Expect analytics-and-experimentation questions (a feature raises DAU but cuts session length, so is it net positive, and how would you decide), statistical foundations (bias-variance diagnosis from learning curves, deriving least-squares from Gaussian-MLE), and applied ML reasoning, alongside SQL. Interviewers reward rigorous experiment design, guardrail-metric thinking, and clean derivations you can narrate. Whether you target Product Analytics or a research-leaning role, the bar is connecting statistics to real product decisions. Prepare A/B testing, metric trade-offs, core statistics, and ML fundamentals, and practise reasoning aloud.
Recruitment process
- Recruiter screen and role alignment
- Technical phone screen on statistics, SQL, and analytics
- Onsite loop: analytics case, statistics, ML fundamentals, and behavioural
- Team match and offer
Interview rounds
- Round 1 (45-60 min): statistics and SQL screen
- Round 2 (45-60 min): product-analytics and experiment-design case
- Round 3 (45-60 min): ML fundamentals and statistical-reasoning round
- Round 4 (45 min): behavioural and Googleyness round
Product sense and rigour framework
- Pair rigorous experiment design and guardrail-metric thinking with clean statistical derivations and ML fundamentals, always tied to a real product decision
Practice questions
- 01Practise
Given a stream of integers, design a data structure that returns the median in O(log n).
Intense - 02Practise
You have a service that's slow only on Tuesdays. Walk me through how you'd debug.
StandardGoogleyness: structured debugging, hypothesis-driven, instrumentation-first. - 03Practise
You're the Staff Engineer rolling out Cursor/Copilot to a team of 25 with mixed seniority. What's your 90-day plan and how do you measure success?
Intense2026-defining tech-leadership question. Wants: phased rollout (junior-pair → senior-pair → solo), guardrails (security review boundaries, never-trust patterns), measurement (PR-review-time, defect-escape rate, NOT lines-of-code). - 04Practise
You launch a feature and DAU rises but session length falls. Design an analysis to decide whether the feature is net positive. What metrics and tests do you use?
IntenseGoogle data-scientist rounds test product-sense analytics: metric trade-offs, guardrail metrics, and rigorous experiment design over rote statistics. - 05Practise
Explain the bias-variance trade-off and how you'd diagnose whether a model is underfitting or overfitting from its learning curves.
StandardGoogle tests ML fundamentals with practical diagnosis. Connect bias-variance to learning curves, regularisation, and data-size decisions. - 06Practise
Derive why maximum-likelihood estimation for linear regression with Gaussian noise gives the least-squares objective.
IntenseGoogle data-scientist rounds probe statistical foundations. A clean MLE-to-least-squares derivation signals genuine depth.
