← Waymo Interview Insights

Waymo·Data Scientist·Technical Phone Screen·Senior

SeniorPrefer not to say
Apr 2026Remote

Summary

Waymo data scientist interview that leaned way harder into probability theory than I expected. The whole session was basically one extended stats problem with a simulation layer bolted on, which was fine until the confidence interval follow-up made me realize I hadn't thought carefully enough about the variance.

Questions Asked (3)

Q1

In a best-of-5 table tennis match where player A wins each game with fixed probability p, derive the probability that the match goes to exactly 5 games.

A/B Testing & ExperimentationAlgorithms & Data Structures
Author's notes

This took me longer to set up than it should have.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

First, clarify that the match ends when one player wins 3 games, so going to 5 games means the score is 2-2 after 4 games. Then compute the probability of reaching 2-2 and multiply by the probability that the 5th game is played (which is 1, since the match must end).

Pro tip: Emphasize that the 5th game's outcome doesn't affect the probability of the match going to 5 games; it's solely determined by the first 4 games. Also, mention that this is a negative binomial or binomial scenario, showing you can connect probability to real-world experimentation.

1. Define the event

The match goes to exactly 5 games if and only if after 4 games, both players have won exactly 2 games each.

2. Identify the distribution

The number of games player A wins in the first 4 games follows a Binomial(4, p) distribution, assuming independence.

3. Compute the probability of 2-2 after 4 games

Calculate P(A wins exactly 2 games) = C(4,2) * p^2 * (1-p)^2 = 6 p^2 (1-p)^2.

4. Confirm the match goes to 5 games

Since the match cannot end before 5 games if the score is 2-2, the probability is exactly the probability of a 2-2 score after 4 games.

5. State the final answer

The probability is 6 p^2 (1-p)^2. Optionally, discuss edge cases (p=0 or p=1) and symmetry.

Key Points to Mention

  • Independence of games and fixed probability p
  • Binomial coefficient C(4,2) = 6
  • The match must end when a player reaches 3 wins, so 5 games implies 2-2 after 4
  • The 5th game's outcome is irrelevant to the probability of reaching 5 games
  • Edge cases: p=0 or p=1 yield probability 0
  • Connection to A/B testing: modeling binary outcomes and calculating probabilities of sequences

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

Q2

How would you write a Monte Carlo simulation to estimate the probability that a best-of-5 match lasts exactly 5 games?

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

Pretty straightforward to describe.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Clarify the assumptions (e.g., constant per-game win probability, independence) and then outline a simulation that repeatedly plays best-of-5 series, recording the number of games. Finally, estimate the probability as the proportion of series that last exactly 5 games, and discuss how to choose the number of simulations for a desired precision.

Pro tip: Mention that you can compute the exact probability analytically (e.g., 2 * p^2 * (1-p)^2 * (something) or via negative binomial) and use that to validate your simulation—this shows you understand the problem deeply and can sanity-check results.

1. Clarify assumptions and parameters

State that each game is independent and has a constant probability p of the stronger team winning (or assume p=0.5 if not specified). Define what 'lasts exactly 5 games' means: the series is tied 2-2 after 4 games, and the 5th game decides the winner.

2. Design the simulation loop

For each simulated series, simulate games until one team reaches 3 wins. Count the number of games played. Repeat this for N independent series.

3. Compute the estimate and uncertainty

Calculate the proportion of series that lasted exactly 5 games. Also compute a confidence interval (e.g., using the normal approximation) to quantify the Monte Carlo error.

4. Validate and interpret

Compare the simulation result to the analytical probability (if p is known) to ensure correctness. Discuss how the probability depends on p and the implications for the business context.

Key Points to Mention

  • Independence of games and constant win probability p (or the need to estimate p from data).
  • The condition for a 5-game series: the first 4 games must be split 2-2, and then the 5th game is played.
  • Analytical formula: P(5 games) = 6 p^2 (1-p)^2. For p=0.5, that's 0.375. This is because the series lasts exactly 5 games if and only if the first 4 games are split 2-2, which has probability C(4,2) p^2 (1-p)^2 = 6 p^2 (1-p)^2.

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

Q3

Given M simulated matches, how would you construct a confidence interval for the estimated probability?

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

This is where I got a bit shaky.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying that the matches are independent Bernoulli trials with a constant probability of success, then present the standard Wald interval using the sample proportion and its standard error. Also discuss the limitations of the Wald interval for small M or extreme probabilities, and mention more robust alternatives like the Wilson score interval or Clopper-Pearson exact interval.

Pro tip: In practice, especially for A/B testing at scale, the Wilson score interval is preferred because it maintains good coverage even for small sample sizes or probabilities near 0 or 1, and it's easy to compute.

1. Clarify assumptions

Confirm that the M matches are independent and identically distributed Bernoulli trials with a constant probability of success p. State that the estimate is the sample proportion p_hat = X/M, where X is the number of successes.

2. Choose a confidence interval method

Select an appropriate method based on M and p_hat. For large M and p_hat not near 0 or 1, the Wald interval is simple. For small M or extreme p_hat, use Wilson score or Clopper-Pearson exact interval.

3. Compute the interval

For the Wald interval, calculate p_hat ± z_{α/2} * sqrt(p_hat*(1-p_hat)/M). For Wilson, use the formula that adjusts for continuity and skewness. For Clopper-Pearson, use the beta distribution quantiles.

4. Interpret and communicate

Explain that the interval provides a range of plausible values for the true probability, with the chosen confidence level. Discuss any caveats about coverage probability and the impact of M.

Key Points to Mention

  • Independence and identical distribution assumption for matches
  • Sample proportion as the point estimate
  • Wald interval formula and its limitations (poor coverage for small M or extreme p)
  • Wilson score interval as a robust alternative
  • Clopper-Pearson exact interval for conservative coverage
  • Effect of sample size M on interval width and precision

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