Three Takeaways for Structuring a Data Science Team From Scratch

As MS in Business Analytics candidates in UC Davis, we have the privilege to embark a certain practicum project specialized in data analytics. Some of us were assigned to partner with one of the leading E-scooter companies to optimize their operation efficiency. During the first phase of the project, besides exhausting industry knowledge and exploring the business data, our main achievement is that we gradually structured our team from the ground and grew towards a highly productive team by building a few simple norms. Although there is no one “right way to Rome” to build an effective team, I would like to share some key take-ways that might help whoever just starts building a data science team.

Setting Necessary Team Roles

In the first week, we found it so hard to push things forward as we needed everyone to present and make decisions together, which made us suffer three nonproductive meetings that week. Then we realized the necessity to assign team roles. With the experiences from previous batch and our current project state, we set two specific roles,

• Project Manager responsible for process management and team organization
• MSBA Industry Partner(MIP) contact as the point of contact with our practicum partner

Most of the practicum teams created the two positions, for we need the project manager on behalf of the team to make decisions and get the team on the right track if anything goes wrong, while MIP contact builds relationship with external client, requesting necessary resource and feeding MIP valuable output from our side. Some teams also tried two project managers without MIP contact, as they had several projects going on simultaneously. Meanwhile, to release decision makers from too many ad hoc tasks, we initiated task based owners,

• Two meeting notes takers, rotating among all team members
• Action owners, depending on task requirement and team member’s strength

With the basic roles settled, our meetings became much more organized. Naturally, some tasks are lying in the grey area, which are not pleasant to deal with. Our practice is that we always discuss a team-level task together, trying to break it down into clear and simple pieces, and involve members with a rotating basis, making sure everyone undertaking fair workload.

Utilizing Assessable Resource

• Regard client as your resource

On one side, MIP is our client who expects our output. On the other side, they also provide valuable input and useful tools to facilitate our productivity. In the early stage of our project when we didn’t get the access to the datasets, our MIP guided us through their business process, the lifecycle of each vehicle and all the current datasets, which turned out to be very useful in later explanatory analysis. During the first meeting with our MIP, I started to picture their business, thus I asked the favor for our team to visit one of their repair warehouses. They made it happen without any hesitation. Obviously, they will always be willing to help for the sake of the business. So is the same between the data science team and business teams that data team provides data or insights to business team and vice versa.

• Reach out to experienced professionals

Furthermore, each team was also equipped with a business mentor, who has years of industry or consulting experiences. Our business mentor helped us out when we were struggling to define the scope and use cases of our data product. As we had thrown out many questions to our MIP and sometimes even duplicate ones, we were still confused about what they wanted. Then our business mentor held a whiteboard sessions leading us to draw the full picture the business. At that time, we realized that we didn’t see the full picture of the project and hadn’t organized useful resource at hand.

Adjusting Team Structure Flexibly

Now our team is facing new challenges after getting the datasets. To accomplish our project target, we have restructured to a “real” data science team. With a clear goal and measurable workload, we changed open roles to function based roles. At current stage, we have two main tasks, Exploratory Data Analysis (EDA) and model building. Thus besides project manager and MIP contact, we have set two types of roles to form two minor teams,

• Data analysts, preparing and making sense of data through statistical analysis
• Data Scientists, developing descriptive and predictive models and deploying algorithms to power data products

Some of us were confused why we need to do EDA for such a long time. Our team member Breno, an experienced banking data science manager, said, “EDA never stops”. The EDA group focuses on getting familiar with every single variable and their correlations. They will process and feed accurate datasets to modeling team, while the later team will keep exploring feasible parameters and building the fit model to measure the business problem. At the same time, two teams will have weekly catch-ups to share findings and ask for inputs. During the modeling process, if any one of the two teams undertakes exceptional burden than the other, we need to transfer members flexibly. With the new structure, our first data product seems promising only within a week. We will continue to practice new methods for high team productivity and collect valuable take-aways.

References:

Rebecca, K(2016), How to Boost Your Team’s Productivity, Harvard Business Review

Angela, B(2018), Managing a Data Science Team, Harvard Business Review

Chuong, D(2017), What is the most effective way to structure a data science team?, Medium

Anonymous(2018), How to Structure a Data Science Team: Key Models and Roles to Consider, altexsoft