← Google Interview Insights

Google·Data Scientist·Technical Phone Screen·Senior

Senior
Apr 2026

Summary

Data scientist interview at Google with two fairly meaty experimentation questions. Both were product-analytics flavored with enough statistical nuance to trip you up if you're not careful about the details.

Questions Asked (2)

Q1

Your PM wants a two-week A/B test to evaluate raising subscription prices. How would you design it, should you even A/B test pricing at all, and how would you decide whether the new price is actually better?

A/B Testing & ExperimentationPricing & MonetizationProduct Analytics & Metrics
Author's notes

Two weeks is almost certainly not enough and I said so upfront, which felt risky but I think was the right call.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by questioning whether a two-week A/B test is appropriate for pricing, given its unique challenges like long-term effects and fairness concerns. Then outline a rigorous design that addresses these challenges, including randomization, metrics, and decision criteria. Finally, discuss how to interpret results and make a recommendation.

Pro tip: Pricing tests can be risky: consider a staggered rollout or geo-based test to mitigate user backlash and revenue loss. Also, focus on long-term metrics like retention and LTV, not just short-term conversion.

1. Assess feasibility and risks

Evaluate if A/B testing pricing is suitable: consider ethical, legal, and user experience issues. Discuss alternatives like historical data analysis or surveys.

2. Design the experiment

Define hypothesis, randomization unit (e.g., user), sample size, duration, and success metrics (e.g., conversion, revenue, retention). Address potential pitfalls like novelty effects and seasonality.

3. Analyze results

Use statistical tests to compare metrics between control and treatment. Check for significance, confidence intervals, and segment-level effects.

4. Decide and iterate

Weigh statistical significance against practical significance (e.g., revenue impact). Consider long-term effects and recommend rollout or further testing.

Key Points to Mention

  • Pricing tests can lead to customer backlash and long-term brand damage; consider ethical implications.
  • Randomization at user level may not be feasible; consider geo or time-based splits.
  • Key metrics: conversion rate, average revenue per user, retention, churn, and customer lifetime value.
  • Two weeks may be too short to capture long-term effects; consider extending or using holdout groups.
  • Statistical power and sample size calculations are crucial to detect meaningful differences.
  • Segment analysis by user demographics or behavior can reveal heterogeneous treatment effects.

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

Q2

A sign-up page has a red button at the top. The team wants to test red vs blue and top vs bottom placement to improve click-through. How do you structure and analyze that multivariate test?

A/B Testing & ExperimentationProduct Analytics & Metrics
Author's notes

Straightforward multivariate setup but I fumbled the interaction effects part.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining the goal and metrics, then design a full factorial experiment with two factors (color and placement) at two levels each, ensuring proper randomization and sample size. Analyze using ANOVA or regression to test main effects and interaction, and interpret results in terms of practical significance for click-through rate.

Pro tip: Don't forget to check for interactions between color and placement—sometimes the best combination isn't the best individual levels. Also, consider sequential testing or Bayesian methods if you need to peek at results early, but be transparent about the trade-offs.

1. Define objective and metrics

Clarify the primary metric (click-through rate) and any secondary metrics (e.g., conversion rate, user engagement). Establish the minimum detectable effect and required sample size.

2. Design the experiment

Use a 2x2 factorial design with four variants: red-top, red-bottom, blue-top, blue-bottom. Randomly assign users to one of the four groups, ensuring balanced allocation and no confounding.

3. Run the test and collect data

Launch the experiment, monitor for technical issues, and collect data on user interactions. Ensure the test runs for a sufficient duration to capture typical user behavior and avoid novelty effects.

4. Analyze results

Use ANOVA or a regression model with interaction terms to estimate the main effects of color and placement, and their interaction. Check for statistical significance and effect sizes.

5. Interpret and decide

Determine which combination yields the highest click-through rate, considering both statistical and practical significance. Recommend the winning variant or suggest further testing if results are inconclusive.

Key Points to Mention

  • Full factorial design to test main effects and interaction
  • Randomization and sample size calculation to ensure power
  • Use of ANOVA or regression with interaction term for analysis
  • Checking for practical significance and business impact
  • Potential pitfalls: multiple comparisons, novelty effects, and external validity
  • Consideration of sequential testing or Bayesian approaches for early stopping

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