Focus on the business impacts of analytics, not methodologies or insights…. keep the business involved in all iterations of an analytics process or project…. and remember, it’s ultimately skilled people that drive successful analytics.
These “secrets” of leading corporate analytical gurus may seem like everyday common sense, but rarely are put into practice. To find out what it takes for organizations to succeed with data analytics, Wayne Eckerson, founder of the BI Leadership Forum, recently spoke with seven exemplary analytics leaders, and distilled their advice in his latest book, Secrets of Analytical Leaders: Insights from Information Insiders.
Advice ranged from approaches to developing analytical models to how to achieve faster delivery of analytics to data quality. The goal of analytics is to positively impact the business – not to create elegant statistical models or penetrating insights, says Ken Rudin, head of analytics at Facebook. “Focus on questions, not answers,” he says. “Most technologies fail to meet business needs because they focus on getting the right answers, but not asking the right questions. Getting answers has become easy, but the real value comes from getting the business to ask the right questions first.”
Sometimes, Rudin adds, “you have to really wrestle with what drives performance. If you can’t come up with anything, then maybe the project isn’t worth doing because you cant influence the outcome. Focusing on impacts instead of insights is a great discipline… I prefer a trivial analysis dome on a napkin that changes behavior to the most brilliant mathematical model that has no business impact.”
While the technology is now available to companies of all sizes for analytics, making the data valuable to the business still requires skilled and knowledgeable analysts, says Kurt Thearling, director of IT and applied analytics at AlixPartners, LLC. “A few years ago, I attended a data mining vendor’s customer advisory board meeting and heard a number of board members say that they were going to make significant jumps in their analytical sophistication in the next two years. A lot of their confidence came from the fact that the marketplace would provide them with the tools, people, and data they needed. I worked at Capital One at the time, and it dawned on me that the competitive advantage we enjoyed from building an analytic capability from scratch at a sizable cost was now threatened. Any company could purchase an analytic capability. Although you can buy analytics, you can’t buy knowledge. A good statistician can be worth millions of dollars per year. That’s the additional value they generate compared to out-of-the-box analytical models.”
Keeping analytic engagement focused on the business was a common thread that ran throughout the comments of the analytical leaders Erickson spoke with. “Despite all the science involved in analytics, judgment and intuition still play important roles,” says Eric Colson, chief analytics officer for an internet startup. “Data scientists need to frame the business problem, choose the right modeling techniques – decision trees, clustering, regression, and so on – select the right data, define desired outputs and actions.”
The challenge is to deliver insights that are not only of high value for the business, but deliver them rapidly as well. This is where the “the principle of proximity” comes in, as related by Dan Ingle, VP of vehicle valuations at Kelly Blue Book. “To deliver great applications, seat your data developers alongside your business users… all the time,” he says. Ingle, for one, is a great believer in the “Scrum” methodology – in which, unlike the traditional “waterfall” approach, where all solutions are delivered all at once at the tail end of a project, solutions are built collaboratively and in constant iterations.
“Many of our Scrum efforts today focus on data acquisition, either acquiring data sources of managing changes to existing ones,” he explains. “For instance, one of our major providers of auction data recently changed its data feed, giving us more granular data. So the data analysts on that data acquisition team worked with statisticians to generate a series of stories related to the changes. The data analysts defined what the new data elements would mean, while the statisticians assessed the value of those elements for their statistical models. Both participated in user acceptance testing as part of the Scrum process and verified that the new data acquisition jobs delivered the proper data.”
Collaborative approaches to analytics such as Scrum only succeeds when it’s rolled out to everybody, “including executives, department heads, analysts and developers,” Ingle adds. “You have to give executives full disclosure about what is happening, failures and all.”
Analytics executives were also asked to make predictions on the future of their craft. As Colson put it: “We’ll start to see analytics that extends beyond the walls of the enterprise. The current approach is to take external data and make it internal. In the future, I suspect we will consider prominent reference data as a virtual extension to our data warehouses. In this way, data can be linked and shared across organizations and applications.”