Jack Noonan Explains How Predictive Analytics Delivers Customer Insight
Defying the Limits: Businesses have gathered lots of data from online operations. Are there emerging ways to put that data to better use?
Jack Noonan: Yes. Most organizations are data-rich but information-poor, particularly with regard to Web data. Transforming online operations data into valuable customer intelligence requires more than the simple statistics that most organizations are using today.
Organizations that are committed to maximizing their Web investments are turning to solutions that combine the predictive power of data mining with sophisticated Web analytics. Solutions of this type provide insights into both the historical and future behavior of customers, both as individuals and as aggregates. By using predictive Web analytics, organizations can automatically discover visitor segments, detect the most significant sequences through the site, understand content and product affinities, and predict likelihood to respond, buy, or leave.
The data-gathering power of the Web requires powerful analytical technologies such as predictive Web analytics to bridge the gap between data and action. Only in this way only can they effectively manage their Web presence and maximize returns on their online investments.
DTL: It seems like predictive analytics is moving down from the cadre of statistical analysts to mainstream users. Is that accurate and, if so, how is that likely to affect CRM solutions?
JN: This is the trend. For several years, predictive analytics has been applied to an expanding number of operational tasks. What was cutting-edge five years ago has now been "operationalized" in systems designed for campaign optimization, customer retention, customer segment migration, and fraud detection, as well as in recommendation engines.
Many systems that previously operationalized the management of these activities now include prediction to help accelerate tasks, provide new insight, reduce costs, and optimize detection. This trend will continue as additional customer insight is interlaced with historical patterns. This integration is happening now with the addition of survey data and real-time response data from call centers; text-mining information is also finding its position in the rich set of operational data that is in place.
Although the statistical analyst can still play a major role in advanced discovery and modeling, the difference today is that application software and new computing architectures allow for the augmentation and deployment of predictive analysis in the workflow of mainstream users, making it less intimidating and, in some cases, totally transparent to the users of today's CRM systems.
DTL: There are recent concerns about privacy and data mining. What's your take on that?
JN: Personal privacy is not an entirely new issue in the analysis of people data, but it undoubtedly has taken on greater significance in the United States during the past two years with the proposed application of data mining to counter terrorism. Naturally, there are valid concerns on all sides of this issue. We need effective tools in law enforcement, but we also need to protect the personal information people disclose in the course of their lives.
Technology can provide solutions that balance these needs, though recent legislative attempts to complement technology have great merit as well. Most of these are aimed at regulating the collection and integration of personal data (a matter of data protection), although some do so under the banner of regulating "data mining," which is a particular analytic technique. The European Union has given us an exemplary data-protection model, and we might do well in the U.S. to take the a cue from the EU and focus on expressly how data are collected, transferred, and disclosed, rather than restricting how we derive knowledge from those data correctly. The customer will then perceive your communications as adding value or empowerment.
DTL: Predictive analytics is best suited for business-to-consumer transactions, as opposed to business-to-business, right?
JN: Actually, predictive analytics, which covers a wide range of techniques, can add value to both types of transactions. Different techniques are used with different types of data. In business-to-consumer transactions, companies are predicting the behavior of individuals using techniques such as decision trees and segmentation. These techniques build mathematical models that match similar attributes, such as the car an individual drives, with anticipated behavior. These are the types of predictive analytics increasingly being applied in CRM systems.
On the other hand, in business-to-business transactions, there is another set of predictive analytical techniques based on time-series data. Product sales information is a good example of this. Each day, there is a new value for the sales of a particular product. Given enough days, a pattern can be mathematically modeled for product sales, and then future sales predicted. The techniques used can be as simple as fitting the data to a pre-defined curve. These are the types of techniques that are increasingly being integrated into supply chain management systems.
DTL: How close are we to a predictive analytics solution that can link up to a CRM suite?
JN: Although it is relatively new to the market, this type of solution currently exists. Today, there are predictive analytic applications that seamlessly integrate with existing CRM systems. These solutions enable organizations to maximize the value of their existing CRM systems by using CRM data to predict key behaviors, such as the likelihood of a prospect to respond, the potential of a Web site visitor to purchase, and the probability of an existing customer to leave, so that action can be taken to positively affect that behavior.
What's important is that these applications provide their powerful predictive analytic technologies within a user interface that makes the technology all but invisible to the end user. These solutions can be used by front-line employees such as marketers, customer service representatives, salespeople, and other business professionals to make decisions that directly affect organizations' revenue and expenses.
DTL: Predictive analytics has been around for 35 years. What's new in predictive analytics that will help companies retain the "right" customers?
JN: Some of the core technologies, such as statistical techniques and algorithms from artificial intelligence, have indeed been in existence for many years. Two things have changed, however, that make them more applicable today to problems such as customer retention. One is that the techniques have been enhanced to be richer, more powerful, and more scalable to the enterprise - something that is essential for handling data for large customer bases.
The other is that predictive analytics, as defined today, is not simply about building predictive models. It covers all aspects of accessing, integrating, and preparing relevant data and, crucially, the deployment of predictive results to operational systems. Only when results drive actions can a true ROI be generated.
For example, in customer retention applications, predictive scores for propensity to leave and measures or predictions of current or future profitability are used to ensure that the "right" customers are targeted - not just the high-risk ones. In these instances, predictive analytics is used to drive the selection of lists for outbound campaigns and to personalize interactions on inbound channels such as call centers or the Web to encourage high-risk, high-value customers to remain loyal.
