Topics: AI | Data Science | Data Visualization | Deep Learning | Machine Learning | NLP | Python | R | Statistics

KDnuggets Home » News » 2021 » Feb » Opinions » Who is fit to lead data science? ( 21:n06 )

Who is fit to lead data science?

Data science success depends on leaders, not the latest hands-on programming skills. So, we need to start looking for the right leadership skills and stop stuffing job postings with requirements for experience in the most current development tools.

By Polly Mitchell-Guthrie, VP, Industry Outreach & Thought Leadership, Kinaxis.

Data Scientist Stool

Would you hire a director of data science if she has been leading analytical teams for 10+ years, her teams won awards for their work, presents actively at conferences, and has a PhD in industrial engineering from a top program? Of course! But would it give you pause if she didn’t have significant hands-on experience with tools/techniques like deep learning, Kubernetes, and Tensorflow? It shouldn’t, but this job requirement persists in many job postings for senior data science leaders, so let me explain who is fit to lead teams and why experience with the latest toys should not be top of the list.

A few months ago, I spoke with the professional colleague who I just described above because she was on the market after years of leading teams for a Fortune 100 company in an industry that competes heavily on analytics and data science. She was lamenting that so many job descriptions want "hands-on" experience with the latest tools. I’ve long seen this gap between what is needed and what so many job postings demand, so I wrote about it on LinkedIn. The post stimulated significant discussion, so I decided it was worth expanded commentary.

I’ll repeat my plea to HR and those hiring data science leaders: if you want to maximize results from your data science teams, please hire people who are good at leading people and have experience delivering results with data science. Stop insisting on years of hands-on experience with the latest tools and techniques, which they won’t touch themselves once they get hired. Instead, value the skills they will actually need, which are far more critical to a team’s success than proficiency in GPT-3.

While you don't want someone out of touch, what you DO want is someone who knows how to frame business problems, help data scientists abstract that business problem into a technical structure they can model, coach them to explain their results to business leaders, considers usability and deployment, and develop careers. I view the skills needed in data science like the legs on a stool. Programming experience, quantitative rigor, business acumen, and interpersonal skills are all essential, but it takes a while for the stool to be balanced. A junior data scientist is part of a team, so you can afford for his stool to be wobbly because others can compensate for areas where he is weaker. Your leader’s stool needs to be balanced, but that balance is the result of skills that developed into steadiness over time.

The balance is illustrated well by an exchange I had years ago with another former colleague, Annie Tjetjep, who relates development for data scientists to frozen yogurt. She argues that people start with one set of strengths, just like the first swirl of frozen yogurt added to the cup. For most junior data scientists, the first legs to develop are in programming and quantitative rigor, which they learn in university or through online courses and practice. And in spite of the fact that research has shown that interest in communication and interpersonal skills dominate ads for data scientists more than any other skills, academic programs continue to shortchange curriculum in these areas.

Annie Eating Ice Cream

Over time Annie might add “…creativity (which I call confidence), which improves modelling, then business that then improves modelling and creativity, then communication that then improves modelling, creativity, business, and programming, but then chooses to focus on communication, business, programming and/or modelling - none of which can be done credibly in Analytics without having the other dimension. The strengths in the dimensions were never equally strong at any given time except when they knew nothing or a bit of everything - neither option being very effective - who would want one layer of froyo? People evolve unequally, and it takes time to develop all skills, and even once you develop them, you may choose not to actively retain all of them.” Swirls of additional flavors of yogurt get added, just like the data scientist adding to her skill set with experience.

As a data scientist moves into more senior leadership roles over her career, she’ll need to develop the legs of her stool, focusing on business acumen and interpersonal skills. By the time she reaches the director level or higher, she hasn’t likely written hands-on code in years, nor should she. She will be busy leading, which means she is immersed in the company’s business, attracting, retaining, and growing talent, identifying where data science can add the most value, ensuring data science moves successfully from the lab to production, and translating that value to non-technical leaders. Even if she misses hands-on programming herself, she won’t have had time to keep up with the latest toys with everything else on her plate.

And yet, here’s a recent job description for an SVP of Data Science: “Strong proficiency with R, SQL, and Python including the Tensorflow, Keras, and XGBoost libraries.” This person was also expected to lead a “large team of Data Scientists” to “enhance the company’s profitability, improve our customers’ experiences, and better our ability to measure and manage risk… extract valuable business insights, translate those insights into real-world benefits, and communicate the results to the executive team...collaborate with business domain subject matter experts in various departments to understand the problem and the objectives.”

The italics are mine because my point is that the italicized verbs are the kinds of things I expect someone at a senior leadership level to be doing, not writing code. The leader needs to understand the business value, technical strengths, and limitations of Tensorflow, Keras, and XGBoost, but she doesn’t need to gain that knowledge by hand. At the senior leadership level, she has spent at least the last 5-6 years rising in leadership ranks since the birth of those tools and is no longer in the trenches slinging code.

The value of hiring fresh talent out of university is they do have that experience, with strength in the quantitative rigor of the latest algorithms like deep learning and programming in the packages-du-jour, like GPT-3. But while the tools will change, what recent grads won’t have in-depth, because as noted, even the best, most well-rounded programs struggle to teach, is the enduring business acumen and interpersonal skills necessary to excel at those verb capabilities like translate, communicate, collaborate, and understand.

Those skills are the result of lessons learned, years of listening to business users complain about their problems, and translating those symptoms into a properly-framed business problem. Junior data scientists may jump into solving their first pass at the problem and throwing the fanciest math they know at a solution. Over time they will improve at abstracting the problem correctly and choosing the right methods (not always the latest) to address it, but they will get there faster if led by someone who already learned those lessons. Experience has taught me volumes, like finally succeeding at explaining operations research to health care leaders who think OR means the operating room or having my team’s models languish because we didn’t involve IT soon enough in planning for deployment.

Data science initiatives fail due to leadership, not because the leader hadn’t added to their GitHub repo in the last year. Consulting firm McKinsey’s list of 10 reasons for failures is all about those verbs I emphasized above – lack of a clear vision, insufficient understanding of the business value, gap in translation, role ambiguity, team isolation, etc. Too many data science projects fail, so let’s set them up to succeed by hiring leaders fit to lead data science teams, not program for them.


Bio: Polly Mitchell-Guthrie (@PollyMGuthrie) is the VP of Industry Outreach and Thought Leadership at Kinaxis, the leader in empowering people to make confident supply chain decisions. Previously she served as director of Analytical Consulting Services at the University of North Carolina Health Care System and in several roles at SAS. She’s been very involved in INFORMS, the leading professional society for analytics and operations research.


Sign Up

By subscribing you accept KDnuggets Privacy Policy