Organizations seeking a competitive edge are increasingly looking to hire data scientists to parse through all of the information they collect and draw actionable insights from it. But building a data science team requires a strategic approach and realistic expectations about what these professionals can actually do, experts said.
“Data science is new to companies, despite assuming that it operates the same as other existing technologies,” said Ryan Johnson, head of data science at GoGuardian. “There’s an assumption that it will operate just like an IT department or engineering department, but it’s fundamentally different.”
Because it is a relatively new discipline, companies often try to dip a toe in the water by hiring a data scientist who is fresh out of school and new to the workforce, because they command a lower salary. “That’s the strategy companies are taking, and it’s not working out for lots of them, and causing some damage to early-career data scientists,” Johnson said.
If you’re an early-career data scientist, Johnson recommends against taking a job as the first data scientist at a given company.
Instead, “they should be looking to join an established team if at all possible,” Johnson said. They should also ask what the data science team looks like, including who it operates under, and if that person has familiarity with the field and the scientific method.
How to create a data science team
The majority of the work involved in building a data science team is not technical, said Meta S. Brown, business analytics consultant and author of Data Mining for Dummies.
“Most of what you’re doing is building bridges with the business,” Brown said. “You have to build something that begins in a way that’s acceptable to the rest of the business and makes them feel that they’re getting something valuable, and gradually enable them to do more and more by building trust.”
Data scientists, or those interested in hiring them, should begin by building trust through small data-driven projects, that make the work people already do a little easier for them, Brown said. After that, they can go about building a team to fit that model.
However, hiring managers need to be realistic about the makeup of the data science team, Brown said.
“In our dreams, we have a team of people who have the specific, technical skills that we need, and they’d all be very comfortable talking business,” Brown said. “But you have to hire real people.”
SEE: The data scientist job interview: Questions to expect and questions to ask (free PDF)(TechRepublic)
A more reasonable approach is to determine what you need at your current stage, and how to build a team of people with different competencies, instead of seeking unicorn candidates who will be difficult to find, she added.
“That may mean that if you need certain kinds of programming skills and certain knowledge of statistics and certain kinds of business knowledge, it may be that you’re going to hire not three magic people who have each of those things, but rather three individual people, each of whom have a primary strength at one of those things,” Brown said. “Maybe one person is primarily a programmer, one person is primarily the statistics expert, one person is primarily the liaison to other departments. And that person might not call themselves the data scientist. That person might be called a business analyst, for example. But the point is not to be called data scientists—the point is to do valuable research for the business, and help them get their work done.”
Avoiding data science failure
Some research suggests that the majority of big data projects fail—one of the reasons being that they are not approached as projects that solve a business problem, Brown said. Instead of educating managers on analytics, data scientists should educate their team on what goes on in the business around them, and how to adapt to fit the business.
To avoid data science failure, companies must first examine the data they actually have, Johnson said.
“Every company thinks they have tons of data,” Johnson said. “I would advise companies to think about not just the volume, but the quality of the data they have, and if they don’t find those to be very to be very high, then they probably don’t even need data science yet. A lot of companies are putting the cart before the horse there.”
Companies also must establish a data engineering team responsible for collecting, storing, and curating data, before hiring a data scientist, Johnson said.
CIOs who are interested in hiring a data scientist or building a data science team should sit down with other C-level executives and think through what the organization needs, where the team would be placed, and how it would work, Brown said.
“In many places, there’s an assumption it just goes under the CIO, and I think for many places, that’s not going to work well,” Brown said. “The CIO typically has no training in analytics at all, and whose focus is on making IT operations work well, so the person is also not necessarily some primary expert in the business problems that the organization wants solved.”
The marketing organization often has the most interest and acceptance in using analytics, Brown said, and may be a good space for a data science team.
Meanwhile, data science job candidates should decide what the role means to them personally before looking for a job, Brown said.
“This is very important, because they better go looking for an employer whose ideas are similar, or they’re going to be very unhappy,” Brown said. “And that’s actually a very common problem. There are a lot of stories about people being hired as data scientists and then not liking their jobs at all. And I think that the fundamental problem there is that they had an idea of what the idea would be, and didn’t fully explore that with their prospective employers. And so they’ve got a mismatch.”