Homegrown Data Science at Red Ventures
Looking internally to meet the demand for data science talent at RV
At Red Ventures, data drives every single decision we make — which is why we’re heavily invested in expanding our data science capabilities.
Traditionally, data scientists join our team as professionals who already have experience in the field but are looking to make a change. We also hire recent graduates who studied data science at the university level. However, because data science is still emerging as a field of study, we’ve found that the supply of qualified candidates just can’t keep up with the current demand.
So in true RV fashion, we decided to invent our own alternative hiring source, a little closer to home: We developed an in-house accelerator program for experienced RV analysts who have the aptitude and appetite for programming and statistics to transition into rockstar data scientists. Like these:
(Editor’s note: The gifs in this post have nothing to do with the accelerator program. They come from a music video the data science team produced for our Annual Meeting in Cancun, which is why it’s remained internal for months.)
(Until now.)
Plus, growing a data science team from within makes total sense for a company like ours. Learning and development is already a huge part of our company culture, which means we already have tons of educational programs in place. Additionally, candidates from this group are already familiar with the culture at RV, are guaranteed to be intrinsically motivated — and even if data science doesn’t work out for them, they can just continue kicking ass in the analyst role.
“The accelerator was an incredible, challenging, and transformative experience for me. In my previous role as an operations analyst at Red Ventures, I had never worked with our data science team or had any exposure to any data science techniques. Having something new to learn made me so excited to come to work every day. Now, as a member of the data science team, I’m continually facing new challenges and new opportunities, so I’m acutely aware of how much is still out there for me to learn. But when I stop and think about it, I’m really impressed with myself and my mentors for how much I’ve learned through the accelerator.”
-Allie (2016 graduate)
Designing a Data Science “Starter Kit”
Our data science accelerator program is narrowly focused on closing the gap between analytical and data science technical skills — such as statistical programming, data exploration and visualization, and machine learning. In order to become successful data scientists, candidates also needed to be able to interface with business teams and stakeholders, develop data science solutions into real-world products, and drive projects forward.
Our goal was that graduates of the accelerator program would feel comfortable writing code to import, clean, visualize and transform data, applying appropriate machine learning algorithms to those data, and diagnosing/validating the results.
Additionally, we expected participants to be extremely self-driven. We like to run lean, which means we don’t have one dedicated resource for data science L&D. Instead, we asked them to research independently and learn new skills by facing challenges head-on, while we provided support and structure.
With those goals in mind, we launched our accelerator program in two phases over a five month period:
Phase 1: Complete MOOCs (through Coursera and edX)
If it had been all on me to provide content, support, assessment, and feedback — there’s a 100% chance the program would have failed miserably. So instead, we used online courses from Coursera and edX to provide content and support, while I focused on assessment and feedback.
We selected coursework from Johns Hopkins University and Stanford University which provided tons of great material — both lectures and print — that covered the entire data science process, from statistical programming, to getting and cleaning data, exploring and visualizing, modeling and statistics, and applied machine learning. Participants were also required to learn the mathematics and theory behind many statistical learning techniques. Our favorite thing about the courses is that they put a heavy emphasis on applying data science techniques, which means candidates were able to demonstrate meaningful skills at the conclusion of each course.
Phase 2: An Applied Capstone Project
The Capstone project was designed to accomplish two goals:
- Give the candidate an opportunity to apply new data science skills to a real RV business problem.
- Give the candidate a real sense for what the day-to-day role of the data scientist is like.
The first goal allows us to assess if the candidate will be successful in the role, while the second goal allows the candidate to make an informed decision on whether a role change is right.
“For my Capstone project, I learned to use natural language processing to build predictive models of unstructured text, which will eventually be used to inform RV Paid Search testing decisions. One of the most valuable aspects of the program was the opportunity to build relationships with incredible mentors on the Data Science Team who passed on insightful nuggets of information throughout the program and have continued to mentor me outside of the accelerator.”
-Ryan (2016 graduate)
Levels of support
Throughout the program, we knew it would be critical to provide participants with support to give them the best chance for success. We designed our support system as a hierarchy:
At the bottom, the Coursera and edX courses have cohort fora where other (non-RV) students collaborate to answer questions. Those fora are moderated by previous students who demonstrated success and an active forum presence.
Within RV, we created a Slack channel for open collaboration during the week and encouraged participants to form study groups. Veteran RV data scientists monitored the Slack channel and chimed in when necessary.
During the week, we set aside an hour during each work week to meet as a cohort, discuss challenges/clarifications from the previous week, and preview the next week’s assignments.
Finally, candidates who started the Capstone project were paired with a mentor from the data science team to provide 1:1 help and feedback.
What we learned
Now that we have one cycle under our belt, we’ll definitely make a few adjustments to improve the experience for the next group. For example, one misstep we made was attempting to provide two different outcomes for candidates (change roles and move to the data science team; or alternatively, to augment their skillset with data science tools that can be applied within their current career trajectory), while really only focusing on a framework for the former.
In order to achieve the latter, we should have given more attention to marrying new data science skills with business problems in the first phase of the program. In future iterations, we should collaborate with business teams to define what this might look like.
“Having a program in house is really beneficial because as you learn basic skills and theories, you can use an RV lens to reinforce those ideas. I was able to learn the ins and outs of collecting data, shaping it to solve a problem, creating a model to predict an outcome, and iterating to optimize that model, all while thinking through it with the data I use every day in my current position.”
-Dan (2016 graduate)
Results
When we launched the program, our top goal was to bring at least one candidate all the way through the program and onto the team as a full-time data scientist.
We had a total of three matriculates of the program, one of whom we might not have identified as a potential data science fit otherwise and two for whom we provided a formal path for transition. All three have officially changed roles and are now full-time data scientists. (Nailed it!)
While these aren’t the first analysts in the building to change roles and join our team, they are the first to be catalyzed in order to quickly become autonomous and impactful members of the team — as opposed to going through a traditional interview and onboarding after joining the team.
Perhaps the greatest measure of success is the fact that we’re planning to do it again in the Fall — with a larger group! We’re excited to continue developing this talent source in order to continue growing data science at Red Ventures from within.
Keith Williams has been a data scientist at Red Ventures for one year. Before that, he taught high school chemistry for Charlotte-Mecklenburg Schools. Outside of work, most of his efforts are focused on teaching two tiny humans all the weird rules of the world (for example: you can’t eat ice cream for breakfast, but pancakes with syrup is totally allowed).
