The Most Common Data Science Interview Mistake
OK, that’s a clickbait-y title. Sorry.
At least I’ll get to the point right away. I think the most common data science interview mistake is premature problem-solving.
In my interviews, I usually describe a business problem and ask for a data-driven solution. For example, our locker network doesn’t always have enough capacity to satisfy user demand. What model would you use to regulate the inflow of shipments?
Bad candidates proceed to give a poor solution. Good candidates reply with a promising idea. Great candidates, however, begin by asking insightful questions.
Premature problem-solving is very common in interviews. That’s understandable: If you hear a question, the natural instinct is to answer it, not ask a bunch of clarifying questions. But, actually, you should ask a bunch of clarifying questions. Real-life business problems are vague, complicated, and messy. You can’t provide an excellent solution without first understanding the problem.
There’s another benefit to asking questions. You’ll always have less information as a job candidate than the folks interviewing you. Sure, you’ve done your research. But you’re still at an informational disadvantage. The unfortunate implication is that many things that seem smart from the outside will sound pretty naive to those on the inside (i.e., people interviewing you).
What are some questions you can ask? That depends on the business context. Here are some relevant questions for the case above:
What are the key business metrics we should optimize for?
What data do we have available?
Are we looking for a quick fix or a long-term solution?
How does inflow management fit into the broader business strategy?
Of course, if you are asked a straightforward question (“what’s the difference between supervised and unsupervised learning?”), you should give a straightforward answer. No need to be like, “but what do you really mean by learning?” Just respond directly.
I already hear you saying: “Sure, but how do I know if I’m being asked a straightforward question?” Fair point. As a rough guideline, most technical questions (“what’s the minimum detectable effect size?”) are “straightforward.” On the other hand, most business-case-type questions are “not straightforward.”
That said, it’s a balancing thing. You don’t want to jump into solving problems you don’t understand—that’s premature problem-solving. However, you also don’t want to over-analyze each question to infinity—that’s analysis paralysis. On an axis of “number of questions asked,” you want to be somewhere in the middle:
So, if you’re preparing for a data-science interview and want to stand out, here’s a tip: Ask insightful questions & avoid premature problem-solving.