Predictive analytics: Seeing beyond the horizon

Predictive analytics uses advanced statistical or machine learning techniques to forecast outcomes and trends. For companies, this fast growing field of business intelligence promises to harness Big Data to devise more effective strategy. But what is the best way to integrate these new tools and models into organizations? What kinds of governance changes are needed in businesses to apply predictive analytics effectively? Michael Goul, chair of the W. P. Carey School’s Department of Information Systems, has explored these and related questions in his research, which has taken him inside some of the largest and most successful U.S. companies. He has identified some of the challenges firms have encountered, as well as the best practices that have emerged as the application of predictive analytics has spread across businesses. A special working group of the Society for Information Management asked Goul to present an overview of his research at the society’s annual meeting in Chicago in early May. The working group, called the Advanced Practices Council, is made up of several dozen chief information officers of major firms. “These companies are just starting to run into issues, and they wanted to know what is happening in the field and what things are on the horizon” Goul said. “When new technologies come along, you have to stop and look at what impact they will have on the traditional ways that things have been done.” New models for business Although predictive analytics has gotten a great deal of attention in recent years, the phenomenon actually has been around for some time. For many years, direct mail marketers have used simple models to determine which potential customers should be sent catalogues. The FICO score, which assesses credit-worthiness of borrowers, is used by lenders to determine who is likely to default. “Businesses can score people and their likelihood to buy certain things,” Goul said. “At the university, we use models to predict graduation rates and to identify attributes associated with graduation. Companies are starting to use predictive models in many different ways. “ The emergence of Big Data has provided the raw material that gives predictive models much of their newfound potency. Torrents of data produced by websites, social media and mobile technologies are now available for use in these models. When a potential customer walks near a store or restaurant, predictive analytics makes it possible for a smartphone app to offer coupons or discounts targeted to the individual customer. “These models are proliferating, and they are being embedded in the computer programs that interact with people,” Goul said. “They can personalize the experience for the individual. They also make good business sense for companies.” From witches’ brew to essential business tool Predictive analytics has made great advances in just the last seven years, according to Goul. In his presentation to the executives from the Society for Information Management, he depicted predictive analytics in 2007 by discussing the witches in the opening scene of William Shakespeare's play "Macbeth," stirring a cauldron and making predictions about the future. “When you move on seven years later, we now have a lot more tried and true methods for predictive analytics,” Goul said. “We no longer are like witches stirring a magic brew. We have people who are trained to do this kind of work.” Companies also have learned from their experiences with predictive models, according to Goul. He cites the evolution of eHarmony, a company he has studied. Predicting customer behavior is not just a tool for eHarmony. It is the essence of the company’s business. To match people who join the site, eHarmony first built a model that determined compatibility. “They found that this wasn’t enough,” Goul said. “Being compatible didn’t mean that people would actually get together. So they developed another model, which tested whether people would be attracted to each other.” The model for attractiveness considered such factors as the poses people used for the pictures they posted on the site and whether the photos were taken indoors or outdoors. The company has strived to integrate the two models, according to Goul. “The models work in tandem, so it’s important that interactions be evaluated in order to assess success.” How to deploy predictive models Companies are learning about the complications that arise when deploying predictive models, according to Goul. “Most everybody is focused on how to build a better model but not necessarily how to deploy it,” he said. Goul found that the executives he met with at the Society for Information Management conference, most of them CIOs of their companies, are very interested in how predictive analytics is deployed. “These are the people who have to make it happen,” he said. Goul explained to the executives a methodology that he and Sule Balkan, a former professor of Information Systems at the W. P. Carey School, developed for analyzing the deployment of predictive analytics in companies. The researchers call the methodology DEEPER, an acronym for design, embed, empower, performance measurement, evaluate and retarget. The DEEPER approach suggests that organizations should design a campaign for deploying predictive analytics, embed the model into their business processes, empower employees to participate in the campaign, measure performance, evaluate the results and re-target future efforts. When companies seek to deploy predictive analytics, they need to understand clearly the complexity of both the external environment and the analytics they are seeking to implement, according to Goul. These two areas of complexity will determine the optimal governance mechanisms for the process, he said. “If your environment isn’t very complicated, it means one thing. I illustrate this with a man wearing a helmet and walking a tightrope that is only six inches off the ground. You can contrast that with a man walking a tightrope across the Grand Canyon with no net. The latter would be a complex external environment in which there are major implications for failure,” Goul said The complexity of the DEEPER process also is important for companies to understand, according to Goul. In the most complex case, an organization will deploy multiple predictive models across units, involving many employees. When complexity is high in both the environment and the DEEPER process, companies have to be ready to react quickly, according to Goul. “You have to be prepared for disasters to happen if one model runs into another or decays. You need need to be able to detect this and respond as soon as you can in order to ward off problems,” Goul said. A guide to action As part of his presentation to the CIOs at the recent conference, Goul prepared a practical guide for governance initiatives that organizations should consider depending on where they fall on the complexity scale for both outside environment and DEEPER. A company with low complexity in both environment and DEEPER should lay the groundwork for future initiatives by encouraging pilot studies and experimentation, and by developing job descriptions for prospective new hires, according to Goul. These companies are just getting their hands dirty with predictive analytics. For a company that has low complexity on the DEEPER scale but high complexity in its environment, Goul suggests actions that build support within the organization for predictive analytics initiatives. Among the activities he suggests are encouraging close collaboration between data scientists and business initiative leaders, establishing project management protocols and developing strategies for dealing with shared infrastructure investment pressures. An organization that operates under conditions of high DEEPER complexity but low environmental complexity should take steps to manage risk, according to Goul. He suggests that they require regular reporting of model performance, establish a predictive model asset inventory and segment leadership responsibility. For a company that falls into the high complexity categories in both environment and DEEPER, Goul suggests a series of activities designed to amplify governance. Among the actions he recommends are repeated testing and monitoring of models, frequent audits, establishment of strict standards and policies and preparation of manual backup plans. Goul advised the executives to assess their current IT governance and to adapt the DEEPER framework for their own organizations. He also advised them to determine where their organizations fit on the environmental and DEEPER complexity scales and to consider taking actions appropriate to their circumstances. Goul plans to continue this stream of research. The working group of the Society for Information Management has already invited him to present his findings at the organization’s 2015 conference. “I think I need to continue to work on the DEEPER methodology and to standardize it more to make it more useful,” he said. “I also want to look at what happens when companies move through the different stages of complexity. I want to find out from companies what works the best.” Bottom line: • Predictive analytics is a fast growing field that uses advanced statistical methods and models to make forecasts about outcomes and trends. • Although many companies are developing predictive models, fewer of them are considering how best to integrate these models into their businesses. • Department of Information Systems Chair Michael Goul has been researching how businesses deploy predictive analytics. He has developed a framework that identifies best practices for introducing predictive models to companies. • The complexity of the outside environment and the complexity of conditions inside a company when it is deploying predictive analytics determine what strategies will be most effective, Goul found in his research.