Churn
michel.bruley@teradata.com
Extract from various presentations: Owens, Telecom Lab, Aster Data …
January 2013
www.decideo.fr/bruley
Churn in Comm. Industry: a bottom line issue

Attracting thousands of new subscribers is worthless if an equal number
are leaving

Minimizing customer churn provides a number of benefits, such as:
– Minor investment in acquiring a new customer
– Higher efficiency in network usage
– Increase of added-value sales to long term customers
– Decrease of expenditure on help desk
– Decrease of exposure to frauds and bad debts
– Higher confidence of investors
www.decideo.fr/bruley
Churn: Why Customers Leave
Quality
15%
Other
5%
Price
15%
Poor
Service
45%
Lack of
Attention
20%
How can I effectively manage customer churn?
Why are my customers churning?
How do I identify key churn drivers across the customer lifecycle?
How can I predict when my customers will churn?
What kind of initiatives can I run to anticipate customer churn and address drivers of churn?
How do I report on churn and retention initiatives?
www.decideo.fr/bruley
Churn management: scoping the problem (1)

Churn can be defined and measured in different ways
– “Absolute” Churn. number of subscribers disconnected, as a
percentage of the subscriber base over a given period
– “Line” or “Service” Churn. number of lines or services
disconnected, as a percentage of the total amount of lines or services
subscribed by the customers
– “Primary Churn”. number of defections
– “Secondary Churn”. drop in traffic volume, with respect to different
typology of calls
www.decideo.fr/bruley
Churn management: scoping the problem (2)

Measuring churn is getting more and more difficult
– Growing tendency for Business users to split their business between
several competing fixed network operators
– Carrier selection enables Residential customers to make different
kind of calls with different operators
– Carrier pre-selection and Unbundling of the Local Loop makes it
very difficult to profile customers according to their
“telecommunication needs”

Other frequent questions for Fixed Network Services
– What if a customer changes his type of subscription, but remains in
the same telco? What if the name of a subscriber changes? What if
he relocates?
www.decideo.fr/bruley
The case: Churn Analysis for wireless services

The framework
– A major network operator willing to establish a more effective
process for implementing and measuring the performance of loyalty
schemes

Objectives of the “churn management” project
– Building a new corporate Customer Data Warehouse aimed to
support Marketing and Customer Care areas in their initiatives
– Developing a Churn Analysis system based upon data mining
technology to analyze the customer database and predict churn
www.decideo.fr/bruley
Business understanding

