What’s the No. 1 challenge organizations face on predictive analytics deployments? According to industry analysts, it isn’t developing mathematically-intensive predictive models or mastering the still-evolving analytics tools that run the models. Instead, they say, the greatest impediment to a successful predictive analytics project is creating a culture that values data-driven decision making and encourages people to have faith in the findings generated by models.
Call it a lack of trust or a gap in understanding -- either way, the fact is that the statistical sciences associated with predictive modeling aren’t readily familiar to the average business professional. As a result, business managers and operational workers don’t necessarily buy into the premise that an arcane math formula will churn out tangible insights capable of delivering real business value, like increased sales or improved customer service.
“Many of these models are not things that can be dissected or communicated -- they’re not always cut-and-dried,” said David Menninger, a vice president and research director at Ventana Research in San Ramon, Calif. “The fundamental problem relates to the fact that most people don’t understand the math that goes into creating the predictive models. So when the model makes certain recommendations, people don’t intuitively understand it.”
The first step toward fostering organizational faith in a predictive analytics initiative, according to consultants and analytics professionals, is to have a clear idea of the program’s business goals from the very start. That means not blindly chasing predictive analytics because it’s a hyped technology but identifying specific business problems that you can have real impact on by using the software.
“Diving into a business problem is the most important part of it,” said John Elder, CEO of Elder Research Inc., a data mining and predictive analytics consulting firm in Charlottesville, Va. “If you don’t, you could be shooting a powerful gun at the wrong target or using the wrong metric, and you’ll get the wrong results.”
Predictive analytics project need: Friends in high places
Having a project champion high enough in the organization to protect the initiative during its incubation period and through the ups and downs of a typical deployment is also essential to winning over the business community.
At Paychex Inc., which provides human resources and payroll outsourcing services to corporate clients, the initial strategy on a predictive analytics project launched five years ago was to work at the executive level to identify key questions for the analytics program to address and then to build predictive models that would resonate strongly with the Rochester, N.Y.-based company’s management team.
“Instead of building cool risk or fraud models around our agenda, we started out building models that would affect top-line revenue or client retention so that various senior business hats would embrace them,” explained Frank Fiorille, director of the company’s risk management group, which is in charge of advanced analytics at Paychex.
At the same time, the predictive analytics team consulted with business managers and workers to gather their input -- a tactic that gave business users a stake in the game while easing their concerns about some of the more intimidating aspects of modeling and statistical analysis, said Erika McBride, the company’s manager of modeling and risk review.
Focusing the analytics initiative on activities that could garner “small wins” and then communicating those successes to the business was another key part of the deployment plan, Fiorille noted. “If you have six or seven modeling projects, choose the one that has the highest probability of working first,” he said. “It buys you capital, and pretty soon, you have people in the business asking you to build more models.”
Predictive analytics vs. gut instincts
One of the approaches used by the predictive analytics team at Catalina Marketing Inc. to build trust in the technology there was testing predictive models with business users and doing with-and-without comparisons on data sets. That enabled the users to see how the predictive insights translated to better business decisions than relying on gut instincts did, said Ryan Carr, vice president of global modeling and analytics at the St. Petersburg, Fla.-based company.
“You have to compare and contrast the two,” Carr said. “Usually the story isn’t that their gut instinct is completely wrong -- maybe their gut instinct is four times better than average. But we can show that we can make them a little smarter with modeling.” Six years after the launch of the predictive analytics program, his team is building about 1,000 models annually for various departments at Catalina, which uses a data warehouse with information on consumer purchasing behavior to provide targeted marketing and advertising services to retailers and manufacturers.
Carr and others caution, however, not to lose sight of setting the appropriate expectations for a predictive analytics project. Once on board with predictive analytics, they said, many business users mistake the software for a magical black box that can be counted on to quickly help jack up sales or improve customer relationships.
“Managing expectations is key in all of this,” advised Rick Sherman, managing partner at Athena IT Solutions, a data warehousing and business intelligence consulting firm in Stow, Mass. “People need to understand that the tool itself doesn’t know about your business and what you’re trying to do.” For example, a retailer couldn’t just deploy predictive analytics software “and figure out how to sell stationery better,” Sherman said. “You have to develop and refine the models. It’s an ongoing process.”
ABOUT THE AUTHOR
Beth Stackpole is a freelance writer who has been covering the intersection of technology and business for 25-plus years for a variety of trade and business publications and websites.