Social network analysis (SNA) uses relationships and influence within social networks to predict customer churn in the telecom industry. SNA can improve existing predictive models of churn by over 10 times by incorporating social parameters. It can also identify influential customers within a network who are most likely to reduce churn if targeted in marketing campaigns. Implementing SNA involves constructing the social network from call data, calculating centrality metrics, and analyzing the data to predict churn and improve retention strategies and marketing.

1. A NimzoSTAT White Paper
Onkerzelestraat 103, 9500 Geraardsbergen, Belgium
+32.473.413177; info@nimzostat.com
www.nimzostat.com
Social Network Analysis
Decrease Churn Rate at Telecom Operators
by Rudy Jacob, Philippe Kerremans
March 2010

2. INTRODUCTION
CONTENTS In today’s highly complex and dynamic market the
ability of telecom operators to retain customers and
cross-sell products and services is of critical
importance. As Telecom penetration is increasing
and even approaching saturation, the focus of
Introduction ......................................... 2
Telecom Business Intelligence is shifting from
customer acquisition to customer retention;
Churn in Telecom is a Problem ......... 2
Telecoms must work harder than ever to hold onto
their customer base and attempt to draw numbers
Added Value of SNA ........................... 3
from their rivals.
Improve Existing Models ........................ 3 Further investments are being made into customer
profiling, to obtain a detailed picture of customer
Predict Churn Standalone ....................... 3
segments and individual customers. However,
Improve Marketing Campaigns .............. 3 traditional methods of generating customer insight
to achieve these goals continue to miss a key
How to Implement SNA ...................... 3 dimension. The key to better targeting and
treatment is to gain insight into the relationships
Identify Influencers in your Network ...... 3 customers have with other people in their (call)
network.
Modules of Social Network Analysis....... 3
SOCIAL NETWORK ANALYSIS (SNA) IS A FAST
Constructing the Network .................. 3
EMERGING DISCIPLINE FOR PREDICTING AND
Deeper Digging ................................... 4 INFLUENCING CONSUMER BEHAVIOR. IT
FOCUSES ON HOW RELATIONSHIPS ARE BUILT
Methodology ...................................... 5 AND HOW THEY CONTRIBUTE TO INFLUENCING
INDIVIDUALS WITHIN SOCIAL GROUPS.
Let us analyze your call data ............. 5
Integrating SNA in marketing campaigns and
Conclusion .......................................... 5 customer retention has shown on several accounts
that churn can be reduced significantly in Telecom
About NimzoSTAT .............................. 6 operators (prepaid/postpaid, fixed/mobile).
About the authors .................................. 6 CHURN IN TELECOM IS A PROBLEM
Churn and customer value are critical to Telecoms. If
a customer spends $50/month and you have 5M
customers then 0.5% churn is $1.25M/mo every
month from then on. Annual churn rates in the
prepaid segment average between a significant 50 to
70 percent. Lowering this churn percentage has a
large effect on the bottom line.
2|P a g e

3. Even a small reduction in churn can mean IMPROVE MARKETING CAMPAIGNS
big savings—the cost of retaining a client is
estimated to be only one-fifth of acquiring By having detailed influence parameters per target,
one. And these consumers could ultimately marketing campaigns can be better segmented and
help decrease the churn within their own therefore cheaper and more efficient.
social circles, amounting to even more
potential savings.
HOW TO IMPLEMENT SNA
ADDED VALUE OF SNA
IDENTIFY INFLUENCERS IN YOUR
NETWORK
Attributes derived from SNA may be used
alone or as input to classic predictive
Influencers are people that are more (or more
models to help improve their accuracy.
strongly) connected. Not only will their impact on
Marketing campaigns aimed at retaining
the network be larger than for people with an
existing subscribers can be enriched by the
average amount of connections, they are also an
intelligence gained through SNA
ideal target for a marketing campaign. Research has
capabilities. SNA helps to pinpoint
shown that network neighbors – customers who are
influential customers who are the most
linked to an existing consumer – adopt a service at a
likely to drop their service or jump to a
rate three to five times greater than customers
competitor.
selected using marketing best practices. Because
those influencers will have contact with more
potential customers, the ROI of a marketing
IMPROVE EXISTING MODELS campaign will increase significantly when targeted to
the influencers in your network.
Integrating the social network parameters
into existing predictive models increases
the model performance by a factor of ten or
more and the retention campaigns see an
MODULES OF SOCIAL NETWORK ANALYSIS
even better uplift (10 – 25 times).
CONSTRUCTING THE NETWORK
This is the basis of all the detailed analysis.
PREDICT CHURN STANDALONE
Data is already available on-site in your Call Data
Even if no other predictive model is database. The processing step can be time
available at a Telco, it is still possible by consuming but social networks do not change
means of simply analyzing existing call data, drastically overnight, and therefore monthly or
to calculate the probability that if one user quarterly processing of the call data records is
churns, that other people in his network will sufficient.
churn.
Several centrality parameters are determined:
3|P a g e

