Hiring data scientists can be a big challenge, partly because the available supply doesn't meet the demand for...
them. But that's only the first of the hurdles organizations face in building data science teams with the technical skills, business acumen and analytics bent needed to take full advantage of all the information flooding into big data systems.
In a panel discussion yesterday at Strata + Hadoop World 2016 in San Jose, Calif., a group of experienced data science team managers offered advice on finding, managing and retaining skilled data scientists, both for internal analytics initiatives and efforts to build data products for marketing to external customers. They said it starts with hiring the right types of people at the right time, then working to ensure the assembled data scientists are both productive for the business and satisfied by what they're doing.
That's easier said than done, though. Here are some of the topics that were addressed during the session, and what the panel members had to say about them:
Don't hire data scientists before the analytics "lab" is ready for them. Monica Rogati, an independent consultant who previously built and led a data science team at San Francisco-based wearable device maker Jawbone, said it's a mistake to "hire data scientists thinking that they're just going to sprinkle learning-pixie magic dust" around an organization and start generating actionable business insights. If the data needed to do that isn't available for analysis, Rogati added, the data scientists can become frustrated and restless -- "and the company feels cheated, too, because they're expensive and they're doing nothing."
Yael Garten, director of data science at LinkedIn, agreed it isn't a good idea to "bring in someone whose goal in life is to implement machine learning algorithms when there's no data available to them." She noted, though, that it can be helpful to have someone with data science skills in-house "who can help lay the foundations" for an analytics program, especially in the case of a startup that's pursuing a data monetization strategy. Otherwise, "there's a lot of technical debt to be paid later on," Garten said.
Expertise with algorithms isn't all there is to being a data scientist. Rogati, who also worked at LinkedIn as a senior data scientist in the past, said technical skills clearly are among the traits she looks for in job candidates. But another that's high on her hiring-priority list "is being grounded and having this very realistic, get-things-done attitude." Garten similarly pointed to a strong business sense as a vital trait of effective data scientists -- an idea of "what's doable, what's feasible and what's important," she said.
In addition, Rogati said data science teams need strong communication skills so they can explain analytical findings to business executives in understandable terms. Admonitions to speak more clearly to execs "used to really make me mad," she said. "But if you don't simplify it, someone else will. So, it's in your best interest to do it yourself."
Data science generalists and specialists both have their place. Early in the process of building data science teams, "when you're going from zero to 80" on the analytics speedometer as quickly as possible, jack-of-all-trades generalists who can work across various business units and departments are good to have along for the ride, said Daniel Tunkelang, a former data science director at LinkedIn. Later, when a team is up to speed and the new goal is making incremental improvements, data scientists who specialize in particular functional areas can be more useful, added Tunkelang, who also has worked at Google and other companies, and is currently an independent consultant.
Commingling data scientists and data engineers can promote togetherness. Rogati said data scientists often talk about having to "bribe" data engineers, who help prepare data for analysis, to do what's needed to enable analytics work to proceed. "You can skip all that by having a common team that has the same goals and is working toward the same thing," she added. Tunkelang said putting data scientists and engineers together on one team can also help "avoid having resentment created on one side or the other if they can't do the work they need to" because of a lack of cooperation across the aisle.
Yael Gartendirector of data science at LinkedIn
It's good to keep data scientists happy -- but not at the expense of business needs. While retaining the data scientists you hire clearly should be a priority, it can't be the only one in building data science teams. Garten said promoting "continuous technical growth," partly by adding new analytics tools and methodologies, can help keep data scientists in the fold by enhancing their professional skills.
She also advocated allotting time for a data science team to do exploratory analytics work that isn't tied to specific business initiatives or parts of a data monetization plan. "But you need to make it clear upfront that the goal is to get things done for the company," Garten said, advising that team managers spell out how much time data scientists should devote to practical analytics versus exploring data for possible insights.
In an interview at the Strata conference today, Bill Loconzolo, vice president of data engineering and analytics at Intuit Inc., said he focuses on the business problem-solving aspects of data science jobs when interviewing candidates for the Mountain View, Calif., vendor of financial and accounting software. "We talk about the impact of the work they're going to do -- for example, how much money they're going to put back in the pockets of people at tax time," Loconzolo said. "That's very attractive to data scientists. They want to solve real problems."
More data scientists, big data pros needed for IoT analytics initiatives
Some organizations are getting by on analytics without data science teams
Data products for monetizing info create ethical issues for data scientists