Gaining a Holistic View of the Customer

When examining the evolution of CRM analytics, I recall the transition of media technologies from radio to television to today's latest HDTV displays. Each of these technologies has a place in the market and excels in certain areas, but the greatest opportunities (and adoption rates) ultimately remain in the medium with the highest resolution and clarity. This is analogous to the CRM analytics technology where standard (yet mission-critical) business intelligence (BI) and reporting technologies are the platform of choice for many marketing analytics applications. Organizations seeking a clearer view of customer behavior turn to predictive analytics to explore more data attributes and provide better and more valuable insights. Finally, leading organizations--the equivalent of that neighbor of yours who just bought a 42" plasma HDTV--are using both text and numeric data in their predictive analytics to gain a crisp, clear view of their customers and future behavior. Major analytical techniques Most techniques for analytic CRM technologies are optimized for either text or numeric data, but not for both. The sophisticated enterprise is using both forms of data simultaneously to create predictive profiles that significantly outperform either source alone. The following highlights the enterprise analytics continuum and the advantages of a new, hybrid approach that combines text and numeric data. BI and reporting BI tools are excellent for exploratory analyses, business process management, and results tracking. Ideally suited for summarizing numeric data, these technologies typically take the form of online reports or dashboards that are customized to reflect the operations of an individual organization. Predictive analytics BI reports a tabulation comparing specified data attributes, while predictive modeling tests a particular hypothesis. For example, predictive modeling is frequently used to predict who among a prospect list is most likely to respond to a marketing offer. Traditionally, predictive modeling has been applied to the numeric data available throughout the organization, including financials, sales history, and transactional data. Text mining Text mining is typically used to search unstructured text documents for key words and phrases, and then to extract relevant information. This information might include the author, title, and date of publication, acronyms defined, or the articles mentioned in the bibliography. For business decision-making text mining can also be used to discover the tone (positive, negative, or neutral) from a range of unstructured data sources like customer comments or contact center interactions. In these cases text mining discerns such subtleties of language as:

indications of sentiment

key themes

specific hot buttons

Taking a hybrid approach A hybrid approach suggests that combining numeric data with text data provides a truer picture of customer behavior. The current availability of computing power allows organizations to combine the best of both worlds and create a more holistic view of the customer they are targeting. Without text data business decisions cannot incorporate the emotional aspects of the customer experience. Without numeric data business decisions do not adequately reflect the actual buying behavior of the customer. Combining text with quantitative data produces superior results. This hybrid approach has proven successful in many CRM, direct marketing, and credit- and risk-management situations. For example, a major airline looking to improving its customer experience needed a more efficient way to route customer comments. Previously, a live person was required to determine if a comment (from phone, comment card, or email inquiry) was a compliment or complaint. Additionally, the airline wanted to weigh routing priority based on the transaction histories. The solution was to deploy a genetic algorithm-based application to model a dataset combining the text information with the numeric data. In building the solution current customer data was parsed into more than 16,000 individual words. Preliminary research indicated that only 500 were important. Information about the 500 words was then integrated with frequent flyer data, flight information, and ticket information. The resulting solution predicted complaints with 93 percent accuracy. Since the scoring code was easily identified from the genetic algorithm, the solution was then available for real-time routing of calls and emails. Conclusion In the world of analytics (as in media technology), many different solutions will thrive in an environment with a diverse set of challenges. Hybrid approaches to analytics are becoming more common as organizations look for ways to gain better insight into the 80 percent of unstructured data that is embedded in CRM and other operational systems. About the Author Doug Newell is president of Genalytics, a leading provider of advanced business analytics software, in Newburyport, MA. With more than 22 years of experience and recognition as an innovator in the world of analytics, Newell provides the vision and direction that serves as the foundation for the company. A successful entrepreneur prior to starting Genalytics, Newell was a founder of Tessera Enterprise Systems, which was acquired by a leading systems integration firm in early 2000. Prior to founding Tessera he served as vice president of analytics for Epsilon Data Management. He can be reached at dnewell@genalytics.com. Related articles: Customer Analytics Tools Are Sprawling How Can Predictive Modeling Boost Sales Results? The biggest bang for a first-time analytical investment can come from a spending model. Driven with Business Expertise, Analytics Produces Actionable Predictions Run data mining as a business activity to generate customer predictions that will have a business impact.