i made this switch two years ago and now i'm on the other side, doing phone screens for our team. the gap between what analyst-to-DS switchers think will trip them up and what actually trips them up is pretty wide.
what they prepare for: SQL, python, the usual. these are fine. most analysts are solid here.
what actually gets them: the modeling depth questions. not 'can you code a logistic regression' but 'you built this model, accuracy looks fine, stakeholders are happy, what's the next question you ask?' if you haven't had to think in terms of calibration, feature drift, business outcome alignment, or how your model's failure mode maps to the business cost, you'll sound like you can run a notebook but haven't shipped anything that mattered.
the other gap: framing your analyst work in terms of decision impact. a lot of analyst experience is reporting-oriented, which is fine, but DS interviews want to hear you say 'the analysis led to a specific decision that had a measurable outcome.' if your work mostly produced dashboards that got nodded at in meetings, you have to be honest about that and reframe: what was the closest thing to an actual decision? lead with that.
for the ML-heavy DS roles, the bar is real. you'll probably need to show you've touched model training, tuning, and evaluation outside of courses. kaggle helps a little, but a deployed side project where you had to actually think about production constraints helps more.
level-wise: most analyst-to-DS switches land at L3/IC2 or equivalent even with 3-4 years of analyst experience. that's real and worth knowing before you start. negotiating back up to your previous level is possible after a year or two but you're usually taking a step down. i went from a senior analyst title to 'data scientist' which was technically a lateral on comp but felt like a step back in title. worth it for me. your math may differ.
anyone navigating this right now, happy to answer specific questions about the technical screen format.