The right data science team can turn silos of diverse data into game-changing business insights. But piecing together the perfect combination of technical skills and personality traits isn’t easy. Joshua Sullivan, a vice president and data scientist at Booz Allen Hamilton, a management and technology consultancy in McLean, Va., says the best data science teams aren’t just a gathering of computer scientists, mathematicians and statisticians but a powerful blend of intellectual curiosity, visualization expertise, storytelling skills and domain expertise.
Like any group of people, that blend of talents will develop its own culture, and Sullivan says a dash of conflict among team members can help the group’s effectiveness. That’s because data scientists are “often naysayers,” known for arguing that a particular mathematical model or algorithm “won’t work” – a voice of opposition that, in the right group dynamic, can give rise to fresh perspectives and new approaches to exploring data. (More on this point later.)
Sullivan would know. He’s spent the past 17 years applying computer science to advance the national intelligence and military tradecraft, specializing in large-scale distributed systems and cyberspace. Here, Sullivan provides 6 strategies for building the right data science team for your organization.
1. Make intellectual curiosity a priority.
It can’t be taught, nor bought, yet it’s “the number one criteria” Sullivan searches for in a candidate. “Technical skills alone are insufficient,” he says. “Intellectual curiosity is the first characteristic I look for.”
Sullivan says organizations should seek candidates who demonstrate an involvement in a wide array of projects, and a constant quest for new domain knowledge. Such a person is more likely to dig deeper into the data, and ask better questions, than someone with a more limited range of experiences.
On occasion, a preference for such experience might mean hiring a naturally curious candidate with limited technical skills. But Sullivan says that shouldn’t be a concern for most companies. “In most cases, machine learning, math and science are easier to teach,” he says. “They’re descriptive and deterministic. Plus, there’s usually documentation so you can look at the mathematical models.”
2. Find techies who also can communicate visually.
When assembling a data science team, it’s easy to fall into the trap of focusing exclusively on a candidate’s knowledge of programming languages. After all, says Sullivan, “An understanding of software languages or scripting languages like Pig or R, and being able to do basic scripting like Java,” is critical to data analysis.
But Sullivan recommends giving equal consideration to “technical skills around visualization.” For example, he says visualization experts know when a chord diagram, which shows the relationship among a group of data sets, can yield greater insights than displaying data in a time series, in which data points are plotted along line charts. “Being able to express ideas about how a business use can best consume the output of data analysis is pretty vital,” says Sullivan.
3. Seek out storytellers.
Chord diagrams and time series are essential to visualizing data. But communicating the significance of a company’s data to business leaders requires “the art of storytelling,” according to Sullivan. It’s a difficult task that Sullivan says is best reserved for a data science team’s more business-minded head honcho. “The leader of the data science team is usually the expert storyteller who can put [the data] in human terms,” he says.
4. Look for domain expertise in your industry.
According to Sullivan, it’s not technical wizards but domain experts with versatile skills sets who play a pivotal role in translating a company’s bits and bytes into actionable intelligence. Expertise may range from a deep understanding of the finance industry to years spent working in the healthcare arena. Either way, Sullivan says, “Domain expertise is the connective tissue that makes sure a data science team’s technical people are able to find insights that will actually make a difference in the operation of a business.”
Yet the question remains: what qualifies a team member as an actual domain expert? Opinions vary but Sullivan says as it can take as little as six months for those in areas such as the airline industry, and as long as 10 years in more specialized fields, such as geneticists and medical doctors, before someone can “fully understand a domain.”
Regardless, Sullivan says it’s worth the wait. “Domain experts are the most valuable people because they add the perspective of reality that you may not have if you’re just crunching numbers.”
5. Keep top talent in steady rotation.
To “continually refresh a data science team’s perspective,” Sullivan recommends rotating a team’s domain specialists from inside the data science team to the organization’s multiple lines of business every six to nine months.
By moving them in and out of a data science team, and throughout the company, Sullivan says that domain experts gain a stronger understanding of the impact of actionable insights on a company’s day-to-day decision-making, and end up serving as “evangelists for the data science team.”
However, by limiting the rotation cycle to six to nine months, Sullivan says “it gives them just enough time to understand the data science trade craft and gel with a team, but it’s not so long that they become removed and can’t go back to where they came from.” Proper “orientation and onboarding” is also key to ensuring that domain experts don’t bristle when asked to lend their expertise to a new data science team.
6. Cultivate a touch of conflict.
It’s not uncommon for domain experts and technical gurus to lock horns on what a particular dataset means for the business. Rather than encourage harmony, Sullivan says the best data science teams embrace conflict and see it as a possible source of surprising insights.
“Some of the biggest breakthroughs have involved a team of domain experts who believed certain things could not be done and statisticians who believed they could figure it out,” says Sullivan.
For example, Booz Allen worked with a large pharmaceutical client where the domain experts believed that certain types of analysis were simply not possible while the statisticians “believed with so much conviction that they could figure it out. There wasn’t harmony amongst the team members but that really motivated everyone and really changed the conversation in the organization,” says Sullivan. “The statisticians ended up winning and achieved some very advanced analysis to speed drug discovery but it took months of engineering effort to prove the domain experts wrong.”
Cindy Waxer, a contributing editor who covers workforce analytics and other topics for Data Informed, is a Toronto-based freelance journalist and a contributor to publications including The Economist and MIT Technology Review. She can be reached at firstname.lastname@example.org or via Twitter: @Cwaxer.