Uber Data Scientist interview typically runs 5 rounds: recruiter screen, 2 technical screens, and 3 final interviews. It usually takes about 4-8 weeks and is notably structured, with a bar raiser in the final loop.
$126K
Avg. Base Comp
$311K
Avg. Total Comp
4-5
Typical Rounds
2-4 weeks
Process Length
We've seen Uber consistently favor candidates who can reason through messy marketplace tradeoffs, not just recite textbook methods. Multiple candidates reported questions that pushed beyond standard A/B testing into synthetic control, difference-in-differences, interrupted time series, and especially switchback testing. That tells us the bar here is less about naming the right framework and more about knowing why a framework fits a rides, delivery, or pricing problem. When candidates stumbled, it was usually because they stayed too abstract or couldn't defend the assumptions behind their design.
A recurring theme is that Uber cares deeply about metrics that map to marketplace health. Candidates were asked to define success for ETA changes, batching, surge pricing, driver assignment, and service expansion, and the strongest answers tied primary metrics to operational outcomes like match rate, completed rides, or order volume, then layered in guardrails like cancellations, complaints, retention, and driver earnings. We've also seen interviewers probe failure modes hard: what happens when an experiment looks off, how to verify results, and how to handle endogeneity when pricing and demand move together. That emphasis shows up again in the econometrics-style questions around price elasticity and surge pricing.
The non-obvious separator is clarity under ambiguity. Several candidates noted that the coding itself was not the hardest part; the harder part was framing the problem, stating constraints, and explaining tradeoffs without getting lost. Even behavioral feedback points in the same direction: the bar raiser expects stories that reflect Uber's values in concrete business decisions, not polished leadership language. In practice, our candidates who did best sounded like people who had actually thought about how a marketplace behaves when incentives, supply, and customer experience all interact.
Synthesized from 6 candidate reports by our editorial team.
Had an interview recently?
Share your experience. Unlock the full guide.
Real interview reports from people who went through the Uber process.
I went through seven rounds total for a Data Scientist role at Uber.
The process started with an online assessment covering SQL and statistics. From there, I had two technical screening rounds: the first included a Python medium-level problem, pandas basics, and an experimentation case study; the second focused on my CV, past projects, and a case study around launching a promotional campaign.
I then had four onsite rounds, which repeatedly tested the same themes — past projects, the statistics behind them, and experimentation and causal inference case studies. The last round was a team matching discussion, which felt more informational than eliminatory.
Questions asked during the process included:
Given two lists, find the minimum absolute difference across all possible combinations. I was able to produce a brute-force single-answer solution; the interviewer did not push for an optimized approach.
A pandas question covering basic filtering, datetime ranking, and row ranking operations.
A case study on breakfast order volume being lower than lunch and dinner. I was asked to work through conversion rates, identify possible causes across different pages, design an A/B test around a lowered cart value threshold, discuss session-level vs. customer-level spillovers, and handle edge cases like a confidence interval that just touches zero.
An in-depth discussion on CUPAC applied to an AOD (Always-On Deals) setting step by step, with context provided around Deliveroo and CUPED. I was also asked about the drawbacks of alternating experiments and how the DS role changes after implementing a switchback design.
A promo campaign case study on Uber Eats: defining goals, identifying metrics, convincing stakeholders to run an experiment instead of a pre/post or used-vs-unused comparison, designing the experiment, handling spillovers at high order volumes, defining significance and power, and deciding what to do when results are not statistically significant or confidence intervals only mildly overlap zero.
A second Uber Eats case study focused on an ads platform, covering how the experience differs between small/medium merchant ads and enterprise-grade setups, what inputs are needed from SMB restaurants during onboarding, how to design an experiment for changing the recommendation algorithm, how to define similar restaurants for new entrants, and how to trade off between retention and ad revenue.
A fourth onsite round focused on switchback experiment design and CUPAC in detail, assumptions of the fixed-effects model, whether fixed effects change the magnitude and direction of the ATE, knowledge reuse across teams (I spoke about triple difference-in-differences), mentoring junior data scientists in stakeholder management and work reviews, and use of AI in incident analysis.
A home feed restaurant ranking case study covering multi-objective optimization, what the objective function should be (GMV vs. orders vs. AOV — requiring a strategic answer), how training data would be structured and at what granularity, what features would go in and how to transform them, model selection (I suggested starting with logistic regression and then trying a random forest), handling class imbalance, and whether to predict class labels or conversion likelihoods.
Share your own interview experience to unlock all reports, or subscribe for full access.
Sourced from candidate reports and verified by our team.
Topics based on recent interview experiences.
Featured question at Uber
Write a query to select the top 3 departments with at least ten employees and rank them according to the percentage of their employees making over 100K in salary.
| Question | |
|---|---|
| Download Facts | |
| Experiment Validity | |
| User Experience Percentage | |
| Distance Traveled | |
| Button AB Test | |
| Weighted Keys | |
| Top 3 Users | |
| Third Purchase | |
| Bank Fraud Model | |
| Maximum Profit | |
| Encoding Categorical Features | |
| WAU vs Open Rates | |
| Sum to N | |
| P-value to a Layman | |
| Google Maps Improvement | |
| Christmas Dinner Ingredient Optimization | |
| Random Forest Explanation | |
| Type-ahead Search | |
| Sort Strings | |
| Random Weighted Driver | |
| Hurdles In Data Projects | |
| Uber User Journey | |
| Cancellation Fees | |
| Dijkstra implementation | |
| Assumptions of Linear Regression | |
| Testing Price Increase | |
| Dice Rolls From Continuous Uniform | |
| Drawing Balls From Bin | |
| Type I and II Errors |
Synthesized from candidate reports. Individual experiences may vary.
The process typically starts with an HR or recruiter conversation about your background, resume, and overall fit for the Data Scientist role. In some cases, Uber also shares interview prep materials ahead of time and uses this stage to set expectations for the technical loop.
This first technical screen combines SQL/Python execution with analytics and experimentation thinking. Candidates have reported questions on SQL patterns like joins, window functions, lag/lead, and CTEs, along with basic Python tasks and an applied case study around product metrics or feature launch design.
The second screen is similar in spirit but often more interactive and case-driven. It can include coding, causal inference, A/B testing, confidence intervals, and product or marketplace reasoning, such as how to evaluate a feature change, define metrics, or think through an optimization problem.
The final loop usually includes a hiring manager interview, a bar raiser behavioral round, and one or more technical or business-focused interviews. Topics reported include experimentation failure modes, experimentation design, experimentation-focused metrics, and econometrics or marketplace cases such as surge pricing, price elasticity, Uber Eats batching, or driver ETA changes.
After the final loop, Uber makes a decision based on technical depth, product sense, and behavioral alignment with company values. Candidates reported hearing back relatively quickly in some cases, with outcomes ranging from offer to rejection after panel review.