Credit by Tony Yiu
Some Unsolicited Advice on How to Improve the Data Science Interviewing Process
I’ve thought and written a fair bit about the recruiting process from the perspective of the job-seeker. But lately, having observed my friends (and fellow boot camp grads) go through interview after interview, I have some constructive feedback for how companies can interview better.
Business Case Questions that the Interviewer Has Not Solved in Advance
What’s the actual objective of a business case interview? It’s to test the ability of a candidate to both think critically and creatively when faced with an open-ended problem.
But as an interviewer how do you assess these things? The thinking critically part is not as hard — if the person is stumbling through basic business strategy issues or making recommendations that don’t make sense (given the context) then it’s probably a no go.
The ability to think creatively and formulate novel solutions to problems is much harder to assess. And I think interviewers (and companies in general) currently don’t do a good job at this — despite the fact that the case interview’s primary purpose is to measure this ability.
The issue, in my opinion, is that interviewers prefer to use known, solved, business cases (often ones that they themselves have worked on) when they conduct interviews. The reasons for this are obvious:
Due to having worked on it, it’s relatively effortless for the interviewer — he or she already knows the answer and how to step through the case.There is no uncertainty (and no risk of looking foolish). People in general, including interviewers, shy away from uncertainty and risk.
But these benefits are also the same reasons why the case interview is not as effective as companies generally think it is.
When are people most open to new and novel suggestions? The answer is — it’s when they themselves are not sure what the answer is. If they already have an answer in mind, especially if it’s one that they came up with themselves, then those people will be anchored and possibly biased. In these cases, if a candidate suggests a solution that is different from yours, you will probably spend the time rationalizing why your approach is better and why their’s is wrong when you really should be objectively assessing the candidate’s answer.
If the interviewer doesn’t know the answer beforehand, the interview becomes more of a back and forth brainstorming session — and a more accurate simulation of how it would be like to work together for both people. As opposed to right now, where case interviews often turn into the interviewer sitting there waiting to hear a specific keyword or concept, while the candidate desperately blathers in an attempt to check all the boxes.
So companies and interviewers, I implore you to give what I suggest a try. If you go into the interview with an unsolved problem and step out even 5% closer to a solution, isn’t that pretty strong evidence that hiring the candidate would result in a mutually beneficial working relationship?
Focus on the Candidate’s Ability to Ask the Right Questions
It’s my personal belief that if you can formulate the right question, then you are 80% of the way there. I would be much more impressed if you knew how to ask the right questions of the data and design the correct experiments (to answer those questions) than if you’d memorized the derivation of OLS.
Yes, fancy math implies a certain degree of knowledge and education. And the ability to map your ideas to Python code reasonably quickly is important too. But we live in a world of Google, Wikipedia, YouTube tutorials, and Stack Overflow.
Think about how you solve an actual problem at work:
First, you transform your complex and open-ended problem into a series of questions that are less open-ended, and that can be tackled by the data that you either already have or are able to collect.Then you collect the data and run your experiments.And while doing so, if there is code you are not sure how to write or an algorithm that you are unfamiliar with, you break out your Googling skills and most likely end up at Stack Overflow.
So having an end-to-end solution already in your head is almost never the case. Thus, you should not expect your candidates to have one either. Rather assess them for their ability to ask the right questions and their ability to be resourceful and learn things on the fly — as those are better long term indicators of the candidate's ability to add value to your organization.
Thanks for reading and cheers!