Customer Data Mining and Predictive Modeling

Compared to traditional market segmentation,  our approach identifies specific combinations of  actions that have the greatest predicted impact on  customer purchasing behaviors in both the short  run and the longer term.

CLIENT-PROVEN ADVANCED ANALYTICS, DATA MINING & PREDICTIVE MODELING TOOLS FOR CUSTOMER BEHAVIORS

Cox Associates offers a uniquely powerful, client-proven set of advanced analytics, data mining, and predictive modeling tools for predicting and optimizing customer behaviors. Compared to traditional market segmentation, our approach identifies specific combinations of actions that have the greatest predicted impact on customer purchasing behaviors in both the short run and the longer term.

Business Simulation

Our Predictive Clustering technology has been included in business simulation models that  provide fully integrated models of marketing, engineering, and operations and their financial  impacts in new businesses. Such models include high-level models of subscriber purchasing  behavior; predicted effects of pricing, advertising, and competition on market share and average  revenue per subscriber; and relative costs and performance associated with different network  engineering choices and build schedules. Reduce by over 50% the time to create and run  financial and strategic scenarios. Improve planning validity and consistency. Help planners to  focus on what can be controlled instead of on what has been assumed.

Cox Associates has developed business simulation models for the wireless, cable, data, and  multimedia (bundled telephony, cable, and data service) industries.  Examples of business  simulation software models developed since 1996 include the following:

  • Multimedia demand and market penetration forecasting model
  • Telephony growth-product introduction strategy model
  • Wireless Switch and Interconnect planning model
  • Cable customer acquisition, attrition and retention model

In each case, the main outputs included multi-year revenue projections and uncertainty analyses  showing how the probable value (NPV) of a business depends on its structure, price plans, timing  and locations of new service and network plant rollouts, changing composition of subscriber or  customer demographics, service performance indicators, and competitor and regulatory actions.

Cox Associates’ Business Simulation Models are distinguished from competing models by their  use of well-validated customer behavior models based on extensive data.  Most competing models  rely on speculative assumptions about how customers (and competitors) are likely to respond to  changes. Our method is based on the following principles:

  • All revenue projections are driven by detailed modeling of customer purchasing, usage,  and attrition decisions in response to company and competitor offerings.
  • Service offerings and service-level performance are linked to investments in network  infrastructure, operations support, and staffing. These decisions also flow into the cost  side of the financial model.
  • Competitor actions are modeled by simple behavioral rules that incorporate typical  behavioral strategies. Sophisticated (game-theory and A-life) modeling strategies have  been investigated by our researchers, but simple behavioral rules appear to be more  practical and realistic.
  • Regulatory changes, technology innovations, mergers, and similar one-time events are  modeled by stochastic binary indicator variables.
  • The preceding elements are integrated and use to clarify our clients’ understanding of  their business through several iterations of (a) Formal influence diagram structuring and  quantification, leading to useful pictures of the business; (b) Statistical estimation of key  empirical relations (e.g., for costs and demand functions) from market and engineering  data. We can supply most starting values based on extensive industry experience and data  when this is needed to make a warm start. (c) Simulation of customer transitions among  behaviors over time, with links to resulting financial outputs. (d) Model validation,  refinement, and communication, to maximize its value to planners and strategists.

We have applied this modeling approach successfully to over a dozen businesses in the U.S. and  abroad, delivering interactive “electronic business cases” that add substantial insight and value  to static paper displays.  Our business simulation models allow business planners and strategists to experiment with  different assumptions and scenarios, to gain insights into effective competitive and technology  strategies, and to examine the sensitivity of business value and viability to both strategic and  tactical decisions. In addition to hypothetical and what-if capabilities, we bring special value to  this area by providing the results from data mining and hard data on customer behaviors and  survey responses in wireless, data, cable, and bundled service markets.

Customer Behavior Models

Our customer behavior modeling tools use “machine learning” techniques from artificial intelligence and computational statistics to predict probable customer behavior in response to different product and service offerings, based on demographics, past purchases, and experiences. Compared to traditional market segmentation, our technique  identifies specific combinations of actions that have the greatest predicted impact on  customer purchasing behaviors in both the short run and the longer term. IMPACT:  Reduce customer churn and retention costs by over 20% by allocating resources to factors  that most affect loyalty.

RECENT APPLCATIONS

More Accurate Churn Prediction

We used a new Predictive Clustering method to predict which customers are most likely to churn based on recent product-purchase patterns, account age and size, and demographics. The predictions were far more accurate than those from previous (logistic regression, neural net, and business rule) models, achieving more than twice the lift in predicting the 10% of customers who were most likely to churn.

Better Targeting and Revenue Delivery

Cox Associates’ Predictive Clustering technology was used to quantify the probability that each of several million customers would purchase each of a company’s products (as well as selected product combinations) in the next 3 months. More accurate targeting based on these predictions quickly led to a 15% increase in revenues in an in-market trial.

Tracking Impacts of Advertising

We applied the Predictive Clustering methodology in conjunction with survey data on ad awareness and Competitrack data on competitor advertising to quantify the impact on customer loyalty and purchasing habits of individual advertising campaigns in specific markets and to identify strategies for more effective advertising.

Service Improvement Planning

Predictive Clustering was also used to quantify the total impact on customer loyalty and revenues of alternative proposed service improvements. The resulting causal model was based in part on Customer Satisfaction survey data validated and refined using several years of actual customer behavior data. In contrast to previous, overly optimistic and insufficiently precise regression models, the Cox Associates predictive models clearly distinguished he expected causal impacts of changes in service metrics on customer behaviors.

Related Publications

  • Cox LA Jr, Wong C. State transition model for customer relationship management.   Direct Marketing Analytics Journal. May, 2006, 9-15.
  • Cox, LA Jr. Predicting and optimizing customer behaviors. Chapter 12 in A. Labbi (Ed.), Handbook of Integrated Risk Management for E-Business: Measuring, Modeling, and Managing Risk. J. Ross Publishing. February, 2005.
  • Cox, L.A. Jr., 2002. Data mining and causal modeling of customer behaviors. Telecommunications Systems. 21(2-4):349-381.
  • Cox, L.A., Jr., and Popken, D.A., 2002. A hybrid system-identification method for forecasting telecommunications product demands. International Journal of Forecasting Volume 18, Issue 4, October-December 2002, Pages 647-671
  • Cox, L.A. Jr., 2001. Forecasting demand for telecommunications products from cross-sectional data. Telecommunications Systems, 16:3, 439-456.