Building a data science team is difficult enough, but growing one without losing the team's effectiveness is a major challenge. Here's why overspecialization is the wrong approach to growth.
You've built a great data science team, and everything's going fine until you realize you need more team members. It's a scary endeavor to mess around with a highly performing data science team, but if you've done your job right, you'll face this challenge at some point.
I've talked before about aligning the structure of your data science team to the size and maturity of your organization. This addresses macro-level concerns as your organization grows; however, there are intra-team design issues to deal with as your data science team expands to meet more localized challenges. When leaders try to grow their data science team, they often throw it out of balance, causing a drop in team effectiveness. As you grow your data science team, be careful of overspecialization, as it can ruin your team.
A balanced profile
Your data science team should represent a balanced set of skills. Whether you realize it or not, the most likely cause of your data science team's initial success is a balanced profile of essential data science capabilities.
Pure data science is the convergence of mathematics (statistics, linear algebra, calculus, etc.), computer science, artificial intelligence, and visualization. In addition, effective data science teams have capabilities in leadership (vision, motivation, change management), management (planning, execution, quality control), and subject matter expertise. A dream team would consist of individuals who all have high levels of talent and skill in all of these areas. In practice, you're more likely to bring together a group of individuals who each specialize in one or two of these skills. This introduces the risk of overspecialization.
Overspecialization happens when the overall composition of the team favors one or two specific areas. For instance, if your team has a member with a PhD in Physics, a member with a PhD in Mathematics, a member with a PhD in Statistics, an entry-level programmer, and a former marketing manager, you don't have a data science team — you have a scientific research team with a couple of frustrated extras. Overspecialization is prevalent in growing data science teams, because it's hard to add one or two people (who are probably specialists in one area or the other) without throwing off the balance of the team. Consequently, this can turn a well-functioning team into an unproductive mess.
If you instinctively want to leave a highly performing team alone, it's for a good reason. When your data science team has momentum, the last thing you want to do is ruin it. Overspecialization throws off the harmony of a highly performing team and, if a dominant group emerges, you'll lose the integrity of your data science.
In Silicon Valley, it's much easier to find a Python programmer than someone who has a firm understanding of Principal Component Analysis (PCA). And although both skills may be desired on a properly functioning data science team, if you try to grow by adding a bunch of inexpensive Python programmers, they will form their own subgroup and overshadow the capacity of the rest of the group, and there goes your team's effectiveness.
When this happens, you'll have a larger issue with organizational change adoption, and your ability as a leader may even come under question. Everyone in the organization is watching to see if you can execute on your vision of incorporating big data analytics into your corporate strategy.
The formation of your data science team is a large component of this, and its initial success will bolster your position as a leader. However, if you cannot scale this success into larger wins, it will compromise your influence. For the sake of your corporate strategy (and possibly your job), it's important that you prevent this from happening.
What makes your team tick?
If you pay attention to the overall composition of your data science team, you can keep overspecialization in check. It's a good idea to maintain data science profiles on each team member, so you can easily combine them to create an overall team profile. You can do this simple exercise with the existing data science team. Brainstorm the skills and talents that are necessary for team effectiveness (don't forget the leadership and management skills), and have each team member do a self-evaluation. Then aggregate the profiles to create a comprehensive team profile. If the team is already effective, most likely you'll see the balance emerge in the aggregate profile.
Before adding additional resources, make sure the resultant profile won't become too imbalanced. With so many competencies to consider (including leadership, management, and subject matter expertise), it's nearly impossible to add resources without introducing some imbalance. That said, there are ways to minimize the imbalance, and to correct any imbalance once the new team is formed.
For instance, before adding someone with a PhD in Statistics to the team, send her through a few classes in computer programming, visualization, and organizational leadership. When she joins the team, she should mentor the graphic artist who's excellent in visualization but a little weak on Probability Theory.
Our fictional company sells hiking gear. The CEO has announced a new corporate strategy of adding an online user experience that recommends great places to hike based on customers with similar profiles.
The first phase of the strategy is a complete success. Over an eight month period, the data science team was able to stand up an online recommendation engine, and build a small but active community of loyal hikers. The recommendation engine is good but not great, and the strategy calls for a breakthrough service and not just a competitive one. The team needs more firepower, but the CEO doesn't want to mess up a good team.
The CEO organizes an offsite with the data science team to determine its profile. A quick brainstorming session produces the key competencies that make the team effective: mathematics, machine learning, Python computer programming, geographic data visualization, web design, hiking, leadership, and management. After an exercise in self-evaluation, it's discovered that individual profiles have spikes in specialization; however, the overall team profile looks balanced. The CEO explains that the team needs to grow, and more resources will be added.
The CEO grows the team in a balanced fashion. Three people are brought in with complementary skills: one is strong in mathematics, machine learning, and geographic data visualization; another is strong in Python and web design; the third is an MBA with strong leadership and management skills. All three people love hiking and know the business well. This new team is not only more powerful, but it's also balanced. Once the new members gel with the current team, the CEO's in a great position to take this strategy to the next level.
Growing a highly performing data science team is a scary proposition, but if done properly, it can dramatically increase its productivity. The key is to identify the individual skills and talents that make the team effective, and then grow the team in a balanced fashion.
Overspecialization, which is unfortunately common in data science teams, introduces an insidious group dynamic that will compromise team effectiveness. I've shown you how to avoid this by being thoughtful about how resources are introduced into your team and being mindful of the resultant team profile.
Take some time today to understand what makes your data science team tick. You've got a good thing going — don't mess it up now.