Report on Telecom Churn Management

Description
Churn rate is an alias for attrition rate. It is one of the primary factors that determine the steady-state level of customers a business will support.

Telecom Churn Management
Data Mining for Business Intelligence

Contents
INTRODUCTION ............................................................................................................................................. 3 Churn Management: ................................................................................................................................. 3 Current Telecom Industry Scenario: ......................................................................................................... 3 Churning in the Indian Telecom Market and its Impact: .......................................................................... 4 Customer is King. ...................................................................................................................................... 4 DATA MINING AND ITS APPLICATIONS: ........................................................................................................ 5 BUSINESS OBJECTIVE .................................................................................................................................... 6 PREDICTION MODEL CREATION PROCESS .................................................................................................... 6 Define Scope ............................................................................................................................................. 6 Data preprocessing ................................................................................................................................... 6 Model creation.......................................................................................................................................... 6 Approach 1 ................................................................................................................................................ 8 Approach 2 ................................................................................................................................................ 9 Main drivers of churn.............................................................................................................................. 10 REFERENCES ................................................................................................................................................ 11

2

INTRODUCTION
Churn Management: Churn rate is an alias for attrition rate. It is one of the primary factors that determine the steadystate level of customers a business will support. In a broad sense, churn rate is a measure of the number of individuals or items moving in or out of a collection over a specific period of time. It is most widely used in business with respect to a contractual customer base. It is a very important factor for business with a subscriber-based service model like Mobile telephone networks and pay TV operators. Churn management is becoming famous in the telecom industry as according to various studies when the market is saturated, the pool of “available customers” is limited and an operator has to shift from its acquisition strategy to retention because the cost of acquisition is typically five times higher than retention Churn Management can be used to achieve following objectives. ? Acquire more loyal customers initially ? Identify the customers who are most likely to churn ? Cluster them into “more profitable” and “less profitable” ? Take preventive measures in order to retain the “more profitable customers” Current Telecom Industry Scenario: With the emergence of new players in the telecom operators in the Indian market, the competition is increasing at a rapid speed among the existing players and upcoming .As all the operators have the similar goals of increasing revenue,ARPU,number of users and decrease churn rate. .Most of the cellular circles in India has up to 6 mobile telephone offerings including at least 4 GSM operators and up to 2 CDMA operators. The telecom operators are coming with many new schemes like free SMS, 1 paisa per second etc every now and then to attract new customers in the new circles. While gaining new customer is a gain for any operator, the flip side is losing customers.

3

Churning in the Indian Telecom Market and its Impact: Customer churning Problem is extremely acute in Indian Telecom market. The usual trend is of customers joining and leaving the operators within a short span of time. According to research firm Gartner, India?s churn rate is highest in the Asia-Pacific region somewhere between 3.5 percent to 6 percent per month .One more finding says that the cost of acquiring a new customer is somewhere near Rs. 6000. Taking the industry average of 2-3%, Indian operators are losing 25-30% of their customers every year. Considering the churn rate and cost of acquiring new customer the loss is immense. Customer is King. Indian operators have started managing their customer base. This basically involves acquiring, development and retention of customers. Given the statistics of churn rate, retention has become the main focus for the telecom majors. The churn problem is more prevalent in the prepaid segment which forms a significant percentage of the mobile users. The prepaid customer is more price-sensitive than the post-paid one. With the low rentals, customers with low usage prefer prepaid cards. Also, students and those who like to experiment with different networks prefer the prepaid offering Reducing the churning rate is a difficult task. Even though Indian market is a growth stage, churn rate is high as many subscribers shift from one operator to another due to brand image, numerous tariff options available to customers, past billing disputes with a particular vendor. Other than this, some of the key factors that encourage churn are inadequate network coverage, which includes dropped calls that occur in places where network coverage is thin and blocked calls that occur when the demand for network services exceeds capacity. The most common way to manage customer churn is „Reactive Retention Program?, where the operators attempt to convince customer who are planning to leave the company to stay. However the down side of this is that this does not bring brand loyalty. Thus a challenge faced by these operators is to increase customers? loyalty before subscribers decide to leave and to aim efforts at customers who are at risk of churning. This all has lead to need for a Proactive Retention Program. Traditionally used in credit card business, predictive modeling technique is now
4

actively used in the telecom sector. It allows operators to determine which customers are most likely to churn and to take appropriate measures to decrease their likelihood of churning. A “Churn Model” analyzes the history of customers who have churned and those that have stayed and determines the probability that a current customer with a particular profile will churn or not. Thus, in general these models have two main functions (1) understanding the factors that influence a customer to churn and (2) identifying customers for whom these factors make it likely that they will churn.

DATA MINING AND ITS APPLICATIONS:
With the help of data mining techniques one can extract hidden predictive information from large databases. In the CRM space, data mining techniques used commonly includes clustering, associations, rule induction, genetic algorithm, decision tree, and neural network. The below table summarizes some data mining functionalities and techniques used in CRM.

