Google Data Analyst Interview Guide

I sat down at my desk fifteen minutes early, staring at my own face on the Google Meet preview screen. My palms were sweaty enough that my mouse was slipping. I had five different tabs open with SQL cheat sheets, STAR method behavioral answers, and Google’s core values. The recruiter had told me it would be three rounds, back-to-back, with short breaks in between.

When the first interviewer logged on, I realized immediately that no amount of memorizing SQL joins was going to save me if I didn’t actually think on my feet.

Round 1: General Cognitive Ability (The "Wait, what?" Round) This wasn't a coding round; it was purely about how my brain worked. The interviewer didn't ask me to write a script. Instead, they asked an open-ended question: "If we wanted to measure the success of a new feature on Google Maps that shows wheelchair accessibility, what metrics would you look at?"

Where I felt confident: I started strong, talking about active users, click-through rates, and session duration. I laid out a neat little framework. Where I started sweating: The interviewer smiled and said, "Okay, but what if the feature is highly successful, but usage drops by 20% in the second month? What data do you look at to figure out why?" My mind went blank for a solid five seconds. I stammered something about seasonal trends before finally catching my breath and suggesting we segment the data by device type or geographical region to isolate the drop-off. I felt like I barely scraped by.

Round 2: Role-Related Knowledge (The Data Sense Round) This was the technical-but-not-quite-technical round. Since it's an apprenticeship, they weren't expecting me to build machine learning models. Instead, they shared their screen, showing a messy, disorganized spreadsheet of hypothetical e-commerce data.

Where I felt confident: The interviewer asked me how I would clean it up to find the top-performing products. Finally, I was in my element. I walked them through filtering out null values, standardizing the date formats, and using basic pivot table logic to aggregate sales by category. I didn't need to write code; I just needed to explain the logic of data hygiene. Where I started sweating: They asked me to explain my process to a hypothetical non-technical stakeholder who thought the data was completely wrong. Trying to explain data cleaning without using jargon like "boolean" or "null values" was much harder than I expected. I stumbled trying to translate data-speak into human-speak.

Round 3: Googliness & Leadership (The Vibe Check) This was standard behavioral stuff: "Tell me about a time you had a conflict in a team," and "Tell me about a time you failed."

What surprised me: I thought this would be the easiest round, but it was actually the most intense. They didn't just accept my first answer. They drilled down. When I gave my rehearsed answer about a project failing, they hit me with, "Looking back, what was the exact moment you realized it was going south, and why didn't you speak up then?" They didn't want the polished, "my biggest weakness is I care too much" BS. They wanted actual vulnerability. I ended up ditching my script and giving a messy, real answer about miscommunicating my bandwidth to a team.

The Real Takeaway The biggest surprise of the entire loop was realizing how little they cared about the syntax I had memorized, and how much they cared about how I handled being stuck. They wanted to see me hit a wall, get frustrated, and then think my way around it.

When I finally closed the laptop, I was completely drained. The LinkedIn version of this story is "Had an amazing and challenging conversation with the brilliant minds at Google!" The real version is that I immediately went to the kitchen, chugged a glass of water, and stared at a wall for twenty minutes wondering if I sounded like an idiot.

Questions asked: Here is the exact breakdown of what was asked, the technical prompts, and the hidden traps. Since this is an Apprenticeship role, the focus wasn't on whether I was already a senior data scientist—it was a test of my baseline logic, attention to detail, and how I approach problems I haven’t seen before.

Here is the brain dump for everyone applying next:

1. The Pre-Interview Assessment (The Take-Home) Before the live rounds, I got an email with a link to a Google Sheet containing a dataset of ~10,000 fictional Google Merchandise Store transactions.

The Prompt: "Analyze this dataset, find three actionable insights to improve sales, and create a maximum 3-slide presentation for a non-technical product manager."

