customer insight development in vodafone italy 21_vienna pag. 1 customer insight development in...
TRANSCRIPT
Pag. 1Seugi 21_Vienna
Customer Insight development in Vodafone Italy
Emanuele Baruffa – Vodafone
Seugi - Vienna, 17-19 June 2003
Pag. 2Seugi 21_Vienna
ContentsContents::
1. Introduction
2. Customer Base Management
3. Customer Insight
4. Data Environment
5. Conclusions
Pag. 3Seugi 21_Vienna
1. Introduction
2. Customer Base Management
3. Customer Insight
4. Data Environment
5. Conclusions
Pag. 4Seugi 21_Vienna
Mobile telephony is one of the fastest Mobile telephony is one of the fastest growing industries in the worldgrowing industries in the world
! 1 billion subscribers around the world
! Over 120 countries have mobile networks
! Further acceleration expected
Source: EITO
Worldwide growth in subscribers (millions)
14 3487
206
479
727
958
1152
1321
1480
1991 1993 1995 1997 1999 2000 2001 2002 2003e 2004e
Pag. 5Seugi 21_Vienna
Italy: Europe�s second biggest mobile marketItaly: Europe�s second biggest mobile market
Country Subscribers Penetration%
Western European TLC market growth by country (%)
5,65,0
4,0
5,7
6,8
5,8
9,1
6,25,6 5,4
6,16,8
0
1
2
3
4
5
6
7
8
9
10
Germany Italy UK France Spain WesternEurope
Source: EITO 2001/2002 2002/2003
Germany 60,300,000 84%
Italy 54,000,000 98%
UK 50,900,000 92%
France 39,000,000 77%
Spain 34,000,000 88%
Sources: internal sources for Italy, Yankee Group for other European countries
Pag. 6Seugi 21_Vienna
Penetration rate in the Italian market Penetration rate in the Italian market
91%
74%
53%
36%
21%11%
0
10,000
20,000
30,000
40,000
50,000
60,000
1996 1997 1998 1999 2000 20010%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%Subscribers (,000)Penetration Rate
96% 98%
2002 2003
Pag. 7Seugi 21_Vienna
Italian market sharesItalian market shares
Total subscribers on 31.03.03
36%
17%
47%
Pag. 8Seugi 21_Vienna
Vodafone Italy Customer BaseVodafone Italy Customer Base
713,0002,460,000
6,190,000
10,418,000
14,920,00017,400,000
Dec. 1996 Dec. 1997 Dec. 1998 Dec. 1999 Dec. 2000 Dec. 2001
18,900,000
Dec. 2002
Pag. 9Seugi 21_Vienna
1. Introduction
2. Customer Base Management
3. Customer Insight
4. Data Environment
5. Conclusions
Pag. 10Seugi 21_Vienna
Customer management strategy! Strategy: Consolidate leadership through customer base management
! Marketing Goals: meet customers� needs and identify the best treatment to each customer at the right time through the most suitable channel at the appropriate cost. This approach helps to increase
! customer loyalty
! customer value (ARPU)
How
! create customer insight building a Customer Centric DB (CKM) that allow to:
! Identify segments for cross-selling actions " Customer Segmentation
! Identify customers with highest value potential for new high value added services such as SMS, Voice Mail " Propensity to uptake VAS
! Identify customers with highest churn probability " Churn propensity score
Pag. 11Seugi 21_Vienna
1. Introduction
2. Customer Base Management
3. Customer Insight
4. Data Environment
5. Conclusions
Pag. 12Seugi 21_Vienna
Approach for developing customer insightApproach for developing customer insight
2. Gather comprehensive
customer information
3. Analyze and segment customer base
4. Data Mining5.
Measure results
6. Fine Tuning
1. Brainstorming
Pag. 13Seugi 21_Vienna
Customer Value ScoreCustomer Value Score
~ 30%
20-30%
High value High value customerscustomers
Low value Low value customerscustomers
% CB % Margin
20-30%
! High value customers make up for majority of margins! Margins are relevant and not revenues because some high spenders also have high interconnection cost! Resources should be allocated proportionally to margins generated
70-80%
~ 70%
Value is calculated on a monthly basis for each customer. Customer value is measured based on a four month average Marpu
Pag. 14Seugi 21_Vienna
Churn Churn ModellingModelling
MACROSEGMENTATION
TARGETDEFINITON
DATA ANALYSIS
MICROSEGMENTATION
MODELBUILDING
MODELIMPLEMEN-TATION
MODELEVALUATION
�Identify relevant market segments
� Example� Consumer# Prepaid# Contracts
� Corporate# Small Accounts# Large Accounts
�Define the event you want to predict
� Example� Inactive SIMfor Prepaid Customers� �Cancellation letter� for Contract Customers
�Data Exploration� Correlation between target variable and explanatory variables� Data transformation#Trends#Grouping
�Identification of groups with similar churn behaviour
� improvement of model accuracy�Focus on:# �Active� customers# High value customers
�Oversampling� Application of data mining techniques
� Logistic regression�Decision tree� Neural Networks
�Measurement of the goodness of fit (model validation on hold-out sample)
�Implement model in a production environment
� Monthly production of churn index
�Ex-post evaluation of model performance
� % of correct predictions
Analyse churn behavior of actual churners to predict churn behavior of the customer base
Pag. 15Seugi 21_Vienna
Churn Propensity ScoreChurn Propensity Score
Model Lift
5.5
3.2
0.5
Red Yellow Green
High Risk to churn
Medium Risk to churn
Low Risk to churn
The likelihood to churn is estimated on a monthly basis for each VO customer. Customers flagged in RED are five times more likely to churn than an average customer.
