Did a BCG MLE loop a couple months ago for their Advanced Analytics team. 8 years in ML, mostly NLP and recommendations. Going in I expected something between a consulting firm and a mid-tier tech company in terms of rigor. That was roughly accurate, with one exception.
What the loop looked like:
Five rounds total. Recruiter, two technical interviews, one case interview, one behavioral panel.
Technical round 1: ML fundamentals and system design (75 min) Opened with ML concepts: bias-variance tradeoff, regularization, how you'd diagnose a model that performs well in training but poorly in production. Standard stuff. Then shifted to an ML system design exercise: design a recommendation system for a retail client. They weren't testing for Google-scale infra. They cared about problem formulation (what are we actually trying to optimize?), how you'd evaluate the system offline before deploying, and what monitoring you'd put in place post-launch. I talked through precision@k vs. NDCG, A/B test design, and the cold start problem. Good conversation.
Technical round 2: coding (60 min) Python. Two questions. One was feature engineering on a sample dataset: clean it, handle missing values, explain your choices. The second was implementing a simple gradient descent from scratch (not from a library). They wanted to see that I understand what's under the hood. Not leetcode-hard. More like a solid ML coding screen.
Case interview (this is the wild card) A client wants to build a churn prediction model. Walk me through how you'd approach it from data to deployment. On the surface this sounds like a technical question. It's actually a case interview in disguise. They wanted structure first: what's the business problem, what success looks like, what constraints matter. Then technical second. I leaned too technical too fast on my first pass and the interviewer redirected me toward the business framing. That's the lesson: at BCG, even the ML case is a consulting case first.
Behavioral panel Two people. Mix of situational questions and some culture-fit ones around working in a fast-changing client environment and handling ambiguity. If you've only worked in slow-moving enterprise environments this round might expose some friction.
I passed and got the offer. The comp was in the $170-190k range all-in for a senior IC level in New York, which is below top FAANG but above many mid-size tech companies. The work itself is applied, fast-paced, and client-driven, which is either appealing or not depending on what you want.