Building a Winning Data Science Team

Print Friendly, PDF & Email

Data is becoming an increasingly mission-critical asset for organizations. How you collect it, move it, clean it, and analyze it can have a real and lasting impact on the bottom line. Organizations are under pressure to be faster, more strategic, and more cost-effective than the competition.

As companies continue to walk down the data-driven decision-making path, many are realizing that one data expert is not enough. It’s too complex and too much for one individual to handle.

Driving data is core to TrueMotion. We get driving data by capturing smartphone sensor data and refining it with machine learning algorithms. We use driving data to help insurance carriers improve pricing, lower claims costs, and increase customer loyalty. Driving data also helps people understand and reduce distracted driving. From the beginning, we knew our business would fail without the right data science team. Today, our team has 15 data scientists strong. We’ve learned some important lessons along the way.

Identify the right team structure

Building a data science team starts with an internal gut check: Why are we building this team? What is their primary role? What won’t their role be? We started with the hierarchy. We looked at the business and identified a model that works for us. This starts with three core teams.

Driver Identification – This team identifies user habits and differentiates drivers from non-drivers. When you drive, they identify your mode of transportation (car, bus, plane, train, etc.). Once they define your mode of travel, they determine if you’re the driver or passenger. They use this information downstream for scoring and deeper analytics. The ability to use smartphone sensors to extract driving data is a key differentiator for TrueMotion, which makes the Driver Identification team critical.

R&D – This team focuses on two areas: distraction and claims. They identify distraction events based on driving data collected during a trip. They identify when people text, use apps, and make handheld and hands-free phone calls.

The R&D team can also identify a crash with smartphone sensor data. For example, in high-speed crashes the airbag deploys, increasing the barometric pressure in the vehicle. The barometric sensor in your phone can measure this change. Our R&D team analyzes this data and identifies crashes. This crash data is key for insurance companies to investigate crash claims.

Analytics & Insights – This team focuses on user analytics and engagement. They analyze driving data and develop insights that help increase driver loyalty and reduce distracted driving with a multitude of interventions, from push notifications to rewards programs.

Tip: There’s no silver bullet, so keep an open mind and don’t just follow the first model you read about.

Invest Heavily in Hiring

Once you’ve developed your model, how do you identify, vet, and train the best and brightest talent? The short answer: It’s not simple or fast. But, with such critical hires, it shouldn’t be.

To get a sense of whether a person is technically sound and a good fit, start with a phone screen. From there, candidates who move forward should get an opportunity to show their process and how well they can handle pressure. This could take the shape of an in-person daylong project – either individual or team-oriented. Once completed, ask candidates to present their conclusions and – more importantly – how they arrived at them. This is the most critical part of the interview process. A stellar candidate will communicate their results clearly and excel at creative problem solving. For instance, critical data science team skills include how they visualize data, if they’re able to identify patterns, and how creative they are during the data extraction process.

Data science teams are an important cultural hire. They tend to work across business lines, similar to operational functions like IT or HR. Candidates that do well in the project phase should meet with several members of the team. Just because the hiring team believes a candidate is a good culture fit, doesn’t mean the rest of the team will.

Tip: Make sure you give all candidates the same project and that it’s something they might encounter if they were a team member.

Ensure Long-Term Success

One thing we learned early on was the importance of cross-communication between sub-teams. We thought that the three teams would be naturally integrated, but found they were working in silos. As we grew, there was an increasing need for the teams to be in lock-step.

To solve for this before it becomes a problem, establish a set of opportunities to foster good communication. A few to consider:

Chapter Meetings. In these short meetings, each team updates the other teams on what they’re working on and challenges they’re encountering. Include a Q&A session and open discussion on improving efficiencies.

Weekly Learning Sessions. Every week, have one data scientist present a topic of their choice. Examples for us have included Reinforcement Learning (RL), Markov Decision Process (MDP), and Neural Turing Machines.

Start a data science book club. With the fast pace of new approaches and developments, teams are well-served to read as part of data science continuing education. The TrueMotion data science team has two reading groups. One group reads a book that is dense and mathematical. The other reads a book about implementation. The groups read, discuss, and create simulations and share code. Encourage the team to read and host informal review sessions. This will foster a sense of support in continuously improving their skills and will also accelerate the ongoing learning process.

Tip: Weekly learning sessions and book clubs should be open to any team member – data science or not.

Like building any line of business team, data science organizations must find the management style that works best for leadership and staff. It starts with identifying the right candidates, training them well and giving them ample opportunity to learn, grow and share.

About the Author

Brad Cordova is co-founder and CTO, TrueMotion where his expertise in data science, artificial intelligence, machine learning and computer programming are central to TrueMotion’s data-centric approach to curbing distracted driving on roads and highways around the world. Nearly killed by a distracted driver, Brad experienced first-hand the magnitude of the distraction epidemic and recognized quickly a lack of solutions to cure it. Since then, he has dedicated his life to making the roads safer for all. He co-founded TrueMotion in 2012. In 2015, he led the company through a $10 million Series A funding round. Brad received his B.Sc. in Physics and Mathematics from Notre Dame and holds a M. Sc. in Electrical Engineering and Computer Science from MIT.

 

Sign up for the free insideBIGDATA newsletter.

 

Speak Your Mind

*