Sponsors
– Marketing dept., IT applications, IT operations

Analysis target
– Residential Customers, subscriptions

Churn measurement
– Absolute, primary churn

Goal:
– Predict churn/no churn situation of any particular customer given
5 months of historical data
www.decideo.fr/bruley
Solution scope
millions of residential customers
Usage patterns analysis
of Voice Services by
single subscriber line
millions of business customers
Usage patterns analysis
of Voice Services by
subscriber line, contract,
company, etc.
millions of customers
Usage patterns analysis
of VAS by single
subscriber line
www.decideo.fr/bruley
Application framework
Reporting OLAP
Data Server
Data Mining
•Campaign Targets
•New product /
services
•Loyalty schemes
•Performance
analysis
Marketing
Data Warehouse
Analytical Applications
Data Preprocessing
Marketing
automation
ETL
Loader
Service
automation
Sales
automation
Front-office
Systems
www.decideo.fr/bruley
Loader
Decision Engine
Customer data
Market data
Sales data
Customer service contacts
...
...
...
Listener
Loader
Data Collection &
Transformation
Contracts
Tariff plans
Billing data
Accounts data
Fraud / Bad debts data
Back-office
Systems
Data understanding
Input Data
xx operational systems
Customer
Data Warehouse
•More than 500 indicators per customer
•Loading: on a monthly basis
•Size: xTB
www.decideo.fr/bruley
• Customer demographics
Basic customer information
• Service Profile
Products/services purchased by
each customer.
• Tariff plans
Details of the tariff scheme in use
• Extra service information
Special plans / rates
Service bundles
• Call data aggregated by month
• Billing data aggregated by month
• Complaint information
• Fraud and bad debts data
• Customer service contacts
• Sales force contacts
• Market data
Modeling with Data Mining tool
Main steps
– Define Concepts, Attributes,
Relationships …
– Select Operators
– Build the execution workflow
www.decideo.fr/bruley
Concepts, Attributes, Relationships
Call data
records
Data about
subscribed
services
Demographic
attributes
Revenue data
www.decideo.fr/bruley
Construction stage output
Data Construction
Feature Selection
16 Raw attributes
45 Derived attributes
www.decideo.fr/bruley
Churn modeling chain
4 Predictive models,
one for each
customer segment
Medium value customers
are selected
training set
decision tree operator
applied to fit predict the
likelihood of a customer to
become a churner in the
month M6
Save output
www.decideo.fr/bruley
Predictive performance
HIGH customer model performance
100
MEDIUM customer model performance
100
94
80
80
60
60
40
19
40
11
20
20
ACTIVE
PRED_ACT
14
CHURNER
6
0
ACTIVE
PRED_CHN
100
PRED_CHN
VERY LOW customer model performance
95
100
82
75
CHURNER
0
PRED_ACT
LOW customer model performance
80
89
86
81
80
67
60
60
25
40
20
40
18
0
ACTIVE
PRED_ACT
www.decideo.fr/bruley
CHURNER
PRED_CHN
5
20
33
CHURNER
0
ACTIVE
PRED_ACT
PRED_CHN
What Is Graph Analysis?
Aster: MapReduce implementations for graph analysis
Operates on Any Transaction or Interaction
Data
• Identifies the individuals or nodes in a network
• Identifies the relationships or edges in a network
In-Memory Graph Structure Allows for Graph
Analytics
• MapReduce creates a graph object that can then be
traversed for analysis
• Traversal of the graph is non-trivial even for simple
graph analysis
Output of Graph Analysis Is
Flexible
•
www.decideo.fr/bruley
MapReduce used to dynamically bind structure to
data on execution
Teradata Aster Graph Analysis
Why this belongs on Aster
•
•
Limitations in SQL Relational DBMSs
-
Set-based SQL is a poor programming construct
for Graph problems
-
Every connection between 2 people is a self-join
in SQL
Aster Metrics:
Social Graph Analysis
Aster Advantages
-
Read & transform data into in-memory graph
structure
-
Perform standard SQL logic or MapReduce on
the in-memory graph structure
-
Influencer Analytics: traverse the graph for
single shortest path – six degrees of Kevin Bacon
Environment:
1 billion rows, 700GB, 11 workers
Aster Response:
180 Seconds
www.decideo.fr/bruley
Graph Analysis & Churn Prediction
Graph Analysis constructs an influence propagation model
- Given persons who churned (initial churners)
- Diffuse their influence into their social environment
- Thus, their friends are at a larger churn risk..
(The two C1s churn; N does not)
- And this propagates to some of their friends’ friends
(C2 affected due to indirect, cumulative influence)
C1
*
I
I
*
*
Output
- List of predicted churners
N
C1
C2
Business Value
- Captures higher order social effects
- Capture the effect of multiple churners on a subscriber
- Does not require profile information
Influence spreading
Indirect influence
- Once the model is created, it can be run quickly & often
- Complements traditional churn models
I
C
www.decideo.fr/bruley
Initial churners (known)
Predicted churners
Understand customers paths to service cancellation
Multi-Channel Path Analysis
Identify most frequent paths to
early termination of service
Analyze specific patterns of
customer behavior
Across multiple channels of
customer engagement – web,
retail, customer service
www.decideo.fr/bruley
Customer Journey across Multiple Channels
call drop outs
data drop outs in web
(PDP) sessions
Path to churn
level of call quality
(voice and data speed)
*nPath analysis
Customer experience
score
Propensity-to-churn
model
Propensity to churn
3G to 2G drop down and
length of time on 2G.
sentiment analysis from
call center records
Good Score = Upsell opportunity
Bad Score = Retention Activities
Variation Score = Explanatory Message/Action
*nPath - pre-packaged SQL-MapReduce function for finding sequences of events
www.decideo.fr/bruley
Analysis across Diverse Sources & Data Types
www.decideo.fr/bruley
Discover Specific Service Cancellation Paths
www.decideo.fr/bruley
Detect & Prevent Customer Churn
Big Data & Churn Prevention
• Enrich Traditional Churn Model
• Graph Analysis
• Multi-Channel Path Analysis
Business impact
• With significantly less effort, know when customers are in
the last mile of considering leaving
• Higher customer retention leading to lower costs and
higher profitability
• Higher customer satisfaction
www.decideo.fr/bruley