Abstract: Hiring for data science roles is difficult due to the drastic Venn diagram of necessary skills—stats, ML, CS, software engineering, etc—combined with a typically huge number of applicants for every open position with highly variable skill levels.
At the same time, data science resumes have begun to blur together, converging on similar formats, keywords, example projects, and role descriptions. Job titles from previous roles have lost meaningful signal, having grown to encompass an unhelpful and barely differentiated breadth of experience and skill levels.
The difficulty of assessing such a broad set of necessary technical skills has led to a great deal of pain for both hirers and applicants. Candidates are asked to do increasingly complicated or arbitrary tasks such as answer arcane statistics questions on the spot, or live code solutions to the sort of data structure/algorithm CS problems that even professional programmers resent. Meanwhile, the sheer number of applications incentivizes hirers to apply increasingly arbitrary and potentially unfair filtering criteria such as educational prestige (or worse).
Even after initial screening, the large set of necessary skills for a role are hard to evaluate holistically in interview settings. Good talkers can often get surprisingly far because it is impossible to test everything and most scenarios involve subjective judgment.
All this interviewing is distracting, wasting engineering time and effort. It turns off the best candidates and it still often fails to prevent mistaken hires.
In this talk, I will present findings from a series of interviews with 20 data science hiring managers at leading organizations across industry (e.g. FAANG, finance, startups). We will discuss common patterns that emerged around both challenges and best practices, and make some actionable recommendations for data science teams looking to improve their hiring processes.
Bio: Isaac is a co-founder and Principal Data Scientist at DrivenData, Inc, where he leads client engagements and spearheads development of the data science competition platform. He holds a master's in Computational Science and Engineering from Harvard’s School of Engineering and Applied Sciences and a BS in Operations Research from the U.S. Coast Guard Academy, and previously spent seven years as a Coast Guard officer serving in a variety of operational and quantitative roles.