Table: Data Mining Functionalities, Techniques, and CRM Applications Source: Applying Data Mining to Telecom Churn Management by Shin-Yuan Hung

5

BUSINESS OBJECTIVE
? To identify the key drivers of churn ? To formulate a new model using sample training data ? To test the data on the validation data ? To predict the churners using an another set of data ? To cluster them into “more profitable” and “less profitable”

PREDICTION MODEL CREATION PROCESS

Define Scope The data are a scaled down version of the full database of a wireless telephone company. There are 71,047 customers in the database, and 75 potential predictors. Churner is defined as a subscriber who is voluntary to leave; non-churner is the subscriber who is still using this operator?s service. The transaction data include billing data, call detail records (CDR), customer care, etc. Data preprocessing The sample database of 71,407 data points is divided into two parts: Calibration data and Validation data. The calibration database had a field „churndep? which give the value whether the user has churned or not. The same field is not present in the validation data. Thus, we further divided the calibration database into two equal parts using random variable to create training and testing database. Model creation The model for churn analysis was created with the help of Logistic Regression on SPSS. Before moving on to the logistic regression part, the 74 independent variables where divided into various categories depending upon their characteristics. The following table shows a sample of division of these independent variables. The dependent variable for this logistic regression equation is „churndep? the variable to be predicted.

6

Demographical Age Children Credit Rating

Behavioral Loyalty Level Usage Rate Attitude toward the service

Psychographic Personality

Service Dropped calls

Geographic

Occupation

User Status

Area/Region

After detailed SPSS analysis (the data & regression analysis is available in appended SPSS file), the number of variables where reduced to 54, taking care of multicollinearity and other similar issues.

As explained earlier, the following techniques are widely found application in “churn” prediction: 1) Regression 2) Neural Network and 3) Decision Tree. We intend to use Regression and Decision Tree to

7

arrive at the prediction model for the collected historical data. There are two approaches predominantly used in churn management in telecom industry. Approach 1: Segment the customers based on certain variables and then create a predictive model for each clusters Approach 2: Consider all the customers as a single segment and then create an apt predictive model using decision tree In our study we intend to use both approaches and to compare them against each other.

Approach 1
Step #1 – Clustering of customers The first step in this approach is to cluster the customer samples in to different groups. The segregated sample data was grouped by following the K-means clustering technique. And it was found that there were totally eleven clusters of the customers. Step #2 – Predictive model for each cluster Once the clustering of customers is done, then each cluster was exercised to arrive at the set of rules that determines whether the customer in a particular group will churn or not. And in each clusters, the customers behaved in a different manner i.e. they exhibited the different behaviour in churning. This is more prevalent if one looks at the decision rules that determine the likelihood of a customer who churns. Cluster 1 2 3 4 5 6 7 8 9 10 Determinant variables MOU, change in monthly usage Months at service, change in monthly usage, possession of refurbished equipment Change in monthly usage and overage rate Possession of refurbished equipment and number of retention calls made Months at service and change in monthly usage Change in credit rating Months at service Overage Mail order availed Credit rating

8

Approach 2
Step #1: Reduction of variables There were totally 74 input variables. First it was tried to find out the multi-co linearity between the independent variables using linear regression technique (SPSS). And, if the correlation co-efficient lies either below -0.5 or above 0.5, one of the independent variables was ignored or eliminated. Thus the total independent variables were reduced to 58, and these 58 independent variables were subsequently used for further model creation using decision tree technique. Step #2: Creation of predictive model using decision tree technique The 20000 customer data points were used as a sample to predict whether the customer is likely to churn or not. And the remaining 20000 customer data points were used to validate the model created. The total 40000 customer data points were randomly split into two parts: a) sample data and b) test data

9

Main drivers of churn

Accordingly, there were four main drivers that determine whether customer is likely to churn or not – a) Months at service, Set Price, Number of retention calls made and Web enabled phone Step #3: Validation of model using the test data Once the model was created using the sample data, the model was validated using the test data. Below is the lift-chart that detailing the comparison between using the valid model and random selection.

10

REFERENCES
? ? ? ? ? ? ? ? Berson, Alex, Smith, Stephen and Thearling, K. “Building Data Mining Applications for CRM”, McGraw-Hill, New York, NY, 2000. Applying Data Mining Techniques to Churn Management, Shin-Yuan Hung & Hsiu-Yu Wang, PACIS 2004 Proceedings. Cybenco, Horink “Approximation by Super-positions of Sigmoidal Function”, Mathematical Control Cignal Systems, Vol 2, 1998, pp. 303-314. Kentrias, Stelios “Customer Relationship Management-The SAS Perspective”,

www.cm2day.com, 2001. Lejeune, Miguel A.P.M. “Measuring The Impact of Data Mining on Churn Management”, Internet research: Electronic Network Applications and Policy Vol. 11, No. 5, 2001, pp. 375-387. Mattersion, Rob. “Telecom Churn Management”, APDG Publishing, NC, 2001. SAS Institute “Best Price In Churn Prediction”, a SAS institute white paper, 2000.

11



doc_821510025.docx
 

Attachments

Back
Top