In the second article in our four-part series on setting up and managing data analytics programs, corporate users and industry analysts discuss how to structure and then support analytics teams.
Table of Contents
Data analytics team’s needs: a business home, leeway on tools and data
Getting ready for an advanced business analytics software project
Creating an advanced data analytics business culture: Tips and advice
If finding and hiring talented workers with analytics skills is the first step in establishing an advanced data analytics team, determining how to structure it in relation to your IT and business intelligence (BI) groups – and how much autonomy to give your analytics professionals – is step two.
Some companies, especially those with highly centralized corporate structures, may be tempted to place an analytics team under the purview of the IT department or a standalone BI unit such as a business intelligence competency center (BICC). But according to analytics experts, most successful analytics initiatives take a more decentralized approach.
Data analytics teams are usually organized by business function or placed directly within a business unit, said James Kobielus, an analyst at Forrester Research Inc. For example, an analytics team that focuses on customer churn and other marketing-related analysis might be part of the marketing department. Risk-focused data analysts are typically best suited to the finance department, Kobielus said.
That’s because developing, testing and maintaining complex analytical data models involves significant domain-specific business knowledge, a requirement that doesn’t lend itself to a centrally controlled analytics program, according to Kobielus. He noted that statisticians, quantitative analysts and other analytics professionals dig deep into targeted, often-thorny business problems.
Many data analysts “are artisans taking pride in their work and [are] not all that eager to give up that autonomy,” he said.
Jeff Jackovich, managing partner at Visibility Resource Group, a Boise, Idaho-based staffing agency that specializes in recruiting and placing SAS data analysts, said most of the companies he deals with are looking for analytics professionals with expertise in a particular business area.
For example, a large retailer with more than 1,000 stores has hired Jackovich to find merchandising analysts who can help determine how best to stock the company’s store shelves. “Most analytics teams are in specific departments” and are not housed under an umbrella BI or IT group, Jackovich said.
Maintain communication between data analytics team, BI staff
That’s not to say that advanced data analytics teams should be completely cut off from their BI and IT cousins, however.
At Royal Bank of Canada, the BI team and a pair of advanced analytics teams that RBC has set up are “close, but they're not side by side,” said Cathy Burrows, the Toronto-based bank’s director of marketing services. “They don't report into the same executive."
Burrows said that for the most part, RBC’s analytics teams are organized by the departments they support. Marketing data analysts focus strictly on marketing analytics, for example, while the risk management analytics team concerns itself strictly with that function.
But the two data analytics teams work with each other, and with the bank’s BI staff, to share best practices when applicable, Burrows said. For example, if marketing analysts create a data model that could also be useful for risk analysis, they might pass it on to the risk management analytics team. “They have to be able to communicate,” she said.
Analytics team preferences: choosing, experimenting with analytics tools
Another impediment to a centrally controlled data analytics structure is the variety of analytics tools that are available – and a desire on the part of data analysts to choose the ones they use themselves. While organizations with an advanced analytics program tend to have a default analytics software provider, many analysts have their own preferred tools and like to experiment with cutting-edge technologies, including open source data analytics tools.
And that experimentation should be encouraged, Kobielus said. He noted that the analytics software market is hardly commoditized, with different tools possessing different strengths. His advice: Data analysts should be allowed, within reason, to use the tools that they find to be best suited for the job at hand.
“There might be lots of circumstances why you don’t shove a particular tool down everybody’s throats,” Kobielus said. “There’s something to be said for letting a thousand flowers bloom.”
John Savage, head of strategic risk analysis at New York City-based insurance firm Chartis Inc., said his team has largely standardized on SAS Institute’s analytics applications. But from time to time, members of the analytics team also use various tools based on open source scripting languages.
For example, the team tried out some Java-based tools when it was looking to build fraud-detection data models. “As a group, we experiment with different technologies,” Savage said.
Letting the data analytics team play in ‘data sandboxes’
In addition, Savage gives his analysts a good deal of autonomy to play around in their own data sandboxes of sorts. While the analysts can’t experiment on live production data per se, they can copy it to their desktop systems and test analytical models and scenarios with client-based SAS apps, he said.
Bill Robinette, manager of business intelligence systems at Advance Auto Parts, is also looking at the data sandbox approach as a way to give analytics users at the Roanoke, Va.-based retailer more control over the data they need to look at. “We’re trying to figure out how to create that,” Robinette said after speaking at a recent event held in Cambridge, Mass., by The Data Warehousing Institute.
Part of the interest in sandboxes stems from data analysts asking for the ability to pull information out of Advance’s data warehouse for their own use. Letting them load data into segregated sections of the warehouse would be more secure, Robinette said. He added, though, that there are some hurdles to get over – for example, analysts setting up their own sandboxes “would have to be somewhat savvy with SQL to know how to combine data and do what they want to do.”
As a stop-gap measure, Robinette has given some analysts in Advance’s merchandising and marketing groups access to transaction-level data stored in the data warehouse. He said this at least lets them run queries against more detailed data than they typically would be able to access.
Robinette is also OK with data analysts using different analytics tools. IBM’s SPSS software is Advance’s standard analytics technology, but some analysts prefer SAS apps – and Excel use “is almost a given,” he said. “In analytics, I think it’s important to let people use tools they’re comfortable with.”
Still, an analytics team should occasionally evaluate the tools it is using and consider dropping those that have outlived their usefulness, Kobielus said. “Look for opportunities to prune the predictive modeling tool bush.”