The Hidden Traps (Gotchas):

Dirty Data: The Date column was a mess. Half the rows were formatted MM/DD/YYYY and the other half were DD-MM-YYYY. If you just threw it into a pivot table without standardizing the dates first, your monthly revenue trends would be completely wrong.

The "Refund" Trap: In the Revenue column, some cells didn't have numbers—they had the string "REFUNDED". If you didn't filter those out or convert them to negative values, your sum calculations would break or inflate.

The Lesson: They don’t want complex predictive modeling here. They want to see if you actually look at the raw data before you start analyzing it.

1. The Technical/Logic Round (Live) They shared a plain, unformatted Google Doc (not an IDE, so no syntax highlighting, which is incredibly unnerving). They gave me the choice to write in SQL, Python, or just explain the logic using spreadsheet functions. I chose SQL.

Question 1: The Join Test

Prompt: "We have two tables. Table A is Users (User_ID, Sign_up_date, Country). Table B is Watch_History (User_ID, Video_ID, Watch_Minutes, Date). Write a query to find the top 5 countries by total watch minutes in the last 7 days."

What they were looking for: Can you do a basic JOIN? Do you remember to use GROUP BY Country? Do you know how to sort using ORDER BY ... DESC LIMIT 5?

The Follow-up: "What if a user signed up but has zero watch history? How does your query handle them?" (They were testing if I understood the difference between an INNER JOIN and a LEFT JOIN).

Question 2: The Logic Puzzle

Prompt: "You have a list of 100,000 search queries. How would you determine which queries are misspellings of 'YouTube'?"

My approach: I started talking about Levenshtein distance (edit distance). The interviewer stopped me and said, "Keep it simpler. No advanced algorithms. Just logic." I pivoted and said I would look at the bounce rate or the "next search" term—if users search "Yutube" and then immediately search "YouTube", the first was a typo. The interviewer visibly liked that answer. They want practical business logic, not just math.

1. The Case Prompt (General Cognitive Ability) No coding here, just a verbal business case.

The Prompt: "Imagine you are looking at a dashboard for Google Maps. You notice that the number of people searching for 'coffee shops' drops by 15% on a Wednesday compared to the previous Wednesday. Walk me through exactly what you investigate to find out why."

The Framework: You have to be structured here. Do not just guess. I broke it down into:

Internal Data/Bugs: Is the data pipeline broken? Did an app update roll out on Tuesday that broke the search bar?

External/Environmental: Was there a major weather event (like a hurricane) on the East Coast keeping people inside? Was last Wednesday a public holiday?

Segmentation: I told them I would slice the 15% drop by Geography (is it global or just one country?), Device (is it only happening on iOS?), and User Type.

The Curveball: The interviewer asked, "What if you check all of that, and everything is normal. What else could it be?" (I suggested checking if a competitor launched a massive promo, or if a backend server for a specific region was timing out).

1. The Behavioral/Googliness Round They use the STAR method (Situation, Task, Action, Result), but they are aggressive about digging into the "Action" part.

Question 1: "Tell me about a time you had to analyze a piece of information and realized your initial assumption was completely wrong. What did you do?"

Tip: Pick a story where you actually messed up and caught it, not a fake humble-brag.

Question 2: "Tell me about a time you had to explain a complex concept to someone who had zero background in it."

Tip: This is a core skill for an analyst. I talked about explaining a spreadsheet model to a family member running a small business.

Small, Weird Details to Remember Silence is okay: During the SQL question, I said, "I need 30 seconds to structure this in my head." The interviewer muted their mic, smiled, and waited. You don't have to fill the dead air with panic talking.

Calculators: I was allowed to use a piece of scratch paper and a pen. Have one on your desk.

No Tricks: None of the interviewers were trying to trick me. When I got stuck on the SQL date function, the interviewer literally said, "Don't worry about the exact syntax, just write [DATE FUNCTION] and explain what it should do." They care about your brain, not your ability to act like a compiler.