Assicurazioni generali Business Intelligence Interview Guide

The first round caught me off guard. I expected a classic case, got a behavioural instead — forty minutes of "tell me about a time when." I'd prepared the wrong thing. I felt the adrenaline hit somewhere around question three, when the interviewer stopped taking notes and just stared at me. Not hostile, just waiting. I slowed down and that helped. Round two was the case. A European retailer losing margin in three markets. I built the structure cleanly — I'd drilled that part — but halfway through the quantitative section I made an arithmetic error I caught too late. I corrected it out loud. The interviewer didn't react either way. That silence is what I remember most. Where I felt confident: the synthesis at the end. Pulling the threads together, making a recommendation with a clear so-what. That part felt natural, probably because it maps to how I actually think. Where I sweated: anywhere the answer required me to sit in ambiguity without filling the silence. I have a habit of talking to think. They're testing whether you can think before you talk. Third round was a partner interview. Fifteen minutes of case, then thirty minutes of pure conversation. That one felt almost easy by comparison — more like a debate than an exam. The thing nobody tells you: the pressure isn't really about whether you know the answer. It's about whether you stay structured when you don't.

Questions asked: The first technical question came out of nowhere: "Walk me through how you'd build a churn prediction model from scratch — data, features, validation, deployment." Standard enough, but they pushed hard on every assumption. Why that metric? What if the classes are imbalanced? How do you explain it to a non-technical stakeholder? The case prompt was deceptively simple: a logistics company seeing a 12% margin drop over two quarters. One slide. No data appendix. You had to ask for everything. That's the test — not whether you can solve it, but whether you know what questions to ask first. The take-home was three days, a messy dataset, and the instruction: "Tell us something interesting." No rubric, no defined output format. That open-endedness is deliberate. They want to see what you reach for when nobody's watching. One question I didn't see coming: "What's a model you trust in production and one you don't — and what's the difference?" Sounds philosophical. It's not. They want to know if you've actually shipped something and felt the anxiety of it running in the real world. The small details that stuck: they always asked a follow-up to the follow-up. It never stopped at the surface answer.