Now that you’ve gone through all of this effort to recruit, interview, and hire a data science team and have structured that team for success, how do you make sure that they stay happy and stay with you?
We’d argue that one of the most crucial factors for retention is understanding a key psychological fact about data scientists: FOMO is a real phenomenon within data science. FOMO—or the fear of missing out—is that quintessential emotion of our social-media age, that anxious feeling in the pit of your stomach that you’re sure something more fun and cooler than what you’re currently doing is happening somewhere else out there in the world. In data science, it can take the following forms: “My company isn’t doing any cool machine learning projects right now.” “We don’t really have big data…we just have a MySQL database, but I hear everyone else is using Spark.” “We’re not even doing deep learning…or reinforcement learning…or GPU-accelerated Bayesian inference for time series modeling.” Or whatever the next big thing is.
Roughly 40% of data scientists say that challenging work and learning opportunities are their top two motivating factors for changing jobs. And in fields that are moving and changing at lightning pace, as data science and machine learning are, FOMO masks a real, valid fear at its core: “If I’m not learning at my job, I’m going to be left behind and soon I’ll be irrelevant.”
So how do you make sure that FOMO doesn’t lead to loathing your leadership and leaving your company?
It’s your job, as a data science leader, to hire data scientists who first and foremost care about having a real, measurable impact on your business, not just about using the newest, shiniest methods and tools. And it doesn’t stop with hiring—you’ll constantly want to teach and mentor this part of the job, especially given that a lot of data scientists come out of graduate school where the push is to stay on the bleeding edge (with the added bonus that you don’t need to stick around to actually maintain any software…).
But assuming that you’ve done that, you’re still going to need to give your team opportunities to practice continual learning. Luckily, there are relatively inexpensive ways to keep your data scientists engaged and learning and feeling like they’re working on the cutting edge, even if your business isn’t. These learning opportunities also give you the chance to upskill a team member in a particular area that your carefully designed interview process might have uncovered as a weakness.
Our first piece of advice is to institute a Journal Club. A Journal Club can be very informal—we usually do them over lunch—and it’s nothing more than people reading a particularly interesting journal article or technical blog post, discussing it, and explaining it to one another. We’ve found success when the person who selects the paper does a short presentation to kick things off, but that’s optional. Open up your Journal Club to anyone in the organization who wants to come. It’s a great opportunity for cross-pollination and for evangelizing data science to interested engineers and other analytics-minded people in your organization.
As a leader, set the tone that reading widely and sharing new learnings is a part of what it means to be on the team by sharing your favorite articles on Slack or an internal team email list. Make sure to participate in the Journal Club. Show your team what’s important to you through your most important tool: setting aside valuable time on your calendar. Show up and ask the dumb questions. Be vulnerable and curious. This will help establish that crucial feeling of psychological safety that we mentioned earlier.
As an alternative or supplement to a Journal Club, consider hosting a data science “movie night” (or movie afternoon, to be more family friendly). Pay for some popcorn and watch a video of a conference talk or tutorial session. Many of the big conferences make their videos freely available (PyData is particularly good about this). We find that people often accumulate a long list of talks that they want to watch but just don’t have the time. Watching them as a team both carves out the time and fosters that culture of continual learning.
Next, we think it’s essential to give your team unstructured “hack time” to actively work on something new or speculative. Of course, Google popularized the concept of “20% time,” and the stories about Gmail and other popular products emerging from it are legend. Over the years, we’ve experimented with various forms of hack time for data science. A half day every week or a full day every other haven’t worked well, which can be explained in part by Google’s dirty secret: it’s just too difficult for people to actually pull themselves away from their day jobs. Plus, it’s challenging to get a meaningful piece of data science done in such short chunks of time. Whole team hackathons can solve this problem, but they don’t let each person explore what they’d most like to learn.
For these reasons, we’ve eventually landed on a preference for individually scheduled “hack weeks,” during which each team member can explore a new software package or language, a new statistical technique or tool, or do something with the company’s data that they just haven’t had time to delve into. Give your team members the guidance that they should treat a hack week as they would a vacation—plan for it far enough in advance to get clearance from any projects or meetings. To be productive, the hack week must have a concrete outcome planned ahead of time—an application, a software prototype, a notebook documenting the research process for others to read, or a blog post.
We’ve found that some light dosing of accountability helps make hack weeks successful. The project should have a small amount of daily planning in a tool like JIRA and daily check-ins on progress with a hack week “buddy.” At the end of the week, mandate a presentation to the rest of the team so that they can also learn from the experience. In addition to giving your team the opportunity to learn something and stay on the cutting edge, we’ve personally had a shockingly large number of projects that started as hack weeks blossom into important new capabilities for our teams. Keep a running list of these, which you can use as justification with your boss for setting aside the hack time.
Hack weeks are also a great opportunity to encourage your team to contribute back to open source. Explicitly condoning contributions to open source is a solid retention mechanism, and your team will expect it if you used open source as a recruiting tactic.
Finally, we think it’s important, to the extent that your company can afford it, to have a clear policy on conference attendance. We encourage abstract submission to conferences and pay to send a team member if their talk is accepted. That’s both a great opportunity for your team member but also for generating PR (and valuable recruiting points) for your company. If it’s in your budget, fund one conference a year just as a learning opportunity. If conferences aren’t in your budget, you can rally your team to speak at and attend local meetups and can also use the aforementioned “movie night” concept as a substitute.
Getting your data scientists outside your company’s walls is good for career growth and for gaining valuable perspective that can (ideally) alleviate FOMO. When your team gets out there and talks about your projects and hears about the experiences of others, team members might just realize that your team is doing cool and interesting work and that the problems it faces are actually quite common across the industry.