Google Data Analyst Interview Questions (What to Expect)
Google data analyst loops (including business/product analyst variants) typically include a recruiter screen, one or two technical rounds heavy on SQL and statistics, a product/metrics case round, and behavioral interviews aligned to Google's hiring attributes. Expect structured, rubric-scored interviews where showing your reasoning matters as much as the answer.
About Google
- Google runs structured, rubric-based interviews and typically evaluates against defined attributes: general cognitive ability, role-related knowledge, leadership, and 'Googleyness'.
- Analyst roles exist across many surfaces (Search, YouTube, Ads, Cloud, internal operations), with SQL and experimentation as common foundations.
- Hiring decisions are made by committee reviewing written feedback from all interviews, not by the hiring manager alone.
- A/B experimentation is deeply embedded in product decisions, so statistical judgment questions are near-universal in analyst loops.
- Interviewers commonly expect candidates to think aloud — the reasoning path is scored, not just the final answer.
Interview questions to expect
Given tables of users and events, write a query for the daily active user count over the last 30 days, and then week-over-week growth.
Fluency with aggregation, date functions, CTEs, and window functions (LAG for growth) is assumed. Narrate as you write, state assumptions about the schema out loud, and sanity-check edge cases like missing days — the narration is scored.
Find the top three products per category by revenue.
The classic ROW_NUMBER/RANK-partition pattern. Strong candidates mention the tie-handling difference between RANK and ROW_NUMBER unprompted — it signals real query mileage, not memorized patterns.
Your A/B test shows a 2% lift with p = 0.04. The PM wants to ship. What do you tell them?
Show statistical judgment beyond the threshold: practical vs statistical significance, test duration and novelty effects, multiple-comparison risk if many metrics were watched, and segment consistency. The job is being the honest broker of evidence.
Explain p-values and confidence intervals to a non-technical stakeholder.
A communication test disguised as a stats test. Have a rehearsed plain-language version that survives scrutiny (what the p-value is NOT is part of it), ideally with a concrete example rather than definitions.
How would you measure the success of a feature like YouTube's 'watch later'?
Structure wins: define the feature's job, propose adoption, engagement, and retention-impact metrics, and name guardrails (does saving cannibalize immediate watching?). End with the one metric you'd champion and why.
Search queries per user dropped 3% after a redesign. Investigate.
Decompose methodically: data/logging changes first (redesigns notoriously break instrumentation), then segment by platform, browser, geography, and user tenure, then consider whether fewer queries might mean better results — metric direction isn't automatically bad. That last point often separates candidates.
Estimate the number of ride-share trips in your city per day (or a similar sizing question).
Estimation still appears in analyst loops. Show a clean decomposition (population → addressable users → frequency), round numbers aggressively, sanity-check the result against a known anchor, and be explicit about your weakest assumption.
You find a data quality issue that invalidates a report leadership already saw. What do you do?
Integrity question — the answer is fast, direct correction with impact assessment, plus the pipeline fix so it can't recur. Interviewers listen for whether you'd quietly patch it forward; that's the wrong answer at any company.
How do you decide between a dashboard, a one-off analysis, or saying no to a data request?
Analyst time is the scarce resource. Strong answers triage by decision value: what decision does this inform, how often, and by whom? Recurring decisions get dashboards, one-time decisions get analyses, and requests without a decision attached get a conversation.
Tell me about a time your analysis changed a decision.
The core impact story — pick one where the decision-makers initially believed something else. Show the analysis, how you communicated it (the artifact, the meeting), and the concrete decision delta. Analysis that didn't change anything isn't impact.
Tell me about a time you were wrong in an analysis and someone caught it.
Intellectual honesty check. The strong version: you thanked them, quantified the damage, corrected publicly, and changed a checking habit. Nobody believes the candidate whose analyses were never wrong.
A stakeholder asks 'why is engagement down?' with no further context. Walk me through your approach.
Googleyness-adjacent: dealing with ambiguity. Show you'd first sharpen the question (which metric, which surface, since when), then run the standard decomposition, and time-box the investigation with a check-in rather than disappearing for two weeks.
What's your approach to making an analysis reproducible and trustworthy?
Craft signal: versioned queries/notebooks, documented assumptions, peer review for load-bearing numbers, and consistent metric definitions against a source of truth. Mentioning the pain of two dashboards disagreeing shows you've lived it.
Present an insight from a past project as if we were the leadership team.
Some loops include a mini-presentation. Lead with the takeaway in one sentence, support with two or three exhibits max, pre-empt the obvious objection, and end with the recommended decision. Analysts who bury the lede in methodology lose the room.
Why Google, and what team or product area do you want to work on?
Generic Google admiration wastes the answer. Name a product area, the kind of decisions its analysts likely inform, and connect it to work you've done. Specificity signals you'll hit the ground with direction.
How to prepare
- Drill SQL until window functions, CTEs, and date logic are automatic — then practice narrating queries aloud, because silent-correct scores worse than narrated-correct.
- Prepare plain-language explanations of p-values, confidence intervals, and power; the explain-to-a-stakeholder question is near-certain.
- Practice the metric-investigation pattern (instrumentation → segmentation → external factors → interpretation) on three different fictional drops.
- Prepare 5-6 behavioral stories mapped to Google's attributes, each with a measurable outcome.
- For every case answer, state your structure first in one sentence — structured thinking is explicitly what rubrics reward.
Practice these questions out loud
Paste the actual job post into Preppara: get a study plan, drill the exact topics, and run a voice mock interview. Your first mock is free.
Try Preppara freeInterview coming up? The Interview Sprint gives you 30 days of everything — one payment, no subscription.