Viral marketing in telecoms alpha customers in phone networks by tls14265

VIEWS: 217 PAGES: 21

									 D1 Solutions AG

 a Netcetera Company




Viral marketing in telecoms: alpha customers in phone networks

Stamatis Stefanakos




October 2008
                                                                           2


Marketing through social networks?

Some possibilities:
 Word-of-mouth product penetration

 Campaign targeting through hub (“alpha”) selection

 Customer risk assessment by social-ties analysis




 Goal: Not a graph-theoretic study of social networks but a hands-on
  evaluation of the possibilities of using social networking information
  in CRM and marketing.
                              3


A case study in telecoms

The company:

   Number 2 in Switzerland
   Mobile & wireline
   1.5M mobile „customers“
                                                                       4


Motivation

Marketing and CRM face one main challenge:

 „Know your customers“

Main activities:

   Design of new products (what do our customers need and how
    much are they willing to pay for it?)
   Launch of promotional campaigns (what should we offer and to
    whom?)
   Customer retention (which customers are considering to leave us,
    what can we do to stop them?)
                                                5


Available tools

The company has a plethora of data available:


   Usage & revenues
   Customer socio-demographics
   Product-customer relationship history
   Network logs (call detail records)
                                                                       6


Traditional actions

   Design of new products
     Guesstimates

     Ad-hoc analyses of customer behaviour

   Launch of promotional campaigns
     Guesstimates

     Selections of target groups based on revenue, products, socio-
       demographics
     Evaluation of previous campaigns

   Customer retention
     Churn models based on revenue, products, socio-demographics
                                                                         7


How can social networking help with this?

   Design of new products
     Analyze social behaviour of customers (e.g. sizes of cliques for
       determining the structure of „family“ products)
   Launch of promotional campaigns
     Identify and target „alpha“ customers (viral marketing)

   Customer retention
     Viral churn models
                                                                         8


Call Detail Records

Every call originating from or terminating in the company‘s network
generates a call detail record (CDR). The CDR contains information on:
 Type of call
 Originating MSISDN, terminating MSISDN
 Date / time
 Duration
 Completion code
 Cell (Starting / Ending)
 Volume (data calls)
 Handset
 ….


CDRs are available for mobile and wireline calls. Terminating wireline
CDRs are currently not available (these are handled in the network of
the incumbent operator).
                       9


The call (sms) graph
                                                                        10


The call graph

Is a directed / undirected graph:
 Nodes are subscribers.

 Edges are calls between subscribers.


In the undirected case calls have to be two-directional in a certain
period of time.

To limit the size we only add an edge if a minimum number of calls with
a required minimum duration is made between two nodes.

The call graph obeys a power law degree distribution and contains a
giant connected component with almost all nodes ( suitable for viral
marketing).
                                                                           11


The call graph: computational limits

The size of the call graph is the main obstacle in any analysis:

As there are millions of calls handled daily, the size of the call graph
can grow beyond computational feasibility.


   Voice graph (for one month):
     No restrictions: 11M nodes, 35.2M edges

     Edges with > 4 calls, 120sec total duration: 3.8M nodes, 7.8M
       edges
   SMS graph (for one month):
     No restrictions: 6.8M nodes, 26.2M edges

     Edges with >4 SMS: 2.7M nodes, 6.8M edges
                                                                 12


Alpha customers

The company bases all customer communication and product offers on
four revenue-based segments.

We could extend this using social-networking information:
 Social-behaviour based segments.

 Alpha customers: the „hubs“ of the network.

 Large 1st and 2nd in-degrees.
                                                                      13


Example campaign: 1000 SMS Special Promo

   An offer of 1000 SMS for a very cheap fix price is sent per SMS to
    ~200K prepaid customers.
   A customer that wants to take the offer has to send a keyword to a
    specific number.
   An amount is deducted from the customer‘s balance and 1000 SMS
    can be sent at no additional cost.
   12571 customers signed up.
   652 (5% of all takers) were not targeted by the campaign.
   Obviously the campaign spread beyond the initial targeted group via
    word-of-mouth.
     (Some customers though might have more than one
       subscriptions.)
                                                                      14


1000 SMS Special Promo: Analysis

   Out of the 652 takers that were not targeted by the campaign 47 had
    another subscription that was in the campaign.
   605 (4.8% of all takers) learned about the campaign via word-of-
    mouth.
   Out of the 605 customers, 339 had social links to the takers of the
    campaign (social links = call in the CDRs over a period of a month).
     We only consider takers, not the whole campaign group.

   The links in the takers group had an “alpha degree” of 194.
   The takers group in total had an “alpha degree” 117.
                                                                       15


Customer churn

Churn rate = measure of customer attrition, defined as the number of
customers that discontinue a service during a certain time period.

Churn control is currently the main focus in CRM in the European
markets: high (mobile) penetration and low barriers to switching
providers make it cheaper to retain customers than to acquire new
ones.

Typical yearly telecom churn rates:
 Postpaid: 10-15%

 Prepaid: 30-40%
                                                                   16


Prepaid churn prediction

Prepaid churn is defined solely by usage; there are no binding
contracts.
 Similar to consumer churn in everyday brands?


Available information on prepaid customers is limited
 No contract  bad data

 Socio-demographics are not always available

 The phone might be used by someone else than the one that
   registered it

Many prepaid promotions are based on cheaper or free calls/sms within
  the providers network.
 Social behaviour might be an influencing factor in prepaid churn.
                                                                      17


Prepaid churn prediction by social ties

Social tie = A calls B and B calls A, 5 or more calls, in Nov 2007.

Viral propagation of churn:
 Churned in June 2008:

     13.1% of their social ties had churned before they did.

 Active in June 2008:

     9.3% of their social ties had churned.
                                                                    18


Marketing through social networks?

   Telecom companies “own” social network information.
   Our simple analyses show that word-of-mouth can be quantitatively
    measured.
   Our experience indicates that social-networking analysis can be
    very useful for improving marketing and CRM activities.
     Gain knowledge about your customer base.

     Understand the value of your customers.

     Implement more effective campaign targeting.

     Improve customer risk assessment.
                                                                 19


Teasers

   Outside the telecom world:
     Do phone networks resemble human social networks?

     How about other networks (email, facebook, …)?

     Can we measure word-of-mouth and product churn rates in other
       social networks?
                                                                     20


Teasers

   From the algorithmic point of view:
     How do we deal with massive data sets?

        (find cliques / clustering / …)

     How to maximize word-of-mouth spreading?

     How about dynamic networks (over time…)?

        (rate of friend acquisition / friend loss)

     How do we integrate business specific information into our graph
       models?
 D1 Solutions AG

 a Netcetera Company




Contact
Stamatis Stefanakos
stamatis.stefanakos@d1solutions.ch, stamatis@gmail.com,
phone: +41 44 435 10 03 | mobile: + 41 78 634 09 24


D1 Solutions AG
Zypressenstrasse 71
CH-8040 Zürich

								
To top