new trailer autoplay behavior on title cards — when users hover for 2 seconds, a 30-second trailer starts playing. The test shows higher play starts but unchanged total watch time. The team is split on whether this is a real UX improvement that helps users discover content faster, or metric inflation that makes the dashboard look better without creating real value.