Uber Data Scientist Interview Loop
The Uber data scientist interview has a 30-minute recruiter call, a technical screen, and a virtual onsite of four rounds: SQL, experimentation and statistics, product sense, and behavioral. A Bar Raiser also runs the debrief and has a big say in your level. Some teams add a take-home. Rides, Eats, Freight and Safety each run slightly different versions, so ask which one you're getting. What sets Uber apart is how deep it goes on marketplace experiments. Every company will ask you to design an A/B test. Uber will ask what happens when drivers in your treatment group take trips away from riders in your control group.
What decides it
Marketplace thinking, and the Bar Raiser. Generic big-tech prep gets you through the SQL. It stalls when every case has two sides.
Practice by round
Live SQL in a plain editor, sometimes followed by a statistics question. Interviewers are friendly and will nudge you. That's a signal too, because if you need the nudge, it gets noted.
Uber's data at Uber's difficulty. You'll work on trips in consecutive months, completions within a 168-hour window, actual versus predicted arrival time, revenue loss by order status, distance per dollar, trip distance by city, and riders who never rode.
| Question | Difficulty |
|---|---|
| Sum Of Numbers | Medium |
| Find The Combinations | Medium |
| Find Pairs with Product Over 11 | Medium |
| Advertising Channel Effectiveness | Medium |
| Highest Total Miles | Medium |
| Vehicle Gear Usage | Medium |
| Find The Most Profitable Location | Hard |
| Most Senior & Junior Employee | Hard |
| Maximum Number of Employees Reached | Hard |
| Trips in Consecutive Months | Hard |
| Positive Ad Channels | Hard |
| Distance Per Dollar | Hard |
| Naive Forecasting | Hard |
A marketplace metric has moved. Maybe completion dropped in one city, or supply dipped at peak. Work out the cause across both sides of the market, and name the trade-offs before the interviewer does. An incentive that fixes supply costs margin. A surge that clears the market loses riders.
| Case study | Difficulty |
|---|---|
| Favorite Product or App | Easy |
| Main Outcome Metrics | Easy |
| Accepting Rides | Easy |
| Determining Origin City | Medium |
| Metrics to Acquire Customers | Medium |
| Uber Black Rides | Medium |
Experiments where the two sides of a market contaminate each other's groups. That includes spillover between treatment and control, and what to do when rider and driver results conflict. There's also spotting an anomaly in a distribution, and time-series forecasting. Explain how things work, as well as what you'd choose.
| Question | Difficulty |
|---|---|
| Anomaly in Distribution | Easy |
| SVM Classifier | Medium |
| Reading Results Across Unequal Markets | Medium |
| Guardrails for a Two-Sided Change | Medium |
| When Rider and Driver Results Conflict | Hard |