← Amazon Interview Insights

Amazon·Data Scientist·Technical Phone Screen·Intermediate

Intermediate
Jun 2026

Summary

Amazon data scientist interview with a product analytics case around an energy-usage dashboard redesign. Pretty standard experimentation setup but the depth they wanted on statistical methodology caught me a bit off guard.

Questions Asked (1)

Q1

The product team is launching a redesigned energy-usage dashboard. How would you design an A/B test to measure whether it improves user engagement, and walk through your choice of metrics, statistical approach, and how you'd determine test duration?

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

This is the kind of question where you can easily go wide and shallow and they'll just keep poking until you fall apart.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the goal: define what 'improved user engagement' means for the energy dashboard, then propose a randomized controlled experiment with a clear hypothesis, primary and guardrail metrics, and a power analysis to determine sample size and duration. Walk through the statistical tests you'd use, how you'd monitor for novelty effects, and how you'd interpret results to make a launch decision.

Pro tip: Emphasize practical constraints like network effects or seasonality in energy usage, and propose running the test for at least two full weeks to capture weekly patterns. Also mention that you'd pre-register the analysis plan to avoid p-hacking.

1. Define the hypothesis and success metrics

Translate 'improved user engagement' into measurable metrics (e.g., daily active users, session duration, frequency of dashboard visits) and specify a primary metric. Include guardrail metrics like page load time or user satisfaction to ensure no negative impact.

2. Design the experiment

Randomly assign users to control (old dashboard) and treatment (new dashboard) groups, ensuring proper randomization unit (e.g., user-level) and avoiding contamination. Consider stratification by key user segments (e.g., heavy vs. light energy users).

3. Determine sample size and duration

Conduct a power analysis to calculate required sample size based on minimum detectable effect (MDE), significance level (α=0.05), and power (1-β=0.8). Use historical variance to estimate. Duration should cover at least two full weeks to account for weekly seasonality and novelty effects.

4. Analyze results and make a decision

Use appropriate statistical tests (e.g., t-test for continuous metrics, chi-square for proportions) and check for practical significance. Monitor guardrail metrics and segment-level results. If primary metric improves significantly without harming guardrails, recommend launch; otherwise, iterate.

Key Points to Mention

  • Randomization unit and avoiding network effects (e.g., if users can share dashboards)
  • Primary vs. secondary metrics and guardrail metrics
  • Power analysis: MDE, α, β, and sample size calculation
  • Test duration considerations: weekly seasonality, novelty effect, and run-in period
  • Statistical tests: t-test, Mann-Whitney, or bootstrap for non-normal data
  • Practical significance vs. statistical significance and business impact

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