There aren't many full write-ups here that make it all the way to the onsite round, so I wanted to contribute one.
Overall timeline:
- Day 1: HR reached out to schedule the interview — sent a link to book a time for a "life story" chat, and along with it an IQ test and a personality test to complete before the chat.
- Day 4: Life story interview. The key points are already covered in other posts — this round is more of a casual conversation.
- Day 8: Pair programming. The Python question is well covered in other write-ups here, nothing to add. The SQL question had one tricky part: one of the tables was structured by start month/end month, but I needed to convert it into one record per month. I got stuck on this, checked an AI tool, and the interviewer hinted at using a range join. I looked up the syntax and got it working, but that ate up a lot of time, so I only had a few minutes left for the Python part and only finished the first question... The ideal final SQL output was: by month, pirated-theme shop count %, pirated-theme revenue %.
- Day 10: Got an email inviting me to the onsite (VO) — honestly hard to believe, I'd already assumed I was rejected.
- Day 12: Three onsite rounds, all in one day. There didn't seem to be a set order, it mostly depended on interviewer availability — and this is also where my interview experience started going downhill...
Round 1: "Tech round" (though really more of a product case). The prompt: suppose Shopify is launching the Shopify App Store — how would you measure success as a data scientist? Mostly about structuring your thinking, defining metrics, and designing experiments. But my interviewer turned out to be someone very much on the data-engineering side (I only realized this later, from his own questions) — he was extremely fixated on asking about data sources. For example, I said I wanted to track clickstream data and payment data, and he kept pushing on where exactly this data comes from, what my "data model" was. It took a while before I understood that by "data model" he meant table structure/schema design. I explained the structure of the data I currently work with, but honestly I don't have much upstream data-processing experience. He spent a huge amount of time drilling into these questions, and even when I kept flagging that we should move on, it didn't help — so there was basically no time left to develop the experiment design and evaluation part, which I think was actually supposed to be the point of this round. He also talked a lot and asked a lot of questions, and liked interrupting my train of thought to make me prove things that felt like very straightforward logic to me. It felt a bit like being deliberately tripped up.
Round 2: Data interpretation. This round is data visualization, but the tool options are Python or Google Sheets. If you want to use Tableau or another tool, apparently you can let the interviewer know in advance and get the data ahead of time — I didn't do that (a bit of a regret, because even though I'm not great with Tableau either, I'm still more comfortable with it than with plotting in Python or Google Sheets). In hindsight I should have spent more prep time on this round, because being limited to those tools really hurt my performance. I spotted an issue and formed a hypothesis fairly quickly, but actually producing the chart I had in mind wasn't simple at all — I wasted a huge amount of time just on the mechanics of plotting, felt like an idiot. Then again, data visualization genuinely isn't my strong suit or something I practice, and I hadn't prepared for it, so it's a bit deserved, haha.
The prompt: you're first shown a three-year trend and notice that, outside of normal seasonality, there's a strange traffic spike — but at the same time the add-to-cart rate drops sharply and conversion rate also dips a bit. You're asked what possible hypotheses could explain this, then have to produce charts to support your hypothesis. The provided table had several variables you could use for category analysis, but by the time you get to that point you usually only have twenty-some minutes left for charting, so it really rewards being fluent in whatever tool you pick — I'd recommend going with whatever you know best. You could also look at things like YoY. Note this round is specifically testing data visualization, so don't just focus on the analysis — make sure you're actually using charts to present it.
Round 3: Tech project deep dive. Walk through one of your own technical projects in detail. You need to keep it high-level enough for someone outside the field to follow, while being ready to go deep into technical details at any moment. The most important thing being evaluated, I think, is whether you can tightly connect the project to its actual impact (I only pieced this together myself after asking for feedback afterward, so it might not be complete). This round really depends on who you get — I was paired with someone more like a senior manager/director type, so there wasn't much technical depth; he mostly asked what model I used, and most of the time went to fairly high-level discussion of logic and cross-team collaboration experience. (If you get someone whose background lines up with your project, you might get pushed a lot deeper technically.) During the Q&A part I asked him how he felt about Shopify, since he hadn't been there that long either — his answer was pretty interesting. Roughly translated, it came down to: Shopify moves fast and the pace is intense, management is pretty top-down/one-person-decides, and the people are smart and very invested/hard-working, blah blah. It felt like he really wanted to vent but was trying hard to keep it neutral — not sure what he'd been through.
The last interviewer told me they'd have a recap meeting two days later to decide yes or no, and if yes, what level. I'm pretty sure I didn't pass — there was nothing that really stood out, and there were several real problem spots: I was slow on the SQL, didn't get to the second Python question so never really showed technical strength, completely flopped the data visualization round, and never got to develop the experiment design/evaluation part, which is actually the area I have the most experience in. Posting this write-up to give something back.
Discussion
Loading comments…