Solving Business Problems with Predictive Analytics Collaboration
To get the most from predictive analytics, encourage collaboration between data and business departments and cultivate hybrid employees who are part analyst, part decision maker.
By Joseph Pigato, Managing Director, Sparked
While World War II battles were being fought in the skies, American mathematician Norbert Wiener was on the ground determining how anti-aircraft gunners could anticipate the future flight path of German planes. His prediction equation determined that for a fired shell to reach its target, gunners needed to know where the plane would be 20 seconds ahead of time.
Once the domain of statisticians, data miners, and mathematicians such as Wiener, predictive learning now includes a new cast of central characters: business managers and analysts. In fact, a recent poll [need citation] of companies using predictive analytics indicated that business analysts are more likely to be involved with predictive analytics tools than statisticians and data scientists. Further, the most commonly cited skills necessary to perform predictive analytics were "knowledge of the business" and "critical thinking."
New predictive analytics software is the root of this shift. It now handles much of the science of predictive analytics, producing powerful but user-friendly tools that business leaders can manage with less input from colleagues skilled in analytics. Today, the most common uses of predictive analytics are direct marketing, cross-selling, and customer retention. Use for risk analysis, fraud detection, quality assurance, and econometric forecasting are expected to grow rapidly over the next three years.
Although predictive analytics is still evolving, companies using the technology face two main challenges today: lack of skilled personnel and inexperience with predictive analytics technology
How can data analysts and business managers work together to solve business problems by leveraging predictive analytics? Here are my suggestions.
Data collection. Without a doubt, data collection is a formidable challenge. Data is typically disorganized, decentralized, and not originally created for optimal analysis (because those who structure the data are not the same people who analyze it). As more holistic deployments are made, people in these roles will work closely, early in the data collection process to ensure optimal data structure prior to analysis.
Data sources. Predictive analytics still relies mostly on structured data, which is easier to assemble from different departments and data sources. However, data sources are becoming more diverse, and use of unstructured data is increasing. For example, enterprises are starting to combine internal text data (from e-mail, call center logs, etc.) and social media text data with structured data to give greater insights as to why things happen beyond simply identifying correlations. Several other data sources are quickly becoming more prominent as well, specifically Web log data, geospatial data, and clickstream data.
Politics. Predictive analytics isn't just about data -- it's about the data owners. They sit in department silos, unaware of what data others collect, and collecting their own data in an idiosyncratic way. Politicking is the business manager's data collection task; data analysts rely on them to enable important data sources to add to their models.
Model management. Many people on the business side have read about the magic of predictive analytics and think you just "turn it on" and watch perfect, actionable insights flow in. In reality, it's a constant, iterative process, where data and business people must work closely together. Predictive analytics is meant to find correlations, not causal relationships. Business people can make inferences about data findings in the context of their business problems to spot such relationships.
Disciplined action. Companies often go down rabbit holes, focusing on discrete data findings and reacting to them myopically. Data analysts too often focus on building better models that become too complicated, when they need better and more data combined with a simple but effective algorithm that allows the data to unveil itself. On the other side, business people rush the modeling process. They try to steer models toward confirming their own premise, or toward results that will make them look good. They're eager to make conclusions when models are still being massaged and different data sources aren't providing stable findings.
For enterprises, predictive analytics translates to dollar signs by helping boost sales, customer retention, marketing effectiveness, and everything that touches the top and bottom line. This journey has been and will continue to be marked by brilliant improvements and difficult growing pains.
The infamous example of Target predicting a teen's pregnancy has an interesting epilogue. Indeed, to the teen's father's chagrin, the Target data folks created a brilliant mode for such insights. When the business folks stepped in, they decided to send relevant coupons when analytics gave them such insights, but to mix these in with other coupons so it wasn't so obvious. The father will tell you they weren't subtle enough.
Ultimately, I believe the companies that take the greatest advantage of predictive analytics will be those that encourage collaboration between data and business departments and cultivate hybrid employees who are part-analyst, part-decision maker.
Joseph Pigato is the managing director of Sparked, which helps companies retain their customers through sophisticated predictive analytics and engagement tools. You can contact the author at firstname.lastname@example.org.