← Amazon Interview Insights

Amazon·Data Scientist·Technical Phone Screen·Senior

Senior
Jun 2026

Summary

Amazon DS interview focused almost entirely on A/B testing, and it went deep fast. Five sub-questions all hanging off one scenario about a signup funnel. Walked out not totally sure how I did.

Questions Asked (5)

Q1

Given a baseline conversion rate of 4.0% and a minimum detectable effect of 0.4 percentage points, what sample size per arm do you need for a two-proportion z-test at 80% power and 5% significance? Show your formula and the numeric result.

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

I knew the formula well enough but fumbled the plugging-in part under pressure.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clearly stating the formula for sample size per arm in a two-proportion z-test, then plug in the given values: baseline conversion rate (p1 = 0.04), minimum detectable effect (0.4 percentage points, so p2 = 0.044), significance level (α = 0.05, two-sided), and power (1-β = 0.80). Compute the required sample size using the standard formula and round up to the nearest integer.

Pro tip: Mention that this calculation assumes no continuity correction and equal allocation; in practice, you might adjust for expected attrition or use sequential testing, but for a standard A/B test this is the right starting point.

1. Define parameters

Identify p1 = 0.04, p2 = 0.044, α = 0.05 (two-sided), power = 0.80, and the corresponding z-values: z_{α/2} = 1.96 and z_β = 0.84.

2. State the formula

Use the standard formula for sample size per arm: n = ( (z_{α/2} * sqrt(2 * p̄ * (1-p̄)) + z_β * sqrt(p1*(1-p1) + p2*(1-p2)) )^2 ) / (p2 - p1)^2, where p̄ = (p1 + p2)/2.

3. Compute intermediate values

Calculate p̄ = 0.042, then compute the terms: sqrt(2 * 0.042 * 0.958) ≈ 0.283, sqrt(0.04*0.96 + 0.044*0.956) ≈ 0.289, and the difference (p2-p1) = 0.004.

4. Plug in and solve

Substitute into the formula: n = ( (1.96 * 0.283 + 0.84 * 0.289)^2 ) / (0.004)^2. Compute numerator: (0.5547 + 0.2428)^2 = (0.7975)^2 ≈ 0.636, then divide by 0.000016 to get n ≈ 39,750 per arm.

5. Round and interpret

Round up to the next whole number: approximately 39,751 per arm. Mention that this is the required sample size to detect a 0.4 percentage point absolute lift with 80% power at 5% significance.

Key Points to Mention

  • Two-proportion z-test formula for sample size calculation
  • Baseline conversion rate (p1) and minimum detectable effect (absolute difference)
  • Significance level (α) and power (1-β) and their corresponding z-scores
  • Assumption of equal allocation and no continuity correction
  • Pooled variance under the null hypothesis for the first term
  • Rounding up to ensure sufficient sample size

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

Q2

You observe control at 4.30% (n=50,000) and treatment at 4.22% (n=49,500). Compute the p-value using both pooled and unpooled variance, build a 95% confidence interval for the difference, and explain what the result means practically.

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

This is where I got a bit tangled.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by computing the pooled and unpooled standard errors for the difference in proportions, then calculate the z-statistics and corresponding p-values. Next, construct the 95% confidence interval using the unpooled standard error. Finally, interpret the results in practical terms, emphasizing the small effect size and business implications.

Pro tip: At Amazon, always tie statistical significance to practical significance—mention that with large samples, tiny differences can be significant but may not justify action. Also, note that unpooled variance is preferred for confidence intervals, while pooled is for hypothesis testing under the null.

1. Compute sample proportions and difference

Calculate control conversion rate (p_c = 0.0430) and treatment conversion rate (p_t = 0.0422). The observed difference is p_t - p_c = -0.0008 (or 0.08 percentage points).

2. Calculate pooled and unpooled standard errors

For pooled: p_pool = (x_c + x_t) / (n_c + n_t) = (2150 + 2089) / 99500 ≈ 0.0426. SE_pooled = sqrt(p_pool * (1-p_pool) * (1/n_c + 1/n_t)). For unpooled: SE_unpooled = sqrt(p_c*(1-p_c)/n_c + p_t*(1-p_t)/n_t).