4. o Degree Centrality: Number of direct o Eigenvector Centrality (Reach) is a measure of
ties to other people in the network. the importance of your nodes in the network.
Google's PageRank is a variant of the
Eigenvector centrality measure.
Depending on the type of marketing campaign one
of the above metrics will determine which customer
to select:
If you simply want to spread your message to as
many people as possible, you want to seed your
campaign with members that have high Degree
Centrality.
o Closeness Centrality: Capacity to IF YOU WANT TO SPREAD YOUR MESSAGE AS FAST
reach the rest of the network. AS POSSIBLE, YOU WILL NEED TO TARGET
MEMBERS WITH A HIGH CLOSENESS VALUE.
If conversion is important you must choose the
influencers that have a large reach.
DEEPER DIGGING
o Find families and head of households.
o Identify rotational churners (spinners).
Sometimes people leave and come back; try
to keep them for good.
o Betweenness Centrality: A node on a o Anomaly detection is not a direct way to
communication path controls the reduce churn, but identifying anomalies in
communication flow, and therefore the network can keep those nodes out of
important (node value .73). the calculations, so that the normal ones
provide cleaner results.
4|P a g e

5. The “Validation & Interpretation” step requires your
METHODOLOGY business knowledge and assistance, together with
NimzoSTAT’s statistical analysis.
NimzoSTAT uses the Fayyad methodology
for Knowledge Discovery in its SNA solution. NimzoSTAT has collaborations with leading scientific
researchers within the field of social network
The iterative nature is important for
analysis and modeling, such as Prof. Dr. Bart Baesens
constant learning and tuning of the model.
from K.U.Leuven.
The centrality parameters in a way, are
“just” numbers, and need to be properly The output can be stored in your database and
weighed for their influence of churn in your integrated in your business process to set up
network. marketing campaigns.
By contrasting these results with your list of Our solution is offered in a SaaS business model.
actual churners and other socio- However, for privacy policy reasons, the solution can
demographic and CRM data, the model can be also be deployed locally within your existing IT
be further optimized. infrastructure.
LET US ANALYZE YOUR CALL DATA
CONCLUSION
NimzoSTAT has the algorithms in place to
Social Network Analysis provides ways to help you to
analyze your CDR data. We discuss with all
lower customer churn significantly and can decrease
stakeholders how to select, pre-process,
the cost of your marketing campaign through better
transform and mine your data for SNA.
selection of the target customer population.
5|P a g e

6. ABOUT NIMZOSTAT
NimzoSTAT is a Belgian privately owned
company. NimzoSTAT innovates with its
own down-to-earth solutions, placing the
field-specific expertise of its customers at
the heart of its know-how and tools.
NimzoSTAT translates its cross-sector
experience in a win-win solution for each
customer. Our goal is to activate the
memory (databases, data warehouse) of
your company to be able to use it to steer
your current and future policy and vision in
an easy and efficient manner.
ABOUT THE AUTHORS
Rudy Jacob has 15 years of international
experience with Telecom BSS systems and is
Managing Partner of NimzoSTAT.
Dr. Philippe Kerremans has 25 years of
experience as an IT professional and is an
expert in virtual world and social network
analytics.
We collaborate with the academic world,
and take the opportunity to thank
Prof. Dr. Bart Baesens of the University of
Leuven, Belgium (www.econ.kuleuven.be)
for his input for this white paper.
6|P a g e