← Google Interview Insights

Google·Software Engineer·Technical Phone Screen·Senior

Senior
May 2026

Summary

Google data science interview with a PM asking a statistics/distribution question about search behavior. Pretty short interaction from what I remember, but the question had more depth to it than I expected.

Questions Asked (1)

Q1

How would you describe the distribution of daily search queries per user?

Product Analytics & MetricsData Modeling
Author's notes

I started talking about a right-skewed distribution pretty quickly, which felt right, but then I kind of stumbled when thinking about how to actually characterize it precisely.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying that the distribution is typically heavy-tailed and right-skewed, with most users making few queries and a small fraction making many. Then discuss how you would model it (e.g., log-normal or power law) and validate with data, noting that the exact shape depends on user segmentation and time period.

Pro tip: Acknowledge that the distribution is not purely power law but often log-normal in practice, and mention that outliers (power users) can dominate aggregate metrics, so median and percentiles are more informative than the mean.

1. Clarify the metric and scope

Define what 'daily search queries per user' means: is it per active user or all users? Over what time window? This sets the context for the distribution.

2. Describe the expected shape

Explain that the distribution is right-skewed and heavy-tailed: most users have low query counts (e.g., 0-5), while a small minority have very high counts (100+).

3. Propose a statistical model

Suggest that a log-normal or power-law distribution often fits such data, and mention that you would test goodness-of-fit to determine the best model.

4. Discuss implications and metrics

Highlight that the mean is pulled up by outliers, so median and percentiles (e.g., 90th, 99th) better represent typical usage. This affects product decisions and infrastructure planning.

5. Consider segmentation and validation

Note that the distribution may vary by user segment (e.g., new vs. power users) and that you would validate assumptions with real data, checking for multimodality or changes over time.

Key Points to Mention

  • Right-skewed, heavy-tailed distribution
  • Log-normal or power-law model
  • Mean vs. median and percentiles
  • Outliers and power users
  • Segmentation by user type or demographics
  • Data validation and goodness-of-fit testing

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