3. Compute z-statistics and p-values

For pooled: z = (p_t - p_c) / SE_pooled. For unpooled: z = (p_t - p_c) / SE_unpooled. Then find two-tailed p-values from standard normal distribution.

4. Build 95% confidence interval

Use unpooled SE: CI = (p_t - p_c) ± 1.96 * SE_unpooled. This gives a range for the true difference in proportions.

5. Interpret results practically

Discuss whether the difference is statistically significant (p < 0.05) and the width of the CI. Emphasize that the effect size is tiny (0.08 pp) and likely not business-relevant, even if significant.

Key Points to Mention

  • Pooled variance assumes equal variances under the null hypothesis, while unpooled does not; use pooled for hypothesis testing and unpooled for confidence intervals.
  • With large sample sizes, even tiny differences can be statistically significant; always assess practical significance.
  • The 95% confidence interval provides a range of plausible values for the true difference; if it includes zero, the difference is not statistically significant at 5% level.
  • Compute the actual numbers: pooled SE ≈ 0.00129, unpooled SE ≈ 0.00129 (similar due to large n), z ≈ -0.62, p ≈ 0.54 (not significant). CI ≈ (-0.0033, 0.0017) includes zero.
  • Consider the business context: a 0.08 percentage point difference in conversion rate may not justify implementation costs.
  • Mention that for Amazon, decisions should be driven by both statistical and practical significance, and possibly by segment analysis or guardrail metrics.

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

Q3

How would you define primary versus guardrail metrics for this experiment, and what multiple comparison correction would you apply if you're also tracking cancellation rate and latency as guardrails?

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

Felt more comfortable here.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by defining the primary metric as the key success measure tied to the experiment's hypothesis, and guardrail metrics as those that ensure no harm to user experience or business health. Then discuss multiple comparison correction, emphasizing that guardrails are typically monitored for directional harm rather than statistical significance, so corrections like Bonferroni or Benjamini-Hochberg may be applied but with caution to avoid over-penalizing. Finally, recommend a practical approach such as using a lower alpha for guardrails or focusing on effect sizes and confidence intervals.

Pro tip: At Amazon, guardrails are often treated as one-sided tests to detect harm, and multiple comparison corrections are applied selectively—over-correcting can mask real issues. Emphasize that guardrails are about risk management, not proving superiority.

1. Define primary metric

Identify the single metric that directly measures the experiment's goal (e.g., conversion rate, click-through rate) and is used to decide success.

2. Define guardrail metrics

Select metrics that monitor potential negative side effects (e.g., cancellation rate, latency) and ensure they do not degrade significantly.

3. Determine correction strategy

Decide whether to apply multiple comparison correction. For guardrails, consider one-sided tests and less stringent corrections to avoid missing harm.

4. Choose correction method

If correction is needed, use Bonferroni for strict family-wise error control or Benjamini-Hochberg for false discovery rate control, balancing power and risk.

5. Communicate trade-offs

Explain the rationale behind the chosen correction and how it aligns with the experiment's risk tolerance and decision-making framework.

Key Points to Mention

  • Primary metric should be directly tied to the experiment hypothesis and business objective.
  • Guardrail metrics are used to detect unintended negative consequences and are often one-sided tests.
  • Multiple comparison correction is necessary when testing multiple metrics to control false positives.
  • Bonferroni correction is conservative and may reduce power; Benjamini-Hochberg controls FDR and is less stringent.
  • For guardrails, consider using a lower alpha (e.g., 0.01) or focusing on confidence intervals rather than p-values.
  • Amazon emphasizes practical significance and effect sizes over strict statistical significance for guardrails.

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

Q4

Walk me through how CUPED or pre-experiment stratification reduces variance in an A/B test, and what effect that has on required sample size.

A/B Testing & ExperimentationTechnical Trade-offs
Author's notes

Said CUPED regresses out pre-experiment covariate variance from the outcome, which shrinks the residual variance and effectively lets you reach the same power with fewer users.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by explaining the core idea of using pre-experiment data to explain post-experiment variance, then detail how CUPED and stratification implement this via regression adjustment or blocking. Finally, connect the variance reduction to the sample size formula, emphasizing the squared relationship and practical implications for experiment speed and cost.

Pro tip: Quantify the impact: mention that a 50% variance reduction cuts required sample size in half, and always check that pre-experiment covariates are balanced and truly predictive to avoid bias.

