Telco Churn

Reducing Voluntary Churn via Predictive Analytics for Telecom OperatorsMaking the business case and determining appropriate retention campaign budgets for mobile subscribers with a high propensity to switch

1

HK & AS2Overview (Slide 1 of 15)Telco Readiness ChecklistSegmentationPredictive AnalyticsAcquisitionCosts of Customer Acq. (COCA)ServicingRetentionVoluntary ChurnCustomer Lifetime Value (CLV)Monthly (Voluntary) ChurnPost vs PreFixed vs MobileChurn PredictionStatistical modellingDemographicsUsage (CDRs)Voluntary Churn ReductionRetention CampaignsBudgetingROI / EVAFUDs Fears, Uncertainties, DoubtsCSF Critical Success Factors

2

HK & AS3Telco Readiness Checklist

How are you defining your customer segments?What is your monthly churn (total and per segment)?Are you tracking reasons for churn?What portion of your total churn is voluntary?What is your monthly ARPU (Average Revenue Per User)?What is your COCA (Cost of Customer Acquisition) per subscriber?What is your definition of an active subscriber?What is your active subscriber base (in millions)?What are your average subscriber tenures (in months)?What is your cost of capital (WACC)?What is the breakup between Pre-paid & Post-paid for all of the above?What about mobile vs fixed line (POTS) breakup?How many months of CDRs do you keep online for call analysis?What is your definition of CLV (Customer Lifetime Value) and its avg value?What financial metrics do you use to determine whether to fund a particular project? (EVA, ROI, discounted payback periods, etc)

If you dont have all the answers above you need to get started on them before going much further on voluntary churn reduction using predictive analytics.We have got to be able to CRAWL before we can FLY!

First we CRAWL Then we WALKThen we RUNThen we FLY!

3

HK & AS4Segmentation (Slide 3 of 15)Most telcos define their customer segments using some of the following top-down approaches:By payment type (pre-paid vs. post-paid/contract)By ARPU (revenue generated)By tenure (age on network)By demographics (location, income, job, gender, etc)By usage VAS, data/SMS/MMS, other non-voice penetrationRoaming, ISD/international, STD/domestic long distance, voice-mailBy handsets/devicesWhile this is an important first step, there are supplementary bottom-up segmentation approaches using statistical analysis and grouping by behavioral similarities that have better predictive power

4

HK & AS5Predictive AnalyticsPredictive analytics encompasses a variety of techniques from statistics and data mining that process current and historical data in order to make predictions about future events. Such predictions rarely take the form of absolute statements, and are more likely to be expressed as values that correspond to the odds of a particular event or behavior taking place in the future.

In business, the models often process historical and transactional data to identify the risk or opportunity associated with a specific customer or transaction. These analyses weigh the relationship between many data elements to isolate each customers risk or potential, which guides the action on that customer.

Predictive analytics is widely used in making customer decisions. One of the most well-known applications is credit scoring, which is used throughout financial services. Scoring models process a customers credit history, loan application, customer data, etc., in order to rank-order individuals by their likelihood of making future credit payments on time. Predictive analytics are also used in insurance, telecommunications, retail, travel, healthcare, pharmaceuticals and other fields.

http://en.wikipedia.org/wiki/Predictive_analytics

5

HK & AS6Predictive Analytics (cont.)The goal is to analyse your mobile customer demographics (fairly static) to drive bottom-up segmentation (correlation to churn propensity)It is assumed you are ALREADY doing traditional top-down segmentation but are reaching its limits of usefulnessThen to take their behavioral/usage data (from CDRs which is quite dynamic) to arrive at a score for the probability to churn within the given time period for each statistical segmentAt least 1-2 quarters (3-6 months) of Call Data Records (CDRs) are needed for the predictive engine to be effective but the more the betterTo capture seasonal variations around festivals/holidays etc, 12-18 months is required (4-6 quarters)This will be filtered against the list of high value (ARPU or profitability/CLV) subs to get those worth retainingThis is but one application for predictive analytics, others include:Cross-sell & Up-sell opportunities (likelihood to buy)Credit scoring for setting dynamic limits and risk managementFraud detection (post-paid only)

6

HK & AS7Acquisition (COCA)COCA = Cost of Customer/subscriber Acquisition (also called COSA)COCA has at least 3 components for most telcos:Channel Margins per customerLower margins are more efficient for COCO (Company Owned & Company Operated) stores ONLYFranchisee/retail partners need incentives (higher margins) to push your products/servicesShould be between 1/4th to 1/3rd of your COCA (25%-33%) for telcos (anything above 34% is a red flag)

Handset SubsidiesOnly relevant if handsets are bundled with contracts (post-paid)Also if handsets are locked to your network (portability)Should be below half your COCA (< 50%) and if its above, thats another red flagAdvertising/MarCom costs per subscriberIncludes all costs of MARketing COMmunicationsShould be below 1/4th (over 25% would again be a red flag)WarningWarningWarning

