Churn prediction and management system

Abstract

A system and method for managing churn among the customers of a business is provided. The system and method provide for an analysis of the causes of customer churn and identifies customers who are most likely to churn in the future. Identifying likely churners allows appropriate steps to be taken to prevent customers who are likely to churn from actually churning. The system included a dedicated data mart, a population architecture, a data manipulation module, a data mining tool and an end user access module for accessing results and preparing preconfigured reports. The method includes adopting an appropriate definition of churn, analyzing historical customer to identify significant trends and variables, preparing data for data mining, training a prediction model, verifying the results, deploying the model, defining retention targets, and identifying the most responsive targets.

1 . A method of designing an efficient customer retention program for managing customer churn among customers of a business having a statistically large customer base, the customer retention program including an analysis of the causes of customer churn and identifying customers who are most likely to churn in the future, so that appropriate steps may be taken to prevent customers who are likely to churn in the future from churning, the method comprising:
 adopting a definition of churn sufficient to encompass all customers in the customer base and which relies on objective factors to determine whether individual customers have churned or remain active;   analyzing historical customer data to identify significant trends and variables that provide insight into causes of churn and to identify classes of customers who are more likely to churn than others;   preparing customer data, including data corresponding to the identified trends and variables, for data mining and predictive modeling;   training at least one predictive model on historical customer data;   verifying the accuracy of the at least one predictive model based on historical data;   deploying the at least one trained model on current customer data to generate a propensity to churn score for individual customers indicating the relative likelihood that the individual customer will churn within a specified time period in the future;   defining characteristics of the target customers to be contacted during the course of the customer retention program; and   compiling a list of targeted customers having the defined characteristics.   
     
     
         2 . The method of designing an efficient customer retention program of  claim 1  wherein training the predictive model comprises:
 assembling a first historical data set that includes prepared customer data from a training period in the past for which churn results are already known, applying the first historical data set to the predictive model to obtain a first set of training results; comparing the first set of training results to the known churn results for the training period; and 
 adjusting the model to compensate for discrepancies between the training results and the known churn results. 
 
     
     
         3 . The method of designing an efficient customer retention program of  claim 2  wherein training the predictive model further comprises:
 assembling a plurality of historical data sets from a plurality of training periods in the past for which the churn results are already known; 
 applying the historical data sets to the predictive model in an iterative process, and; 
 comparing the training results to the known churn results for each iteration, and adjusting the model accordingly. 
 
     
     
         4 . The method of designing as efficient customer retention program of  claim 3  wherein the plurality of historical data sets are taken from different but overlapping training periods. 
     
     
         5 . The method of designing an efficient customer retention program of  claim 1  wherein verifying the accuracy of at least one predictive model comprises:
 assembling model verification data set that includes prepared customer data from a verification period in the past for which churn results are already known; 
 applying the verification data set to the predictive model to obtain a set of verification test results; 
 comparing the verification test results to the known churn results for the verification period, and 
 determining whether the verification results are satisfactory. 
 
     
     
         6 . The method of designing an efficient customer retention program of  claim 1  wherein preparing customer data for data mining and predictive modeling comprises calculating derived variables from customer data, the derived variables being applied to data mining and predictive modeling. 
     
     
         7 . The method of designing an efficient customer retention program of  claim 6  wherein calculating a derived variable comprises calculating an average value from a plurality of data values based on multiple observations of a single data variable. 
     
     
         8 . The method of designing an efficient customer retention program of  claim 6  wherein calculating a derived variable comprises calculating a trend line that represents a best fit among a plurality of data points based on multiple observations of single variable, and calculating a slope of the trend line. 
     
     
         9 . The method of designing an efficient customer retention program of  claim 6  wherein calculating a derived variable comprises calculating a distribution of customers based on a value of a data variable associated with individual customers, and classifying customers based on where they fall within the distribution according to the values of the data variable associated with each individual customer. 
     
     
         10 . The method of designing an efficient customer retention program on  claim 1  wherein preparing customer data for data mining and predictive modeling comprises assembling and analytical record including data from a plurality of customers, the data including variable values associated with individual customers for the variables that have been identified as being significant for analyzing and predicting churn. 
     