1. Explain the variance problem in A/B tests

Describe how high variance in the outcome metric increases required sample size and reduces sensitivity. Mention that pre-experiment differences or noise can obscure true treatment effects.

2. Introduce CUPED and pre-experiment stratification

Define CUPED as using pre-experiment covariates (e.g., user's prior behavior) to adjust post-experiment outcomes via regression, and stratification as blocking on pre-experiment variables to create homogeneous groups.

3. Detail the variance reduction mechanism

For CUPED, explain that the adjusted estimator subtracts the portion of variance explained by the covariate, reducing variance by a factor of (1 - ρ²). For stratification, explain that within-stratum variance is lower, and the overall variance is a weighted average of stratum variances.

4. Connect variance reduction to sample size

Show that required sample size is proportional to variance (n ∝ σ²/Δ²). Therefore, reducing variance by a factor k reduces sample size by the same factor, assuming power and effect size are fixed.

5. Discuss practical implications and trade-offs

Highlight that CUPED can achieve large variance reductions (e.g., 50%) if covariates are predictive, leading to faster experiments. Mention that stratification requires careful implementation and may have limited gains if strata are not predictive.

Key Points to Mention

  • CUPED uses pre-experiment covariates to adjust post-experiment outcomes, reducing variance by (1 - ρ²).
  • Pre-experiment stratification divides users into homogeneous blocks based on pre-experiment variables, reducing within-group variance.
  • Required sample size is directly proportional to variance: n ∝ σ²/Δ², so variance reduction directly reduces sample size.
  • A 50% variance reduction halves the required sample size, significantly speeding up experiments.
  • CUPED is easy to implement and often yields substantial variance reduction without bias if covariates are pre-treatment.
  • Stratification can be combined with CUPED for even greater variance reduction, but requires careful design to avoid imbalance.

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

Q5

What are the risks of peeking at results before an experiment ends, and how do you control type-I error inflation? Also, what do you do when a global average looks flat but segment-level results show opposite directions?

A/B Testing & ExperimentationAdaptability & Ambiguity
Author's notes

Two-parter and I think I handled the first better than the second.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by explaining the statistical risks of peeking (inflated type-I error) and methods to control it, such as sequential testing or alpha spending. Then, address the segment-level paradox by discussing heterogeneity, multiple comparisons, and the need for pre-registered subgroups or post-hoc validation. Emphasize a structured, data-driven approach that balances statistical rigor with business impact.

Pro tip: Mention that at Amazon, decisions are often made with incomplete data, so it's crucial to quantify the risk of false positives and use techniques like sequential testing to allow early stopping without inflating error rates. Also, highlight that segment-level insights should be validated with holdout data or a follow-up experiment to avoid overfitting to noise.

1. Explain the risks of peeking

Describe how repeatedly checking results during an experiment increases the chance of false positives (type-I error) because each look is a new hypothesis test. This can lead to premature and incorrect decisions.

2. Control type-I error inflation

Discuss methods like sequential testing (e.g., group sequential designs, alpha spending functions), Bonferroni correction, or Bayesian approaches to maintain the overall false positive rate at the desired level.

3. Address flat global average with opposing segments

Acknowledge that this is Simpson's paradox: the overall effect can mask heterogeneous treatment effects. Investigate segment-level results to understand the drivers, but be cautious of multiple comparisons.

4. Validate segment findings

Use pre-registered subgroups or apply corrections for multiple testing (e.g., Benjamini-Hochberg). If segments are post-hoc, validate with a follow-up experiment or holdout data to confirm the effect.

5. Make a decision balancing statistical and business significance

Consider the practical implications: if segments show opposite effects, decide whether to target specific segments or iterate on the experiment. Communicate uncertainty and recommend next steps.

Key Points to Mention

  • Type-I error inflation and the multiple comparisons problem
  • Sequential testing methods (e.g., alpha spending, group sequential designs)
  • Simpson's paradox and heterogeneous treatment effects
  • Pre-registration of subgroups and corrections for multiple testing (e.g., Bonferroni, Benjamini-Hochberg)
  • Bayesian approaches as an alternative to frequentist sequential testing
  • The importance of validating segment-level findings with follow-up experiments

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