Reading Time: 4 minutes

Data science is all that I do. I’m vested in this space. I want all data science engagements to prosper! In that spirit, I hope this list helps shed some light on some very common pitfalls I’ve seen over and over and helps you course correct within your company. Because data science done right can be magical.

ONE: Turnover — With such short supply and massive demand, they’re hard to keep around!

I had a nice conversation tonight with a recruiter I respect a lot. She was talking about placing data scientists and how it’s happening more and more lately. It’s true, the field is (still) exploding. The LinkedIn Workforce Report for US (August 2018) reports that in the US alone, we have a shortage of 151,717 people with data science skills, with particularly acute shortages in New York City (34,032 people), the San Francisco Bay Area (31,798 people), and Los Angeles (12,251 people).
There are the usual reasons for turnover, like being bored at work or not liking management and all the normal issue, but data science has its own particularities.
Your data might not be all that! Munging is 80-90% of the work they do. And the majority of data scientists I work with fall into one of two camps:

  1. Data engineers or infrastructure experts
  2. Modelers

Neither one of these titles is going to be happy cleaning data all day for weeks/months, especially with typical result-driven deadlines.

Another problem retaining data scientists I see is a problem space that is too narrow to keep long-term interest. On the opposite side, a company like Trunk Club has such a wide problem space, it’s an incredible challenge to work there. There are data science touchpoints throughout the company, and most are really exciting to data scientists, from product recommenders to marketing and sales optimization. At big, diverse companies like that, there’s a lot to work on, and little to get bored with.

For companies with a narrow or single product offering, where margins are much tighter, such as in ad tech, data scientists can get caught in endless optimization of small parameters, the shape of which are often dictated by how marketing behaves. In this setting, data scientists can get caught chasing unpredictable marketing or consumer changes day in and day out. These are things that (generally) can’t be modeled or predicted. That sort of tail-chasing tedium can get tiring and burn out in those positions is common.

TWO: Data scientists are exceptionally difficult to manage

A scenario I’ve seen before: a company hires a couple data scientists, sets them up with nice desks, gives them computers, company mugs and maybe a couple company tee shirts left over from the trade show last week. And they’re off!
But doing what? And if the “what” is clear, then is there a “why? And who will be there to measure success and provide guidance? Someone must be in charge of them!

This is so much trickier than it sounds, because, as the headline says, data scientists are exceptionally difficult to manage. Not because they’re inherently different than anyone else, but because their skills and tools can be difficult to understand. And without knowledge of the skills or tools available to a resource, it’s challenging to know what direction the person will go in to solve a problem. Even framing the problem and articulation the objective(s) is something I rarely see done well in all but the most veteran teams.

THREE: Your data scientists are on the wrong team

Data science is done right when the team is carefully positioned next to or within the product team. After all, data science is almost always there to make product more profitable. This alignment is golden and serves two wonderful purposes:

  1. Being tied to product success is incredibly motivating to data scientists and
  2. The company benefits from the most efficient access to data scientists

The point here is to avoid thinking about data science like a technology. It’s not. It’s pure product enhancement or optimization. I like to think about various company stakeholders as customers for a data science team.

What if Fulfillment needs help maximizing sales while minimizing inventory? Sounds like a perfect data science project. How about when the marketing teams needs help personalizing its outbound marketing materials? This is another perfect opportunity for data science. But in neither of  those cases is data science a pure technology; rather it’s an enhancement to each of those product needs. It’s not just semantics; over and over again, I’ve seen that data scientists work best when they report to product.

FOUR: Lack of relevant mentorship

There are tons of online data science courses and books available. There are tons of good programmers around, and project managers to talk to. But nothing can replace tested insights from a senior colleague who’s “done it all before”.

I’ve been lucky enough to have had a wonderful mentor during my many years at SAIC. Now, I do my best to provide that to those I work with. It’s valuable in both directions, and I believe, necessary. Mentorship can foster the right kind of growth, and without it, job satisfaction dwindles.
Providing a budget for conferences, online courses and books is fantastic. Please don’t stop that. But it can only go so far.

FIVE: Your CEO doesn’t care about data

I know. How could that be? Especially today! But it’s out there. There are still tech companies running initiatives where the final approval to buy an ad or sell an asset or market to a particular demographic, comes from a person. A gut check, rather than an informed analytical model. Even at scale. I’ve seen it.
This is one of my first questions when I start working with a new firm, “How do you use data to make decisions?”

It’s not fail-safe; The answers can be deceptive, because everyone looks at Tableau each morning (you do, don’t you?!) before starting your day. You want to know that everything’s ok and the main KPI’s are tracking and a bug hasn’t taken the car off the rails. That’s fantastic, but it’s not driving innovation and profits by building and leveraging intelligent models that work repeatedly, even when the CEO has the flu 🙂

Culture starts at the top, and when data driven initiatives don’t come passionately from the CEO, it’s going to be difficult to drive meaningful ROI from data science.

What do you think? Have I missed any? Leave them in the comments; I’d love to hear from you!

Zank Bennett is CEO of Bennett Data Science, a group that works with companies from early-stage startups to the Fortune 500. BDS specializes in working with large volumes of data to solve complex business problems, finding novel ways for companies to grow their products and revenue using data, and maximizing the effectiveness of existing data science personnel. bennettdatascience.com