← Meta Interview Insights

Meta·Data Scientist·Technical Phone Screen·Intermediate

Intermediate
May 2026

Summary

Meta data scientist interview with a probability and stats question centered on ad impression modeling. Pretty focused, just the one problem but they pushed on the math.

Questions Asked (1)

Q1

You have Y ad impressions distributed randomly across X users. What is the expected number of impressions per user, and what is the probability that any given user sees at least one impression?

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

The expected value part is straightforward, just Y divided by X.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the assumptions: each impression is independently assigned to a uniformly random user, so the number of impressions per user follows a Binomial distribution. Then compute the expected value and the probability of at least one impression using the complement rule, and discuss practical implications for ad delivery and experimentation.

Pro tip: Mention that in real systems, impressions are not perfectly independent due to user targeting and frequency capping, so the Binomial model is a simplification. Acknowledging this shows you understand the gap between theory and production.

1. Clarify assumptions

Confirm that each impression is independently and uniformly assigned to one of X users, and that Y is fixed. This leads to a Binomial model for each user's impression count.

2. Define random variable

Let I_i be the number of impressions user i receives. Then I_i ~ Binomial(Y, 1/X).

3. Compute expected impressions per user

By linearity of expectation, E[I_i] = Y * (1/X) = Y/X. This is the average number of impressions per user.

4. Compute probability of at least one impression

Use the complement: P(I_i >= 1) = 1 - P(I_i = 0) = 1 - (1 - 1/X)^Y. For large X and Y, approximate with 1 - e^{-Y/X}.

5. Discuss implications and extensions

Relate to ad delivery: if Y/X is small, many users see zero ads; if large, users see multiple ads. Mention that in practice, targeting and frequency capping alter these probabilities.

Key Points to Mention

  • Binomial distribution for impressions per user
  • Linearity of expectation for the mean
  • Complement rule for probability of at least one
  • Poisson approximation when X is large and Y/X is moderate
  • Independence assumption and its limitations in real ad systems
  • Practical relevance: ad reach, frequency, and experiment design

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