← Google Interview Insights

Google·Data Scientist·Technical Phone Screen·Senior

Senior
Jun 2026

Summary

Google DS interview with a meaty causal inference question about a shipped product change. No A/B test existed, which made it genuinely hard. The kind of question where you realize mid-answer how many layers there actually are.

Questions Asked (1)

Q1

Engineering shipped a new version of a product meant to reduce disconnections, but there was no holdout group. How would you rigorously evaluate whether it actually worked? Walk through your choice of method, identification assumptions, how you'd handle confounders, and how you'd communicate uncertainty to stakeholders.

A/B Testing & ExperimentationProduct Analytics & MetricsTechnical Trade-offs
Author's notes

This one wrecked me a little.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Acknowledge the lack of a holdout and propose a quasi-experimental design, such as difference-in-differences or synthetic control, to estimate the causal effect. Emphasize the importance of validating assumptions, quantifying uncertainty, and communicating limitations transparently to stakeholders.

Pro tip: Frame the analysis as a retrospective causal inference study, not an A/B test, and proactively suggest a follow-up randomized experiment to confirm findings. This shows you understand both the limitations of observational data and the value of experimentation.

1. Clarify the goal and data

Define the target metric (e.g., disconnection rate) and identify available pre- and post-launch data, including potential control groups (e.g., unaffected regions, similar products, or users who haven't upgraded).

2. Choose a quasi-experimental method

Select an appropriate causal inference technique like difference-in-differences, synthetic control, or interrupted time series, based on data structure and plausibility of assumptions.

3. State and test identification assumptions

Articulate key assumptions (e.g., parallel trends, no spillovers) and perform diagnostics (e.g., pre-trend tests, placebo tests) to assess their validity.

4. Address confounders and robustness

Control for observed confounders via matching or regression, and conduct sensitivity analyses (e.g., different control groups, model specifications) to check robustness.

5. Quantify and communicate uncertainty

Report effect estimates with confidence intervals, discuss limitations, and use clear visualizations to convey uncertainty to stakeholders, recommending next steps like a randomized experiment.

Key Points to Mention

  • Difference-in-differences or synthetic control as alternatives to A/B testing
  • Parallel trends assumption and how to test it (e.g., pre-period trends)
  • Handling confounders through matching, regression adjustment, or propensity scores
  • Sensitivity analysis to assess robustness of results
  • Communicating uncertainty with confidence intervals and caveats
  • Recommending a follow-up randomized experiment to validate findings

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