← Uber Interview Insights

Uber·Data Scientist·Technical Phone Screen·Senior

Senior
Apr 2026

Summary

Uber DS interview focused entirely on a surge pricing notification system for drivers. The whole thing was one big open-ended case with several layers peeled back one at a time, which I wasn't fully prepared for.

Questions Asked (6)

Q1

What are the business pros and cons of sending surge-pricing push notifications to nearby drivers?

Product StrategyPricing & MonetizationProduct Sense & Ideation
Author's notes

Started okay.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the objective of surge-pricing push notifications: to increase driver supply in high-demand areas. Then, structure your answer by analyzing the business pros and cons from multiple perspectives—driver behavior, rider experience, and platform economics—and conclude with a recommendation or metric to evaluate success.

Pro tip: Acknowledge the trade-off between short-term supply gains and potential long-term driver dissatisfaction or rider backlash due to perceived price gouging. Emphasize the importance of testing and monitoring key metrics like driver acceptance rate, rider conversion, and overall marketplace efficiency.

1. Clarify the objective and context

Confirm that the goal is to incentivize drivers to move to surge areas, reducing wait times and increasing completed trips. Consider the broader context of Uber's dynamic pricing model.

2. Identify key stakeholders and their incentives

Analyze how drivers, riders, and the platform are affected. Drivers may respond to higher earnings, riders may face higher prices, and the platform aims to balance supply and demand.

3. List business pros

Highlight benefits such as increased driver supply in surge zones, reduced rider wait times, higher trip completion rates, and increased revenue from surge multipliers.

4. List business cons

Discuss drawbacks like driver frustration if surge disappears upon arrival, rider dissatisfaction with higher prices, potential for gaming the system, and notification fatigue.

5. Recommend evaluation metrics and next steps

Propose metrics to measure success (e.g., driver response rate, rider wait time reduction, net revenue impact) and suggest A/B testing to optimize notification strategy.

Key Points to Mention

  • Impact on driver behavior: increased supply, but risk of disappointment if surge ends before they arrive.
  • Rider experience: potential for higher prices and perceived unfairness, but also faster pickups.
  • Platform economics: surge pricing can increase revenue but may reduce demand if prices are too high.
  • Operational efficiency: reducing wait times and improving marketplace balance.
  • Potential unintended consequences: drivers ignoring non-surge areas, notification fatigue, and regulatory scrutiny.
  • Importance of data-driven testing: A/B tests to measure incremental impact on supply, demand, and overall marketplace health.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q2

Design a ranking system that decides how many drivers to notify and which specific drivers to target for surge alerts.

System DesignProduct Analytics & MetricsAlgorithms & Data Structures
Author's notes

This is where it got real.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the business objective and constraints, then propose a two-stage system: first predict the number of drivers needed to meet demand, then rank and select drivers based on expected impact and cost. Emphasize the use of real-time data and machine learning models, and discuss how to evaluate and iterate on the system.

Pro tip: Frame the problem as a resource allocation optimization: maximize incremental completed trips per notification while minimizing driver disruption. Mention the importance of experimentation (A/B testing) to measure the true causal impact of alerts.

1. Clarify Objectives and Constraints

Ask questions to understand the goal: is it to reduce rider wait times, increase driver utilization, or balance supply-demand? Identify constraints like budget, driver fatigue, and real-time latency.

2. Define the Prediction Problem

Predict the number of drivers needed per region and time window to meet expected demand, using time-series forecasting or regression models. Then, for each driver, predict the probability of responding to an alert and the incremental value of their response.

3. Design the Ranking and Selection Algorithm

Rank drivers by expected incremental value (e.g., predicted probability of acceptance times expected earnings) and select the top K drivers, where K is determined by the predicted need. Consider fairness, diversity, and exploration-exploitation trade-offs.

4. Implement and Evaluate

Deploy the system with real-time data pipelines and monitor key metrics like acceptance rate, completed trips, and driver satisfaction. Use A/B testing to measure the causal impact of the alerts and iterate on the models.

