ALEXANDRIA, Va. – Like many advanced and emerging technologies, predictive analytics software has a certain degree...
of coolness associated with it. When corporate and business executives see that the technology can accurately predict which customers are likely to buy what products, they get excited.
But what good is a prediction if companies don’t do anything with the insight? Not much, according to Dr. Eric Siegel, president of consulting firm Prediction Impact Inc. and chairman of this year’s Predictive Analytics World conference.
As predictive analytics starts to gain more traction and deployments increase, the technology must be used to tie insight to action to be truly effective, Siegel said. Companies must devise business rules that trigger specific actions when predictions are made, he added.
Insurance companies, an early adopter of predictive analytics technology, are a good example of this, Siegel noted. Insurers use predictive analytics software to determine the riskiness of taking on a particular customer, he said. The potential risk is then tied directly to the price of the insurance policy being offered to that customer.
At a retail organization, connecting predictive analytics to action could mean triggering marketing campaigns based on a customer’s likeliness to purchase a certain item or service, Siegel said. At financial services companies, the technology could be used to identify potential fraud and then prompt an audit.
Whatever the industry, predictive analytics software used in isolation doesn’t do anybody much good. But that’s not all that companies considering predictive analytics projects need to keep in mind, according to other speakers at the conference.
Predictive analytics demands significant data prep work, user buy-in
There is significant prep work that must go into a successful predictive analytics initiative, said Paul Coleman, director of marketing statistics at retail giant Macy’s Inc. He estimates that getting data prepped before even applying predictive analytics technology to it is about 80% of the job.
“Building [predictive data] models is at least as complex as your business,” Coleman told attendees. And, he cautioned, “the models are only as good as the data” that goes into them.
Jean Paul Isson agreed. Isson, vice president of global business intelligence and predictive analytics at Monster Worldwide Inc., said data governance and data quality are key to successful predictive analytics projects.
At Monster, for example, company executives first had to decide on the definition of “customer,” Isson said. Initially, they came up with seven possible definitions. Not until they agreed on a single one could the provider of online job listings and career management services move forward with predictive analytics, he added.
Isson also said that internal change management is important when deploying predictive analytics technology. Most predictive analytics initiatives fail not because of faulty predictive data models but from a lack of executive buy-in and poor end-user training, he said.
Marketing executives who are hitting their numbers will likely be reluctant to adopt a new technology such as predictive analytics, Isson said. As a result, he advised, it's important to show them how the technology can improve their success rates and then train them on how best to use the associated tools.
Jason Fox, an information system and portfolio manager in Paychex Inc.’s enterprise risk management division, told attendees that finding subject matter experts from business operations was crucial to the company’s predictive analytics initiatives.
“We identified subject matter experts to ensure that business conditions were met,” Fox said. He also sought out champions of the technology in Paychex’s sales department – people who could tout the benefits of the predictive analytics software to their colleagues and help boost end-user adoption.
Technical obstacles to predictive analytics success
There are also technical factors to consider, Coleman said. Data contained in flat files, for example, is relatively simple to model but then difficult to change, he warned. Data in relational databases, on the other hand, is more flexible to work with but can be limited by data volume constraints, according to Coleman.
Companies should consider the type of data that they plan to exploit and how it's stored before starting a predictive analytics initiative, he recommended. Those factors might also play a role in determining the type of workers that a company hires to oversee its deployment and use of predictive analytics software.
In the end, however, all of the required efforts are worth it because of the business insights that can be gained through the use of predictive analytics tools, the conference speakers agreed.
“Inside this data, there’s a customer in there someplace,” Cole said.
Correction: This article was updated to correct a quote by Eric Siegel.