Predictive analytics is hard to do; just ask a data scientist

Where do you rank in using predictive analytics? New research and two Ivy League data scientists suggest that if you do it all, you're way ahead.

Despite the hype and hopes surrounding the use of predictive analytics, a relatively small number of enterprises consider it a critical element of their business intelligence (BI) strategies today, new research shows.

Only 13% of businesses say they use predictive analytics, ranking it dead last among a group of 10 BI capabilities, well behind spreadsheets (60% of businesses), BI (49%), analytics databases (41%) and custom-built systems (34%).

The first thing I did was pull the BI team out of IT, and they suddenly got productive and started actually thinking like scientists.

Worse news for CIOs? Traditional BI teams rank behind both data scientists and business analysts in their ability to deliver predictive analytics, according to a new study from Ventana Research Corp.

"Predictive analytics remain a specialist tool," said David Menninger, former research director at San Ramon, Calif.-based Ventana Research and author of its Business Analytics Benchmark. He presented the study's findings, which were based on a study of 2,600 organizations and the types of analytics they perform, at a recent Mass Technology Leadership Council (MassTLC) seminar, Data Science -- A Practitioner's Perspective.

"I have some theories as to why [analytics] is dead last. Personally, I think it is very hard, and the math involved is beyond the capabilities of many people or their training today," said Menninger, who recently left Ventana to head business development at EMC Corp.'s Greenplum data analytics division.

The study does suggest that the potential payoff of predictive analytics is high. Of the 13% of the study's respondents using predictive analytics capabilities, 80% ranked its capabilities as "important or very important" to their organizations. Not surprisingly, much of the work using predictive analytics homes in on revenue-related activity. Among the top five uses of predictive analytics, forecasting and marketing ranked first (72%), followed by marketing analyses (70%), customer service or support (45%), product recommendations (43%), and fraud detection (34%). Within a year, however, the largest growth area for predictive analytics will be around social network analysis, the study showed.

Big data and advanced analytics

The term predictive analytics refers to a loosely defined discipline that involves using statistical and other complex mathematical analyses of data to predict the future. The field goes beyond data mining, industry experts are quick to point out, and it's not about data mining for data mining's sake. Predictive analytics identifies patterns in the volumes of data generated by a business and by relevant external sources. Those patterns can point the way to probable outcomes, and that allows businesses to adjust their strategies.

Now that the volume, velocity and variety of data -- or big data -- are a reality for many businesses, the value of predictive analysis has appreciated. At the same time, it's become even more difficult for companies to obtain. That doesn't mean CIOs or their companies should count predictive analytics out, Menninger said, offering a prediction of his own.

"I think it's rising. And part of the reason I think it is rising is because of big data," Menninger said "I don't think you can look at those volumes of data and just browse through a billion records and find something of interest. You've got to use advanced analytics." Indeed, in another Ventana research study on technologies associated with big data, advanced analytics was the most common reason respondents gave for using Hadoop.

One potential worry for CIOs? When study respondents were asked who does the best job of using predictive analytics, the BI and data warehouse team came in last (59%). It was outdone by line-of-business analysts (65%) and specialized data scientists, statistical or data mining resources (70%). Strongly correlated to dissatisfaction, if not its cause, was the complaint (which often is lodged against BI programs in general) that data models are not updated often enough to be useful. Indeed, in the results, the satisfaction level associated with predictive analytics programs that updated models daily (81%) knocked the socks off the level associated with programs that were updated less frequently (48%).

That finding didn't surprise audience member Andy Maddocks, formerly head of strategy and IT for In Touch Ministries, a global charity based in Atlanta. "The first thing I did was pull the BI team out of IT, and they suddenly got productive and started actually thinking like scientists," said Maddocks, now president and chief product officer of R2integrated LLC, a digital marketing startup in Cambridge, Mass.

The view from real-life data scientists

One of the chief reasons companies struggle with advanced analytics is that a majority of organizations (57%) don't provide adequate training in the concepts and techniques involved in predictive analytics, the study showed.

Just how hard it is to become adept in this field -- or even decide which tools to use -- became apparent from comments during the MassTLC seminar by two scientists who specialize in big data analytics: Michael Kane, associate research scientist at Yale University's Yale Center for Analytical Sciences, and Ian Stokes-Rees, a researcher in Harvard Medical School's NEbioGrid organization and now at Harvard's School of Engineering. Both agreed that businesses' success in predictive analytics is limited by skills and the small number of available tools (Tableau, SPSS, and MATLAB and R computing languages notwithstanding).

More about predictive analytics

Analytics technology is racing past our ability to 'gestalt' it

Data silos in big data analytics: Now you see them, now you don't?

Business intelligence technology 3.0 is all about control, but whose?

Whether the BI team resides in or outside IT, businesses can't expect IT professionals focused on IT infrastructure to be able to jump into business informatics, Stokes-Rees said. "It's nuts. There is a high skills gap. You need to partner with data scientists," he said. The pressures faced by business to perform are not conductive to the exploratory nature of probing big data for predictive insights, he added.

Still, even these two data analytics experts working in academia find themselves up against the same kind of very unscientific demands often made by results- and gut-driven business people. Advanced analytics often is seen as simply as a quick way to add heft to a hunch. Said Kane, referring to the scientists he works with: "Basically, what they say to me is, 'Can you make math agree with me so I can publish this paper?'"

Let us know what you think about the story; email Linda Tucci, Senior News Writer.

Dig Deeper on Enterprise business intelligence software and big data