5. Address Scalability and Edge Cases

Discuss how to handle high throughput, low latency, and dynamic conditions (e.g., sudden demand spikes). Consider fallback strategies and how to prevent over-notification.

Key Points to Mention

  • Two-stage approach: demand prediction and driver ranking
  • Use of machine learning models for response prediction and demand forecasting
  • Optimization objective: maximize incremental completed trips per notification
  • Real-time data and low-latency serving architecture
  • A/B testing and causal inference to measure impact
  • Driver experience and fairness considerations

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q3

The current approach of using a fixed radius to determine which drivers to notify is inadequate. Why, and how would you improve it with data-driven methods?

Product Analytics & MetricsTechnical Trade-offsData Modeling
Author's notes

Radius ignores actual road geometry, traffic, and driver direction.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by diagnosing the limitations of a fixed radius: it ignores spatial heterogeneity, driver supply, demand density, and behavioral factors. Then propose a data-driven solution that predicts the probability of driver acceptance and optimizes notification radius dynamically, using causal inference and optimization techniques.

Pro tip: Frame the problem as a resource allocation optimization: maximize expected accepted notifications subject to driver fatigue and notification budget constraints. Mention that you would run an A/B test to measure the incremental impact of the new system on key metrics like ETAs and driver utilization.

1. Identify shortcomings of fixed radius

Explain that a fixed radius treats all areas equally, ignoring variations in driver density, demand patterns, and driver responsiveness. This leads to under-notifying in sparse areas and over-notifying in dense areas, causing inefficiencies.

2. Define success metrics and constraints

Establish metrics such as acceptance rate, time-to-accept, driver utilization, and rider wait time. Set constraints like maximum notifications per driver per hour to avoid fatigue.

3. Build a predictive model for driver acceptance

Use historical data to model the probability that a driver accepts a notification as a function of distance, driver characteristics, time, location, and current supply/demand. Consider gradient boosting or logistic regression with spatial features.

4. Optimize notification radius dynamically

For each request, select the set of drivers to notify that maximizes expected accepted notifications or minimizes expected pickup time, using the model predictions and optimization algorithms (e.g., greedy or integer programming).

5. Validate and iterate with experiments

Run A/B tests to compare the new system against the fixed radius baseline. Measure impact on key metrics and refine the model and optimization periodically.

Key Points to Mention

  • Spatial heterogeneity: driver density and demand vary by location and time.
  • Driver behavior: acceptance probability depends on distance, earnings, and personal factors.
  • Dynamic optimization: use real-time data to adjust radius per request.
  • Causal inference: account for confounding when estimating treatment effects.
  • A/B testing: validate improvements with controlled experiments.
  • Scalability: ensure the solution works in real-time for millions of requests.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q4

What would you use as a proxy for driver ETA, and what metrics would you define and compute around it?

Product Analytics & MetricsData Modeling
Author's notes

Blanked for a second on the "proxy" framing.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the business context and the definition of ETA, then propose a proxy that is measurable and aligns with the true ETA. Define a set of metrics that capture accuracy, bias, and reliability of the proxy, and explain how you would compute and monitor them.

Pro tip: Acknowledge that the best proxy depends on data availability and business goals; for example, if actual ETA is unavailable, use the difference between predicted and actual trip duration as a proxy, but validate it against ground truth when possible.

1. Clarify the objective and constraints

Understand why ETA is important (e.g., rider experience, driver allocation) and what data is available (e.g., GPS pings, trip records). Identify if a ground truth ETA exists or if you need a proxy.

2. Propose a proxy for ETA

Suggest a measurable quantity that correlates with ETA, such as the time between driver acceptance and arrival at pickup, or the difference between predicted and actual trip duration. Justify its validity and limitations.

3. Define metrics around the proxy

Outline metrics like Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), bias (mean error), and percentage of trips within a tolerance window. Consider segmentation by city, time of day, and driver.

4. Explain computation and validation

Describe how to compute these metrics using historical data, and how to validate the proxy against ground truth if available. Mention techniques like cross-validation or A/B testing.

5. Discuss monitoring and iteration

Emphasize the need for continuous monitoring, alerting on degradation, and iterating on the proxy as more data becomes available or business needs change.

