Many managers of data science teams become managers because they were great individual contributors and not necessarily because they have the skills or the training to lead a team. (I include myself in this group.) But management is a skill in itself, and relying on your experience as a successful individual contributor is not enough to ensure that you are able to retain and develop great talents while providing learning, products and results to the organization. Great data scientists have career options and won’t put up with bad managers for long. If you want to keep great data scientists, you had better commit to being a great manager.
What does it take to become a good manager? Volumes have been written on this subject, of course, including from HBR. But in my experience, there are a few areas that are particularly important for those leading data science teams. Good management means taking care of your team members, connecting their work to the business, and building diverse, resilient, and high-performing teams.
Build trust and be candid
Trust, authenticity and loyalty are essential for good management. This is especially true in data science where confusion around the discipline and its role in the organization means that the team leader is responsible for isolating team members from unreasonable requests and explaining the role of the team to the rest of the organization. Your team must be confident that you will have their backs.
Having the back of your employees does not mean blindly defending them at all costs. This means making sure they know you value their contributions. The best way to do this is to make sure that your team members have interesting projects to work on, and that they aren’t overloaded with projects with vague requirements or unrealistic deadlines (which is all too common given the high demand for data scientists.)
To build trust over time, you need to invest in the franchise. Data scientists are intelligent people who are trained to query and process information. Therefore, my heuristic is to be about 20% more direct and candid than you might think. Be transparent with the good guys and the bad throughout the entire process, from recruiting to onboarding, day-to-day, performance reviews and when discussing team, department and organizational strategy. It is painful but essential for success. The moment you start to “be nice” to avoid a difficult conversation, you and your team have started to lose.
Lastly, comments need to be consistent and two-way, and big data scientists will smell bullshit a mile away. If you say you believe in the franchise but get on the defensive or (worse!) Don’t actually act on the feedback, then your best connections will want to go.
Connect work to business
To get the most out of a data scientist’s time, he must have a clear understanding of the business purpose behind the project. One of the most important jobs of a data science manager is to anchor the work of your team in the context of a larger organizational strategy. Unfortunately, this is not always easy to do.
Data science projects often start with a question from someone outside of the team. But often the question the person asks is not exactly what they really want to know. Much of managing data science involves discussing and refining stakeholder questions to better understand what information they actually want and how it will be used. Don’t let questions or requests become projects for your team until you know exactly what the stakeholder wants to understand and how they will use it. Having very clear goals for the data-related questions that come your way is one of the most important things you can provide to your team.
Of course, stakeholders cannot always answer these questions on their own. They might not have a clear idea of what a finished data science product would look like (or how they would apply it). To fill this gap, ensure that members of the data science team are regularly invited to product and strategy meetings. That way, they can participate in the creative process rather than just responding to requests.
Build large teams
Many professionals are trying to break into the “sexiest profession of the 21st century ”and so, as a data scientist, you will have a lot of applications and need to be picky. Take the opportunity to be picky about good manners. Take care of your hiring process.
One of the main areas where people fail as managers is the trade-off between the short and the long term. For example, it’s easy to start thinking that you don’t have time to recruit. This is a huge mistake. If you don’t have the time to find good team members and carefully review your maintenance and onboarding processes to make sure you have good ones in place, then you don’t have the time. time to manage a new direct report. Creating a great hiring process will pay off in the long run.
What does a good recruiting process look like? On the one hand, it doesn’t just focus on technical skills. Social skills such as empathy and communication are underestimated in data science and the disciplines that data scientists typically emerge from, but they are essential for a team. Make it part of your hiring (but not in a way that is tantamount to hiring only for “the cultural fit” and reinforces your affinity and confirmation biases). Instead of wondering if you can get along with a candidate, ask yourself if there is a lens through which that person sees the world that expands the boundaries of the team’s knowledge sphere and place equal importance in this dimension you attach importance to other attributes such as technical capacity and domain expertise. This is why it is important to favor diversity. This includes the diversity of academic disciplines and professional experiences, but also lived experiences and perspectives.
A few areas in particular stand out as important to data science. First, don’t just hire seniors. Not only are they in high demand and expensive, but less experienced employees have the “luxury of ignorance” and can ask “dumb” questions. These aren’t exactly silly questions, of course, but aren’t cluttered with the usual assumptions that more experienced professionals cease to realize they make. It’s not hard to get enthusiastic about a particular way of doing things and forget to wonder if a privileged approach is always the best solution to a new task.
Second, data scientists come from a variety of academic backgrounds: computer science, physics, statistics, and many more. What matters most is having a creative mind paired with top notch critical thinking skills. I have a member of the team who has studied marine biology and this diversity of expertise has proven to be extremely valuable. (The ability to translate domain knowledge into the behavior of pods of dolphins in the wild can be surprisingly useful when modeling a robot fleet.)
Third, it’s important to hire people whose strengths complement each other, rather than building a team that all excel in the same field. A “big view” person, someone who can articulate stories with data, and a visualization assistant working together can collaborate to produce things that no one else could do independently. To get the most out of these complementary skills, it’s important to make sure that the team is genuinely working as a team and collaborating. You want your team to work with each other and not just side by side. Asking members to read each other’s code and reports on a regular basis and encouraging team activities centered on technical discussions allows you to make the most of this type of diversity.
Finally, it’s also important to build a team that reflects the people whose data you’re analyzing. This is the only way to ensure that you have a resilient team that will ask better questions and a greater openness of perspectives from which to ask those questions. In this way, each individual’s blind spots are covered by the past experiences and skills of another.
When to specialize
One final tip: When a data science team starts out, everyone will “wear multiple hats” and do many types of data science. It’s okay, it’s like when someone joins a startup. But as your team matures and proves its value, recognize that roles will become more defined and some activities will be transferred to other teams (infrastructure, operations, etc.).
Having said that, I caution against specializing too early. Specialization only works when well-defined and clear requirements are available to compensate for coordination delays and costs associated with collaborating multiple teams. Full stack data scientists are very difficult to find, but it is possible to find intelligent and motivated data scientists who can learn, with a little dedicated coaching, how to frame a problem appropriately, manage a small project, develop and train a model, integrate it with APIs, and go into production.
If you have done your job as a manager well, this development will go relatively well. You will have been picky in your hiring and created a great team with balanced skills. Your employees will trust you and understand how the changes support the organization and its goals.