darren whittingham - Fotolia

Analytics VP shares best practices for hiring data scientists

Data scientist jobs can be notoriously difficult to fill, but businesses can turn to tools that fill in gaps in staffing and take care of their analytics needs.

It's no secret hiring data scientists can be difficult, but resources are out there for those who know where to look.

Huey Antley joined KAR Auction Services as vice president of data science solutions about a year ago. During his time with the business-to-business wholesale vehicle auction company, he has been busy building a team of data scientists.

He has hired eight people for his team with a mix of PhDs and master's degrees. Some came straight out of academic programs, while others have been working in data science for a while.

But building out KAR's data science capabilities hasn't just been about hiring data scientists. Antley has also leveraged available services to extend his team's capabilities. These include outsourced service providers, consultants and automated tools that perform some of the data science tasks.

"It's challenging to find people, but my strategy is to have a core internal team, but have external resources, as well," Antley said.

In addition to occasionally working with consultants on specific projects, Antley works with an offshore data science contractor in India. He said these resources allow the team to scale up quickly based on project demand, without committing to large ongoing investments. His team also uses a machine learning platform called DataRobot that automates the selection of modeling techniques based on specific features in data sets. Antley said this allows his in-house data scientists to skip some of the more mundane tasks of machine learning.

He's also focused on hiring data scientists with a mix of skills. A small team means that everyone has to do a little bit of everything. For example, KAR has no formal data management team. Data scientists must be able to get the data they need themselves. Most of the data the team uses is in SQL databases, so querying the data doesn't demand a unique skill. But SQL query skills have to be paired with more complex data analysis skills, including the use of R programming and Python, the two primary analytics tools used by the team.

To date, the team has built two operational predictive models: a channel optimization tool that helps vehicle sellers determine the best times and locations to list their vehicles, and a price optimization tool that analyzes historical and real-time sales data, as well as economic and market trends to determine the optimal price for vehicles.

Antley said it's becoming more and more important for enterprises, regardless of the industry, to have people on staff who can build predictive models and perform other types of advanced analytics. Even though KAR isn't a traditional technology company, investing in this kind of predictive analytics has given the company a leg up on the competition and made the organization more productive.

"Almost every company is beginning to realize that they are collecting data that gives them a unique view of their market," he said. "Leveraging something that has been thought of as exhaust in the past, but [that] can be turned into intelligence now, is important." 

Next Steps

Data science jobs don't measure up to the hype

Can't find traditional data scientists? Look for physicists

Don't assume BI staff can transition straight into data science jobs

Dig Deeper on Predictive analytics