Professional Documents
Culture Documents
Peter Christie et al, 2011. Applied Analytics Using SAS Enterprise Miner Course Notes. Cary: SAS Institute Inc.
SAS Institute Inc. Churn prediction in the telecommunications industry. A SAS White Paper. Cary, SAS Institute Inc.
1999.
Sacha Schubert. Customer Attrition Management in the Financial Services Industry. A SAS White Paper. Cary, SAS
Institute Inc. 2002.
Gordon S. Linoff and Michael J. A. Berry. Data Mining Techniques. Wiley, latest edition.
Also…
CHURN PREDICTION A SIMPLE CHURN DEFINITION
(20 euro / month * 12 months) * 1,000,000 customers * 4% churn rate = 9,600,000 euro.
If we detect 50% of the churners in advance, and we manage to keep 25% of them, the value
of the model is:
Time
Learning Window Learning Window Prediction Window
- -
Customer Characteristics Customer Churn/ Non - Churn
Behavior
Categories of
Customer Characteristics
Customer Attributes
Demographic attributes
Churn
ID
(1 / 0)
I1 I2 I3 I4 I5 ….. In
Delta on ratio variables ….. ….. ….. ….. ….. ….. … ….. …..
RFM Metrics
How recent is the last contact with the “hot-line” (customer support)?
How recent is the last purchase of the individual?
CHURN PREDICTION THE RFM APPROACH FOR CUSTOMER PROFILING
F (“F” means “Frequency”):
Ratio variables:
These variables encode the ratio between two related RFM variables:
What’s the ratio of calls “outside the network” compared to calls “inside the network”?
What’s the ratio of MOU “outside the network” compared to MOU“ inside the network”?
What’s the ratio of “international” MOU compared to MOU “inside the network”?
CHURN PREDICTION DELTA VARIABLES FOR CUSTOMER PROFILING
Delta variables:
These variables encode the difference in RFM variables between two time periods:
How much increase (or decrease) in terms of MOU (minute-of-usage) between the last time period
and the previous one?
How much increase (or decrease) in terms of expenses (in euros) between the last time period and
the previous one?
CHURN PREDICTION DELTA ON RATIO VARIABLES FOR CUSTOMER PROFILING
How much increase (or decrease) in terms of ratio of MOU “outside the network” compared to
MOU“inside the network” between the last time period and the previous one?
CHURN PREDICTION FEATURE ENGINEERING FROM SNA (COMMUNITY MINING)
C = Churners
NC = Non Churners
Case ?
%C %NC Existence Existence Avg Age Avg Income Duration of Duration of Dur of Calls to C
(Relational) (Relational) of C of NC of Neighbors of Neighbors Calls to C Calls to NC Dur of Calls to NC
678
Case ? 3/5 2/5 YES YES 25 12.567 678 203
203
CHURN PREDICTION IN THE MOBILE
TELCOS INDUSTRY CASE STUDY
Steps to Follow
17 Attributes
Customer Attributes
Churn
ID
(1 / 0)
CustServ
Day_Calls Day_Charge Day_Mins ….. Night_Calls Night_Charge
Calls
Indicates how many call the customer made to customer service during the period in
CustServ_Calls Interval
examination.
Day_Calls Indicates the number of calls the customer made during the day. Interval
Day_Charge Indicates how much the customer paid for the day calls. Interval
Day_Mins Indicates the number of minutes the customer used the service during the day. Interval
Eve_Calls Indicates the number of calls the customer made during the evening. Interval
Eve_Charge Indicates how much the customer paid for the evening calls. Interval
Eve_Mins Indicates the number of minutes the customer used the service during the evening. Interval
Intl_Calls Indicates the number of international calls the customer made. Interval
CHURN PREDICTION AVAILABLE HISTORICAL DATA (18 VARIABLES)
Intl_Charge Indicates how much the customer paid for the international calls. Interval
Intl_Mins Indicates the number of minutes the customer used in international calls. Interval
Night_Calls Indicates the number of calls the customer made during the night. Interval
Night_Charge Indicates how much the customer paid for the night calls. Interval
Night_Mins Indicates the number of minutes the customer used the service during the night. Interval
Vmail_Message Indicates the number of voice mail messages the customer sent. Interval
Vmail_Plan Indicates whether the customer has a voice mail plan. Binary
Customer Attributes
Churn
ID
(1 / 0)
CustServ
Day_Min Day_Charge Day_Calls ….. Night_Calls Night_Charge
Calls
Churner/
Frequency Percent
Non - Churner
1 707 14%
0 4,293 86%
CHURN PREDICTION MODEL DEVELOPMENT
USING PREDICTVE ANALYTICS
• They help us find which attributes or characteristics of the customers are important for classifying
churners from non - churners.
• They help us predict which customers are likely to churn and with what probability.
As you remember for each customer we have 17 characteristics e.g. evening calls, international calls, etc. The decision tree will
tell us which of those characteristics are important for classifying a customer as churner or not churner.
Day_Mins Indicates the number of minutes the customer used the service during the day.
Night_Charge Indicates how much the customer paid for the night calls.
Intl_Plan Indicates whether the customer has an international plan.
Vmail_Message Indicates the number of voice mail messages the customer sent.
Night_Mins Indicates the number of minutes the customer used the service during the night.
Vmail_Plan Indicates whether the customer has a voice mail plan.
The Decision Tree Model
The Decision Tree Model
If p1 >= 26.67% then customer = churner
CREATING A PROFIT MATRIX
In the case study under consideration, there are four outcome/ action combinations:
Give Incentives
Do Nothing
to Stay
Churner
Non - Churner
Give Incentives
Do Nothing
to Stay
Churner
Non - Churner 0
...
CREATING A PROFIT MATRIX (2)
In order to persuade a customer not to churn, the expected profit from giving
incentives to stay should be greater than the expected profit from doing nothing. So:
𝑝1 > 26.67%
Give Incentives
Do Nothing
to Stay
Churner 700 -1.500
...
DEMO OF CHURN PREDICTION
USING SAS VIYA
Also customers have become more demanding, there is high cost in acquiring new customers,
product lifecycles are shortening, customer loyalty and profit margins are decreasing and the
market has become saturated.
These characteristics have transformed the mobile operators from fishermen to farmers in the
sense that their biggest priority is to retain as many of their customers and hence maintain their
market shares.
This section demonstrated how we can use a state of the art software tool – SAS Viya – to predict
customer attrition or customer churn in a mobile using operator using analytics - based methods.
THANK YOU!