Key Points to Mention

  • Definition of ETA and its business impact
  • Proxy selection criteria: correlation, measurability, timeliness
  • Metrics: MAE, MAPE, bias, accuracy within tolerance
  • Segmentation by geography, time, and driver
  • Validation against ground truth when possible
  • Monitoring and feedback loops for improvement

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q5

What additional real-time or historical signals should influence which drivers receive a surge push notification?

Product Analytics & MetricsData ModelingProduct Sense & Ideation
Author's notes

I listed driver acceptance rate, recent trip completion status, time since last notification (to avoid spamming the same driver), current heading relative to the surge zone, and shift duration.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by framing the goal: maximize incremental completed trips from the notification without harming rider or driver experience. Then segment signals into real-time (e.g., driver location, current supply/demand) and historical (e.g., driver responsiveness, past surge acceptance) and explain how each would inform targeting. Finally, propose a model or experiment to validate the impact.

Pro tip: Emphasize that the notification should be sent only when the expected incremental value (e.g., additional trips) exceeds the cost (e.g., driver annoyance, notification fatigue). This shows you think about ROI and long-term engagement, not just short-term lift.

1. Clarify the objective and constraints

Define success metrics (e.g., incremental completed trips, driver acceptance rate) and constraints (e.g., notification fatigue, driver satisfaction). This ensures your signal selection aligns with business goals.

2. Identify real-time signals

List signals available at the moment of decision, such as driver's current location, distance to surge zone, current earnings, time since last trip, and local supply-demand imbalance.

3. Identify historical signals

Consider past behavior like driver's historical responsiveness to surge notifications, acceptance rate during similar conditions, typical working hours, and long-term engagement metrics.

4. Prioritize and model signals

Explain how to combine signals into a predictive model (e.g., propensity to accept) and prioritize based on expected incremental value. Mention techniques like uplift modeling or A/B testing.

5. Validate and iterate

Propose an experiment design to measure the impact of the notification strategy, including holdout groups and long-term metrics, and discuss how to iterate based on results.

Key Points to Mention

  • Real-time supply-demand imbalance and driver proximity to surge zones
  • Driver's historical responsiveness to surge notifications and acceptance rates
  • Driver's current state: time since last trip, earnings goal progress, and fatigue
  • Notification fatigue and the risk of uninstalling or disabling notifications
  • Incremental value: measuring the lift in completed trips versus a control group
  • Contextual factors: time of day, day of week, weather, and special events

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.

Q6

If neighborhood-level supply-demand imbalance is used as a feature, how would you detect and quantify that imbalance?

Product Analytics & MetricsData ModelingA/B Testing & Experimentation
Author's notes

Probably my cleanest answer.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining what supply and demand mean in the Uber context at the neighborhood level, then propose a method to quantify imbalance using a ratio or difference metric. Discuss how to aggregate data over time and space, and how to validate the feature's usefulness in a model or experiment.

Pro tip: Emphasize that imbalance should be measured relative to a baseline or expected level, and consider using a smoothed metric to avoid noise from small sample sizes in neighborhoods.

1. Define supply and demand

Clarify what constitutes supply (e.g., available drivers) and demand (e.g., ride requests) at the neighborhood level, and specify the time window (e.g., hourly).

2. Choose an imbalance metric

Select a metric such as supply/demand ratio, difference, or log ratio, and justify why it captures imbalance effectively.

3. Aggregate and smooth

Aggregate data to neighborhood-hour level, apply smoothing (e.g., moving average) to reduce noise, and handle missing data.

4. Validate and refine

Test the feature's predictive power or impact in an A/B test, and refine the metric based on performance and business relevance.

Key Points to Mention

  • Definition of supply and demand in Uber's context (e.g., drivers vs. ride requests)
  • Choice of imbalance metric (ratio, difference, log ratio) and its properties
  • Spatial and temporal aggregation (neighborhood level, time windows)
  • Handling of sparse data and smoothing techniques
  • Validation through model performance or A/B testing
  • Consideration of external factors (e.g., events, weather) that affect imbalance

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.