Google Senior+ Data Scientist Interview Experience — Cold-Called Into the DSR Track, All Strong Hire

I'd read a small handful of interview reports, and even those helped a lot. A bit of background first: I'm at a big company, I've been coasting for a few years, and it started to feel like I was wasting my time, so at the start of this year I decided to look for a new job. It wasn't smooth — I got rejected by hiring managers, rejected on coding or SQL, rejected at onsite, rejected at the final round, and eventually I landed an offer at Google. So if you're as lost as I was when I started job hunting, this might help a little.

I wasn't short on interviews, but the DS job market overall is rough right now — way more candidates than openings — so nothing about this process was smooth. Here's roughly how the Google timeline and process went:

• End of August: submitted my resume
• Early October: HR called me directly out of nowhere — I was literally at work when it happened. Two teams had picked out my resume, and after some internal discussion they had me interview for DSR. I hadn't done enough homework going in — I'd just applied to a handful of postings without really knowing that Google DS internally splits into different tracks.
• Late October: phone screen, two rounds. Heard back a week later that I'd passed.
• Mid-November: virtual onsite (VO), three rounds.
• Two days after the onsite finished, I got word that I'd passed — multiple teams were interested.
• End of November: team match.

I also interviewed at Google five years ago and got knocked out at the onsite that time. On the interview content itself — maybe it's because of how many years of experience I have now, but I wasn't asked any of the basic probability/distribution derivation stuff anymore. It was all business use-case application. The closest thing to a "textbook" stats question was something like how to AB test and reduce variance; everything else was applied performance-measurement questions — experiment design (e.g. how would you test latency's impact on metric X), causal inference (can you explain how you'd use propensity score matching in this scenario?), and on the modeling side, mixed models, random/fixed effects, and so on. On the product side it was things like how to diagnose a ratio metric change, defining a metric for the downstream impact of YouTube latency, and other pretty standard analytics questions. The coding was honestly an insult to my intelligence — way too easy, to the point that I had 15 minutes left with nothing to do and just had to make small talk. It covered things like simulating a uniform distribution, aggregating a table for some purpose, and some Fibonacci sequence manipulation.

My overall take is that the questions were pretty standard, and the difficulty felt a lot lower than the "count the colored balls" probability problems I got asked years ago. I can't go into more specifics on the actual questions, and since I don't grind old interview reports and couldn't see most of them anyway, here's my slightly unconventional prep advice: read books, and ask ChatGPT whenever you're stuck. Everyone probably has their own way of prepping for coding, and DS coding is basically just SQL + Python — generally not that hard, so I won't get into it here.

All told I read at least four or five books for this. Obviously cramming doesn't work, so this wasn't some last-minute strategy — I read these in my free time because they were useful for work, and it turned out they were extremely useful for interviews too. I'd recommend them to anyone; at minimum they'll get you through the DS track just fine.

I'd strongly recommend "Statistical Inference" by George Casella and Roger L. Berger — it goes from the basics all the way up to advanced material and lays out the foundations and reasoning behind all of statistics really clearly. There were parts of this book I didn't fully get back in school, and going back to it now, a lot more of it clicked. It's also genuinely useful for reading papers. A lot of the concepts it covers — bootstrap, delta method — come up constantly in interviews. For example, I got asked "can bootstrap help reduce variance," and the book has a detailed example and comparison for exactly that.

Causal inference: I've personally read a lot of books and papers on this, but if you're just prepping for interviews, Scott Cunningham's "Causal Inference: The Mixtape" is enough — tons of examples, and it walks through a lot of great papers. If you're actually into causal inference as a research area, the Stanford lecture series "Machine Learning & Causal Inference: A Short Course" on YouTube is a genuine treasure. I love it.

Product analytics: this is the thing I'm worst at, even though I deal with it every day at work — interviews for this, honestly, are all about knowing the playbook. I personally don't like Meta's interview style, and I've bombed with pretty much everyone I interviewed who came out of Meta. I genuinely couldn't find my footing — sometimes I'd get the metric wrong, sometimes the HR feedback was that I hadn't clarified the question. I talked to a friend who's at Meta and suddenly realized I'd been an idiot the whole time. I'd recommend these two books — useful for interviews, and useful for the actual job too:

"Ace the Data Science Interview" by Kevin Huo and Nick Singh — the product sense section is written really well. It lays out exactly how to approach each type of question and how to break it down, and it gives you a framework you can genuinely reuse, with a ton of examples.

"Lean Analytics" by Alistair Croll and others: fundamental product analytics. Different industries have their own metrics — what matters for a two-sided marketplace, what matters for SaaS, how to analyze a social network, how to define your key metrics, and so on. No matter how much AI changes things, business models have basically stayed the same for a decade-plus, so this stuff still applies no matter what comes at you.

AB testing: last one — one book is enough here, "Trustworthy Online Controlled Experiments" by Ron Kohavi. It's a well-known book, and I get something new out of it every time I reread it. Kohavi is really good at explaining genuinely confusing concepts, and he's written a lot of papers too, which are also worth the time — search around if you're interested.

On ChatGPT: I'm genuinely grateful for it. Whenever I didn't understand something, I'd just ask, and it's really good at summarizing the question and the reasoning accurately. My usual first step was to use it to dig into a company and the business it's in — where the revenue comes from, where the customers come from, industry characteristics, and so on — and combine that with the industry-specific metrics from the books above, and basically the main product metrics would fall out of that. AB testing and product analytics questions, which are tightly tied to metrics, stopped being a real problem after that. Since I'd read so much, I could naturally anticipate product-related questions and the traps an interviewer might dig, and after talking it through with ChatGPT a bit more, there generally weren't any major issues left.

Last thing — my personal takeaway is that this whole job search process taught me more, faster, than the last several years of actually doing the job — which maybe just says something about how coasty my org has been. But throughout the prep I could really feel myself getting sharper and could feel the concrete effect of adjusting my strategy along the way. A few takeaways:

• If you have the connections, talk to friends at your target companies and see how they think about these problems. Some blogs online are genuinely well written — I'd recommend Meta's in particular. Before one interview, HR sent me a prep doc, and looking back now, it was actually a goldmine.
• After every rejection, always, always debrief on why you failed. Most of the time, right after the interview, you already have a sense of where you didn't do well. If you get a kind recruiter, seize the chance to ask for feedback. Make sure you don't fall into the same hole twice.
• I don't personally recommend endlessly grinding through old interview reports. Even if you get lucky and hit the exact same question, you might not actually understand it. If you can, look past the specific question to what it's actually testing and the strategy behind solving it — that's the real answer.
• Interviewing is a skill. There are patterns and methods you can follow, but all of that only comes from practice. Unless you're some kind of prodigy who passes every single interview, you're going to miss out on some companies — so if you get the chance, interview a lot, and keep a learner's mindset. I went from being completely lost at the start to making it to the final round, and eventually getting all "strong hire" feedback — every bit of that came from stepping into more pits than I can count along the way. Without those earlier failures, with what I actually knew, there's no way I'd have landed this so directly — but I'd rather the people who come after me learn from what I went through and take fewer wrong turns than I did.