Data Science at MagicLinks Is All About Effective Communication
“What information do you wish you had but don’t?”
Will Baumgardner, the director of data science at MagicLinks, asks all stakeholders this or similar questions when getting buy-in for complex projects.
“Communicating with cross-functional stakeholders is absolutely essential,” he recently told Built In. “As it tells us whether or not what we as data scientists are building is actually the right thing.”
But recognizing the importance of cross-functional communication is just the first step. Next comes grounding data science in reality. After all, creating models of real-world concepts is an essential part of the process. The data science team at MagicLinks is acutely aware of this necessity as well as the benefits of quality communication.
“It helps us interact with our stakeholders, not just because we can explain better what it is we are doing; we can interact with them in a way that is much more collaborative,” Baumgardner said.
MagicLinks is a social commerce platform that uses videos to create marketing initiatives for creators and brands.
How does frequent communication with cross-functional stakeholders — especially non-technical ones — benefit your team?
We first need to understand the business problems, the workflows and the potential methods of implementation. For example, I could create the most sophisticated supervised recommendation algorithm the world has ever seen, but if the base assumptions are not based on strong business logic or there is no way to easily action those recommendations, then it is nearly worthless.
When it comes to updating non-technical stakeholders on your team’s latest efforts, how do you translate the complexities of your work into digestible language?
In my mind, what we are trying to do with data and algorithms is to use numbers to create an imperfect model of some kind of real world concept. Each feature that we use in a supervised algorithm, for example, is just representing some idea that we should be able to intuitively understand or describe in language terms.
In my work, I may use a feature that is something like “normalized and weighted likes + shares + comments per view over a given period of time represented as a percentile of their comparison set,” which is kind of just a mathematical way of answering the question, “how engaged is this audience of this social media person?”
If we are skilled at doing this, it helps us interact with our stakeholders, not just because we can explain better what it is we are doing; we can interact with them in a way that is much more collaborative. We can say: “Look, we will handle the statistics, math and coding, but we need your help on the most important part, which is model structure and feature generation, as we consider your insider knowledge critical to understanding what’s actually important.”
Data science is like screenwriting. You want to show, not tell.’’
When have you needed to get buy-in for a data science initiative from a non-technical stakeholder?
Data science is like screenwriting. You want to show, not tell. So when I needed to get buy-in for a project related to measuring the total potential of creators to drive sales, the priority was to sprint to a viable MVP and get it in the users hands.
First, I met with my stakeholders and asked them questions like: “What are the most important piece of information you look at when considering a creator?” and “What information do you wish you had but don’t?” From an initial requirements gathering session, I created a simple tableau dashboard that allowed them to see how the model was performing, and show them some of the new information and capabilities it would provide them.
From a cultural standpoint, I think it’s always important that projects are framed from the perspective of helping stakeholders do their jobs, and to remain humble about our level of domain expertise compared to them. This is done through several rounds of feedback and iteration. In an unsupervised exercise, like this one was, the best way to get feedback on model performance is from the experts in that area. Through this effort, there was a lot of confidence when the model went into full production.
