News Stay informed about the latest enterprise technology news and product updates.

Successful data science process not simple to set up, sustain

Data science teams face a mix of process and cultural challenges in organizations, according to experienced analytics managers who offer advice on how to overcome the hurdles.

Cisco set up a corporate data science team in 2013 to analyze data for business units and develop advanced analytics...

skills and best practices throughout the company. Four years and three revamps later, the team is finally where it wants to be -- almost.

The long and winding road traveled by the Cisco unit illustrates the challenges organizations face in creating an effective data science process that meets business needs for actionable information -- and a data-driven culture that's predisposed to take advantage of analytics info. Those hurdles typically are bigger than the technical ones companies encounter in deploying analytics systems and tools, according to speakers and other attendees at the 2017 TDWI Leadership Summit in Las Vegas.

At Cisco, "we've had a lot of bumps and bruises along the way," senior data scientist Anu Miller said during a session. The corporate team, originally called the analytics center of excellence, was meant to take a broader view of business issues than data analysis teams in the networking technology vendor's business units, with a focus on data science applications. But early on, the team got bogged down in conventional business intelligence and data visualization work, she said.

Looking to bring the team's data science efforts more to the fore, in 2014 Cisco put a new manager in charge and added more people with high-level analytics and technical skills. Those changes also gave the team leeway to start building an internal data science culture and community, Miller said, pointing to meetups, online forums, competitions and other activities for workers involved in analytics efforts.

Data science and business disconnects

By late 2015, though, it became apparent that the corporate unit's own analytics work wasn't connected enough to real business issues. "We were coming across as a little too academic and ivory tower, and people really couldn't relate to us," Miller said.

We were coming across as a little too academic and ivory tower, and people really couldn't relate to us.
Anu Millersenior data scientist, Cisco

So, the data science team changed course again, recasting itself as an analytics consulting operation there to support business units without bias "for our own agenda or any particular outcome" on data science projects, she said. The unit also started working to produce analytical findings in a single day, and changed its name to the predictive analytics and decision solutions team "to emphasize that we were practitioners with a business focus."

But even then, many business managers still had a nebulous view of the data science process, Miller added. To try to clear things up and also better meet business needs, the team last year adopted so-called design thinking concepts to focus discussions about projects on topics such as what goes into business decisions and how that info can help team members craft analytics algorithms.

After all the changes, the data science team, which includes 11 analysts, data scientists and data engineers, is close to delivering the combination of analytics speed and business value it has aspired to from the start, Miller said. Things haven't progressed in a straight line, she acknowledged, "but we keep learning."

Evolution of the data science process

LinkedIn Corp.'s data science team hasn't experienced as many twists and turns as Cisco's did, according to Yael Garten, the social networking company's director of data science. Even so, it took about two years for the data science process at LinkedIn to evolve into a fully solid state, Garten said during another conference session.

Like at Cisco, LinkedIn's data scientists -- who are assigned to and embedded in specific product teams -- were initially asked to do a lot of "very simple data pulls" that didn't make good use of their analytics skills, Garten said. Her team's analytics work was in danger of getting "bottlenecked" because of the number of such requests, she added.

To avoid that, LinkedIn deployed self-service analytics tools so business executives, product managers, developers and other end users could run their own queries. Garten said that in addition to freeing up the data scientists to do higher-level analytics, the move helped foster a more data-driven mindset at LinkedIn, which is based in Mountain View, Calif., and was acquired by Microsoft in December.

Embedding the data scientists in product teams helps to reinforce that culture, Garten said. In addition, all of the key participants in product development projects must agree on target metrics for analysis before new-feature tests are launched on LinkedIn's website. And the data science team prioritizes analytics speed in order to get feedback on tests to the product teams as quickly as possible.

In a series of roundtable discussions, conference attendees cited a variety of other roadblocks in data science and big data analytics initiatives, including concerns among business managers that data scientists running machine learning algorithms will take over the decision-making process. On the flip side, there's the specter of analytical models being ignored and left on the shelf by business units.

All talk and no action on analytics

"There's a lot of lip service paid to being a data-driven organization," said Mark Madsen, president of consultancy Third Nature Inc. in Portland, Ore. He recommended that analytics and data science managers get the ball rolling with business execs who really do want to use data analytics to drive decisions, and then turn those applications into showcases for what their teams can do.

At the same time, data scientists shouldn't get overly enamored with their analytics tools and methodologies, Madsen cautioned. As the experiences at Cisco show, data science work needs to focus on things that actually matter to business units, he said. He compared data scientists to carpenters in that regard; for both, the priority has to be building solid structures, Madsen said.

That process is just starting at Export Development Canada (EDC), the Canadian government's export credit agency. Ben Okudolo, IT director at the Ottawa-based agency, said his team is beginning to build out a big data platform based on Hadoop, Spark and other technologies to support advanced analytics applications.

EDC's leaders have approved the analytics strategy -- now comes the challenge of making it all work, Okudolo said. "We know what we want to do," he noted. "We're at the point now where we're trying to figure out how to make that happen."

Next Steps

More tips from analytics managers on creating data-driven organizations

A lack of data science skills remains a big challenge for analytics managers

Advice on how to get better business value from data science projects

Dig Deeper on Business intelligence team

PRO+

Content

Find more PRO+ content and other member only offers, here.

Join the conversation

1 comment

Send me notifications when other members comment.

By submitting you agree to receive email from TechTarget and its partners. If you reside outside of the United States, you consent to having your personal data transferred to and processed in the United States. Privacy

Please create a username to comment.

What are the biggest challenges you face in implementing data science and analytics processes?
Cancel

-ADS BY GOOGLE

SearchDataManagement

SearchAWS

SearchContentManagement

SearchCRM

SearchOracle

SearchSAP

SearchSQLServer

SearchSalesforce

Close