Micah Melling, director of data science, outlines the traits leaders of data science teams should maintain to inspire growth and innovation.
By Micah Melling
Spring Venture Group is incredibly open to original ideas and innovations. Even when I first started as a data scientist at the company, I had the freedom to implement new strategies and methodologies. That’s specifically why I enjoy being a data science leader here: I am able to give that same autonomy to my team and watch smart people do outstanding work that directly impacts the company’s performance.
From my experience, an effective leader in data science knows when to help and when to back away — this space between support and independence is when breakthrough innovation occurs. Beyond providing technical prowess, I believe leaders should break down hierarchies, provide frameworks to conceptualize problems, and be willing to dig into any and all work.
By definition, data science is a fusion of multiple fields. Leading a data science team overlays another blanket of complexity onto the role. Below are six traits I believe to be distinctions of next-level data science leaders.
Be open with people and willing to dive into the weeds
Oftentimes, fusing multiple minds makes hashing out a programming or systems design challenge more tenable. Leaders should be willing to collaborate, roll up their sleeves, and dive into the work at hand, no matter what it is. Having a gritty mentality lets the team know that we’re working for them and always willing to jump into the mix.
This mindset also encourages open dialogue among all parties. When the team is comfortable pitching new ideas and speaking frankly about roadblocks, we’re more likely to find practical solutions. No one has a monopoly on effective solutions; the hierarchy of ideas should be flat.
Openly sharing company metrics and performance data is also a crucial lynchpin. This action encourages buy-in and spurs everyone to become invested in the company’s success. When people see their direct impact on the company’s bottom line, their perspective changes. They become increasingly creative, curious, and motivated to innovate.
Regular team meetings, idea-generating sessions, and purposeful project management are all important aspects of being a data science leader. While data science leaders are not in the weeds as often as their team members, taking a back seat to the daily work is not a viable option. The job pivots the leader into a different role: gathering requirements, setting general parameters, and supplying the right amount of feedback.
Without solid project management and clear desired outcomes, the team won’t know the problem they’re solving. In data science, we have an incredible number of tools at our disposal. Give a data scientist a problem and they can likely find dozens of ways to solve it. The key is to provide the framework for conceptualizing the challenge and communicating the desired end state. After we have the task correctly framed, we can unleash the data scientists to find the optimal solution with their vast toolboxes. Leaders don’t need to explain how to put together the puzzle, but they do need to make sure the team has all the pieces.
Maintain mutual trust
Data science teams tend to be filled with brilliant folks who enjoy experimenting. Many disciplines have prescriptive recipes for doing their work — data science is comparatively unstructured. General steps assuredly exist: we will always need to clean the data, model development should follow certain best practices, etc. That being said, data science is always ripe for trying new methodologies and processes, and many data scientists actively seek this opportunity. This often pushes us into new waters.
For continuous experimentation to succeed, we have to trust the team is working every day to solve the problems assigned to them. We also have to have confidence that we hired the best team we could. On the other side, the team should see that we’re working for them as diligently as they’re working for us. By building a mutual level of trust, the team becomes motivated to experiment and succeed.
Trust should not be blind, however. Software systems reveal when people are producing — we can see the code. On our data science team at Spring Venture Group, we manage production-grade systems, like intelligent lead routing and lead bidding algorithms. I am able to see firsthand if my team is innovating and generating new methodologies.
One of the hallmarks of exceptional data science is vast curiosity. Infinite ways to improve machine learning systems exist, which can provide a constant challenge for data scientists — and especially managers. An effective leader should be constantly seeking out new methodologies and considering how they can be applied to the team’s work.
Without curiosity, growth will be at a minimum. When we maintain a culture of creativity in our data science departments, we are in essence creating an environment that supports high-impact experimentations. Many data scientists are naturally curious; allowing them to search the depths of their minds is imperative. As leaders, our personal curiosity feeds into how the team tackles their daily challenges, and it buoys their collective creative energy in developing forward-thinking systems.
Keep up-to-date with the larger business context
As leaders, we must stay abreast of both trends in our field and our business. As companies grow, their systems become increasingly dependent and complex. Keeping a strong pulse on these intricacies is paramount.
In many environments, the work of a data science team has an echo effect across the company; our work interplays with other company-wide systems in potentially intricate ways. Without knowledge of systems integration, we might make adjustments to a system that we believe is isolated, but those actions might have a cascading effect on other systems, processes, and people.
Viewing the high-level impact and context is crucial — innovation cannot happen in a silo. A great idea for one company may be a flop for another. By having a firm grasp of the business along with its objectives and systems, we can ensure our work is impactful, practical, and integrated.
In business, specifically software development, efforts will not always pan out. In large code bases, bugs will exist — it’s unavoidable since no one is perfect. As leaders, we have to understand that systems are going to break at some point and that projects will not always succeed with flying colors. Our goal is to provide swift patches and prevent the same issues from happening repeatedly.
Putting a machine learning model into production is a whole new beast. When a model is in the wild, it’s fully different than when it’s on our laptops and we’ve run it against clean, historical data. A myriad of issues can happen in production that we can’t anticipate. The only way we can see how the model actually performs and impacts the bottom line is to be daring and deploy it. When we have integrated, complex systems, the only objective evaluation is an empirical test in a true production environment. We should still work through simulations and backtests, but again, those environments operate on the safety of our isolated laptops and cleaned training data.
When we shy away from uncertainty, we will neglect worthwhile opportunities. Implementing systems that do not behave as we hoped doesn’t mean we’ve failed — it likely means we’re closer to being correct or knowing when to pivot. This is especially true for data science. The top leaders know when to risk “breaking things” but they also possess the know-how to put the pieces back together swiftly.
The moral of the story: Go fast, break things, and trust your people. We are better served by rapidly innovating, tolerating work that doesn’t go quite right, and continuously experimenting rather than moving with caution. No one has ever made a splash without first diving in.