Pag. 16Seugi 21_Vienna
Segmentation Methodological Road Map
Segmentation Data
Mart
Data Analysis
Clustering
Segmentation
Algorithm
Build a data mart with all segmentation variables•Demographicdata•Voice Traffic(peak-offpeak, network, discounted tariffs)•Service usage(wap, gprs, sms, mplay, music, …)•Propensity to VAS uptake
Cluster analysis•Started from a large number of micro-clusters•Then reduced/ aggregated clusters to a “manageble” and meaningful number
Cluster validationwith market research
Application of segmentation rules to the entire customer base•Customers are assigned to the nearest cluster (rule: minimum Euclidean distance from cluster centroid)
FactoralAnalysis to identify main dimensions of segmentation
Elimination of outliers
Customers are assigned to segments based on their usage of mobile services, attitude towards technologies and their lifestyles
Pag. 17Seugi 21_Vienna
Developed and implemented statistical models to predict on a monthly basis the likelihood of a VO customer to uptake five different Value Added Services.
Propensity to Uptake Propensity to Uptake VASVAS
SMSSMS No Users SMS
Low Users SMS
ADVANCEDADVANCEDSMSSMS High Users SMS
not using Flash SMS
WAPWAPWAP handset owners not using WAP
VOICE VOICE MAILMAIL
No Users Voice Mail
INTERNET INTERNET SELF CARESELF CARE No registered
customers on Vodafone Italy web site
High propensity customers are on average three times more likely to start using a service than an average customer
Pag. 18Seugi 21_Vienna
1. Introduction
2. Customer Base Management
3. Customer Insight
4. Data Environment
5. Conclusions
Pag. 19Seugi 21_Vienna
Customer Knowledge Management Customer Knowledge Management System (CKM)System (CKM)
DWH (Oracle) (Oracle)
CKM(Oracle)
daily + monthly
DSS Analysis (Microstrategy)DSS Analysis (Microstrategy)
DWH Campaign ManagementSystem
Campaign ManagementSystem
Raw dataDemographicsTrafficService UsageHandsetLoyalty����
Data Mining (SAS) ORACLE + File System SASData Mining (SAS) ORACLE + File System SAS
End User Data Mining (SAS Miner, SAS Stat, ...)
End User Data Mining (SAS Miner, SAS Stat, ...)End User (Web)End User (Web)
Pag. 20Seugi 21_Vienna
CKM Modelling EnvironmentCKM Modelling Environment
Fase 2Fase 2
Fase 1Fase 1
CKMCKM
DWH
Campa-gne
Campa-gne
��.��.
Segmentation & Sampling
DataTransformation
DataTransformation
Model Development
Model Development
ScoreScore
RulesRules
Metadata
Anagrafica
AnagraficaAnagrafi
ca
Anagrafica
SAS EnterpriseMiner ®
SAS EnterpriseMiner ®
SAS WarehouseAdministrator ®SAS WarehouseAdministrator ®
SAS AccessOracle ®
SAS AccessOracle ®
SAS Connect ®SAS Connect ®
SAS CKM Application
Pag. 21Seugi 21_Vienna
SAS CKM ApplicationSAS CKM Application! A Graphical User interface: has been developed to interact
with the Customer DataMart.
! Some of the functions of this GUY:
$ Dynamical datamart: extraction and definition of sub-universe for a given temporal interval with the maximum flexibility
$ Mining :
1. Development of a statistical model with SAS/Enterprise Miner®
2. Export of the model
! Assessment of predictive models when deployed: allow to verify performance on model (to measure degradation) and doing ex-post analysis on redemption.
Pag. 22Seugi 21_Vienna
1. Introduction
2. Customer Base Management
3. Customer Insight
4. Data Environment
5. Conclusions
Pag. 23Seugi 21_Vienna
In Summary�In Summary�
A successful strategy needs to be based on a good understanding of customer needs by customer groups
Develop Customer Insight simple to understand in order to become a real working tool for all parts of the organization (customer care, marketing, sales).
Take actions on customer insight and fine tunemodels to improve results over time
Reasons to Fail:% Lack of strategy
% Lack of data
% Lack of statistical skills
% Lack of commitment from Marketing CRM/Customer operations