     
         11 . The method of designing an efficient customer retention program of  claim 10  wherein preparing customer data for data mining and predictive modeling comprises assembling a first analytical record for input to a clustering data mining operation to in which significant groups of customers are identified based on common behavior characteristics, and assembling a second analytical record for input to the clustering data mining operation to identify significant groups of customers based on common value characteristics. 
     
     
         12 . The method of designing an efficient customer retention program of  claim 1  wherein defining characteristics of target customers to be contacted during the course of the customer retention program comprises establishing a threshold churn propensity score, and targeting customers having a churn propensity score greater than the established threshold. 
     
     
         13 . The method of designing an efficient customer retention program of  claim 12  further comprising identifying a customer characteristic other than a customer's churn propensity score, and further filtering targeted customers based on the other characteristic. 
     
     
         14 . The method of designing an efficient customer retention program of  claim 13  wherein the other customer characteristic is customer value. 
     
     
         15 . A method of identifying targets for a customer retention program, the method comprising:
 identifying a set of customer data variables from which a customer's propensity to churn during a future period may be estimated based on values of the identified customer data variables associated with the customer;   providing a data mining tool with predictive modeling capabilities, the tool supporting at least one predictive model for estimating the propensity of individual customers to churn during the future period;   training the at least one predictive model on historical customer data for which churn results are known such that the at least one predictive model may be refined based on a comparison of the estimated churn propensities of individual customers against actual churn results;   deploying the trained model on current data to estimate churn propensities of individual customers for the period;   selecting targets for the customer retention program based on said churn propensities.   
     
     
         16 . The method of identifying targets of a customer retention program of  claim 15  further comprising:
 receiving customer data in monthly installments; and 
 defining a prediction horizon such that the predictive model calculates customer propensities to churn during the prediction horizon based on customer data installments received in previous months. 
 
     
     
         17 . The method of identifying targets of a customer retention program of  claim 16  further comprising compiling a first historical data training set including a plurality of historical customer data installments, an historical data analysis month, and an historical prediction horizon, all corresponding to a period of time in the past, the first historical data set including actual churn results accumulated during the historical prediction horizon. 
     
     
         18 . The method of identifying targets for a customer retention program of  claim 17  wherein training the at least one predictive model on historical customer data comprises applying the first historical data training set to the predictive model to predict churn events expected to have occurred in the historical prediction horizon and comparing the predicted churn events with the actual churn results accumulated during the historical prediction horizon, and refining the predictive model based on any discrepancies. 
     
     
         19 . The method of identifying targets for a customer retention program of  claim 18  further comprising compiling a second historical data training set substantially similar to the first historical data training set but wherein the historical customer data installments, the historical data analysis month, and the historical prediction horizon of the second historical data training set are offset in time from the historical customer data installments, the historical data analysis month, and the historical prediction horizon of the first historical data training set. 
     
     
         20 . The method of identifying targets for a customer retention program of  claim 18  further comprising compiling a model verification data set substantially similar to the first historical data training set but wherein the historical customer data installments, the historical data analysis month, and the historical prediction horizon of the model verification data set do not correspond in time with the historical customer data installments, the historical data analysis month, and the historical prediction horizon of the first historical data training set. 
     
     
         21 . The method of identifying targets for a customer retention program of  claim 20  further comprising applying the model verification data set to the predictive model to predict churn events expected to occur in the historical prediction horizon of the verification data set, and comparing the results predicted by the predictive model with the actual churn results accumulated during the verification data set prediction horizon. 
     
     
         22 . The method of identifying targets for a customer retention program according to  claim 15  wherein the data mining tool comprises an SAS Data Miner. 
     
     
         23 . The method of identifying targets for a customer retention program according to  claim 15  wherein the data mining tool comprises a KXEN data mining tool. 
     
     
         24 . The method of identifying targets for a customer retention program according to  claim 15  further comprising defining at least one derived variable and calculating a derived variable value for individual customers. 
     
     
         25 . The method of identifying targets for a customer retention program according to  claim 24  wherein calculating a value for the derived variable comprises:
 selecting a base variable for which individual customers have a corresponding value each month; 
 calculating a base variable average value for individual customers based on individual customers' base variable monthly values over a number of months; 
 calculating a customer distribution based on individual customers' base variable average values; 
 classifying customers based on their position within the distribution; and 
 storing individual customers' classifications as the customers' derived variable values.