Follow Us:

Blogs Blogs
« Back

How to Build a Data Science Team from Scratch

According to numerous news reports, data scientists are in high demand. Chris Pouliot, Netflix director of algorithms and analytics, knows that only too well. When he started at Netflix five years ago, he was the company’s lone data scientist. Nowadays, he has responsibility for managing and building entire teams.

Who are these data scientists? How do you find them? How do you interview them, and structure them in the organization? These are some of the questions Pouliot answered for attendees of Rock Stars of Big Data at the Computer History Museum Tuesday.

First and foremost, he said, is understanding what a data scientist is, since there are varying ideas on the definition. Data scientist has been called today’s sexiest job—a label that is endlessly repeated. “This is the data scientist’s favorite quote,” said Pouliot. “We like to be called sexy. I’ll take that compliment every day.”

In reality, what people must do to become a data scientist is more about years of hard work than innate sex appeal. Pouliot said the data scientists that Netflix hires typically have a master’s degree or Phd in a quantitative discipline. Those with an undergraduate degree only shouldn’t entirely be ruled out, but they would need to be exceptional.

They should have experience with regression and time-series analysis, as well as hands-on experience in gathering data. “That’s really hard to learn on the job. It takes years and years of school to learn this,” Pouliot said. “We’re really looking for creative data scientists who understand how each component of the algorithm works.”

Experience in gathering data using tools such as SQL, Hive, Pig, or Python is also important, he said, since gathering their own data also helps them manipulate the data and understand potential problems.

Pouliot said it’s not a good idea to build a team with clones of the best data scientist you currently employ. Rather, team members should have diverse backgrounds and approaches. Still, it’s important that they have some common base of knowledge. Team members may come from electrical engineering, statistics and math, or physics and use different tools and processes. Yet it’s important for them to possess a common understanding of programming, deep math, and predictive modeling.

If team members have some commonalities and differences, it enables better creative brainstorming. “When these people brainstorm on a white board, it’s really a beautiful thing,” Pouliot said.

Horizontal data science teams lead to better brainstorming and better career paths for participants. In addition, they make it easier to shift people around to meet demand and manage the team. Vertical teams, meanwhile, provide deep business context. They also tend to produce less friction.

 In terms of hiring data scientists, Pouliot suggests starting with resumes, then drilling down by asking in-depth questions. Frequently, applicants may have worked on interesting projects, but don’t possess a very deep understanding of the technology. Next, it’s a good idea to get then to engage in brainstorming to gauge their ability to think creatively.

For data scientists just graduating from college, Pouliot advises entering contests to prove your abilities or gaining in-depth knowledge on several topics. “The first job’s the hardest to get. You just have to get that first job,” he said.

Trackback URL:


Follow CS_Conferences on Twitter