Catalysis, by Matthew Schall, August 28, 2012
In a rare interview with Catalysis, Gregory Piatetsky-Shapiro, editor and chief scientist of the world's foremost data mining information site, KDnuggets, shared his views on the future of data mining. Now that data mining (also called predictive analytics and data science) is becoming mainstream, we asked Gregory about its best uses and its future.
Q: What are some of the greatest benefits companies can expect to derive from investing in data mining, and how will those benefits evolve in the next five to ten years?
Gregory Piatetsky-Shapiro: One of the main applications of data mining is to understand and predict customer behavior, and one of the top applications is reducing churn. A typical telecom like Verizon may have churn rates of 2% a month, meaning that out of 1,000,000 customers 20,000 will churn each month. Data mining cannot predict all the customers who will churn, but it can narrow the focus to perhaps the top 100,000 likely churners, where the churn rate is not 2% but 14% - seven times higher. With a smaller target group, messaging to reduce churn becomes cost effective, where targeting all of the customers would be prohibitively expensive. The typical analyses are done over time using methods like neural networks or support vector machines to identify patterns of behavior that are more likely to lead to churn. If we catch churners before they make that decision, we have a better chance of retaining them. This is a good example of the promise of data mining: we increase accuracy and ROI.
I also see great benefits from segmentation and prediction analyses, which can be used to increase customer loyalty. Both analyses may be used to understand what is important to each individual customer. When we understand what drives customer behaviors, we can identify what products will appeal to them and increase the number of products bought, which increases wallet share and profitability.
Data mining tools combined with simulations are also a great way to create and understand the value of new markets, and understand marketplace trends.
Finally, because data mining methods can be tuned to find rare events, I expect a continued increase in the use of data mining to identify and reduce fraud.