BACKGROUND IMAGE: iSTOCK/GETTY IMAGES
The data scientists at Scotiabank, a financial services firm based in Toronto, aren't attached to any particular business units. Instead, they're part of a standalone team -- officially called the decision sciences group -- that does advanced analytics work for all of the bank's departments. But standalone doesn't mean stand apart: Andrew Storey, the bank's vice president of decision sciences, said he and other managers work to ensure that the data analytics projects taken on by the team aren't just abstract exercises with no practical value to business strategies and operations.
"Just because we can do something doesn't mean we should do it," Storey said during a session at the 2015 TDWI Executive Summit in Las Vegas. "What we're really doing is rooting ourselves in the business and supporting the projects they're working on. You can get detached from reality pretty quickly in a unit like this, so we're trying to keep ourselves grounded."
Storey does want his 30-plus analysts to think creatively as they run predictive analytics and data mining applications against sets of customer and pricing data in an effort to help optimize marketing campaigns, promotional offers and product pricing and identify financial connections between different customers. In fact, fostering a culture of innovation is one of his core tenets for managing a successful analytics team. "We should always be looking for a better way of doing things," he said, adding that empowering his workers to do so has helped in retaining them.
At the same time, though, Storey lets business managers at Scotiabank take the lead in deciding what his team should explore, or he works with them to figure that out. Analytical findings need to be embeddable in operational systems and processes -- a predictive model "is completely useless if we don't make decisions based on it," he said.
Explanations in order on data analytics
Team members also must be able to explain the analytics techniques and methodologies they're using to business execs to help get buy-in on making use of the results. And to try to streamline the analytics process, Storey steers the data scientists away from reinventing the predictive-modeling wheel. He encourages his team to reuse algorithms from other financial services firms and organizations in other industries, adapting them to the bank's needs.
Balancing acts such as the one at Scotiabank are a common element of managing analytics teams, and they're becoming more crucial -- and more challenging -- as big data analytics programs expand the scope of the work data scientists are doing and the types of information they're looking to analyze.
That's turning into a bigger issue as big data analytics projects become more prevalent, as survey results show. For example, in TechTarget's IT Priorities Survey for 2015, 25% of the 2,212 respondents worldwide said their organizations were planning big data analytics deployments this year, ranking it among the top five planned software initiatives (see chart). Meanwhile, 40% of 302 business and IT professionals surveyed in June 2014 by consultancy Gartner Inc. said their organizations had invested in big data technologies, up from 30% the year before; another 33% cited plans to do so within 24 months.
Collaborative and collegial approaches are a must in managing big data analytics efforts, said Mike Lampa, managing director of consultancy Archipelago Information Strategies. "I think the proper mindset is how do you guide the process, not control it." Lampa warned that skilled data scientists are likely to recoil -- and perhaps look for new jobs elsewhere -- if they think their work is being excessively controlled. He said managers should work with their teams to focus analytics work on useful projects and implement clear guidelines on using data and vetting analytical models -- then get out of the way.
Analysts get the keys to the data vault
That's the sort of approach Netflix has taken with its data science team. The Los Gatos, Calif., company uses a variety of systems running in the Amazon Web Services cloud -- including Hadoop, a Teradata data warehouse and Amazon's Redshift and Simple Storage Service technologies -- to store multiple petabytes of data for use in analyzing customer interactions with its online streaming-media service.
Kurt Brown, vice president of Netflix's data platform, said during a presentation at the Strata + Hadoop World 2015 conference in San Jose, Calif., that the data analysts are responsible for building their own queries, algorithms and models -- and that his goal is to enable them to do what they need to do on data analytics projects "with as little impediment as possible."
Brown's platform managers consult with the analysts and promote development best practices, but they don't put up any gates when it comes to doing the development work. That sometimes results in coding errors and data issues, but he said trying to keep bad code out of analytics systems is "a fool's errand" in a company like Netflix. After the fact, one of his staffers looks for code that needs to be cleaned up, then sends the information to the analysts who were responsible for the errors so they can make the fixes themselves. "It wouldn't scale for us to do it," Brown said. "It has to be a shared responsibility."
Get real-world advice and examples in our guide to managing Hadoop projects
Read tips from analytics managers on creating successful analytics processes
Consultant Rick Sherman offers a checklist of big data analytics to-do items