7

HK & AS8Servicing (Slide 7 of 15)There are costs associated with servicing your customersThe number of times they contact your call centre could mean the difference between a subs worth retaining or not at the SAME ARPUIf you have retail outlets, each time they walk-in, you will incur costs which need to be accounted forOnly after these ongoing costs have been factored in will you get a true picture of your customers profitability or CLV

8

HK & AS9RetentionLoyalty programs (Unsustainable Competitive Advantage)Shotgun approach like airline miles or credit card reward ptsEasier to implement (quick win)Lazy approach thus less effective over the long termWhat will you do when your competitors also offer the same rewards (bribes)?What kind of mercenary behavior are you really encouraging from your customers (blackmail / threatening to quit)?Churn prediction & reduction (Sustainable Competitive Advantage)Only focused on those who are likely to leave you which can be lower cost or higher value offers at the same total campaign budget (since the money will need to be divided among fewer subs)Better ROI / EVA but slow win (no results in the 1st quarter)

9

HK & AS10Monthly ChurnThe is the portion of your active subscriber base that goes inactive (via passive/implicit cancellations) each month or explicitly cancels your connection/accountIf this value < 1%/month, you are better off spending your money on other enhancements that your customers are demanding (higher ROI projects)Its critical to track the reasons for churn of your subscriber baseIn most cases the churn for pre-paid is higher than that for post-paid subscribers

10

HK & AS11Churn PredictionVarious statistical models will have varying levels of performance as far as predictive ability goes based on the data you feed them but most should have some kind of feedback loop (self-learning/continual refinement) since you dont want to keep changing your models every year as your customer profile driftsModel PerformanceThis is the models ability to correctly identify customers about to churn out voluntarilyFor telcos this is usually between 65% - 85% with lower rates being more fiscally conservative (lower ROI)Strategy EffectivenessThese are the ratio of churners who actually take up your retention offerThis varies between 5%-15% for most telcos with lower values being more financially conservative

11

HK & AS12Voluntary Churn ReductionYou can only reduce voluntary churn NOT eliminate it entirely (due to diminishing returns)You CANNOT do anything about involuntary churnDeath of the customerMoving / relocating outside your service areaChanging jobs/employers (for company connections)Once you have the reasons for churn, you can focus on the voluntaryUntil you get the reasons, a rule of thumb is that voluntary churn is usually around 2/3rd 3/4th of total monthly churn for most telcosThe higher your voluntary churn, the more room for improvement (better ROI)You should target to bring down the ratio of voluntary churn to about: in the short term (1-2 quarters)1/3rd in the medium term (1-2 years)1/4th in the long termIf your voluntary churn is already below 25% of total churn, spend your resources elsewhere (you are in pretty good shape for now) !

12

HK & AS13Fears, Uncertainties, Doubts(FUDs)Why should a model built in the West work for Africa?My customers are uniquely differentWhat if I spend money to find out something I already know?What if the retention campaigns do not reduce my churn?How will my staff get trained so we are not dependent on outsiders/vendors to keep going?Has this been implemented anywhere else in Africa?Who are the reference clients?Who are the partners?

13

HK & AS14Critical Success Factors(CSF)Retention MUST be a top-down initiative since it requires the assistance of many different deptsThe CFO, CMO, CIO/CTO must be involved at all stages of the project for support & buy-in by forming a steering committee that meets weekly initially and then monthly to review progress/milestonesThe CMD/CEO must lay out the vision and drive the organisational changes needed to support this initiative

14

HK & AS15Our Partners (Slide 14 of 15)Cranes Software14 years old with 600 employeesUS$ 60 Million in revenuesStatistical Consulting (Predictive Analytics)Bangalore, IndiaSiemensGurgaon, India

15

HK & AS16SummaryReview the Telco Readiness Checklist before jumping into predictive analyticsMeet the benchmarks (for this effort to make financial sense for your enterprise):Base >= 1 million active subsVoluntary Churn >= 1%/monthVoluntary/Total Churn > 25%3-6 months CDRs minimum for analysis and modelingThe CSF are prerequisites for any such initiative to kickoff

16

HK & AS17Questions?

Thank you!

HK & AS

2007 April

17

HK & AS18AppendixWhat is CLV / LTV and its components (HBR)?What is Customer Equity and how do we measure it (HBR & Wikipedia)?What is Economic Profit (McKinsey)?What are Social Networks (SN & ASN)?What is Data Mining of CDRs (DM & KD)?What is Residual Customer Value (RCV)?

18

HK & AS19Churn Attrition DefectionCustomer attrition, also known as customer churn, customer turnover, or customer defection, is a business term used to describe loss of clients or customers.Banks, telephone service companies, Internet service providers, pay TV companies, insurance firms, and alarm monitoring services, often use customer attrition analysis and customer attrition rates as one of their key business metrics (along with cash flow, EBITDA, etc.) because the "...cost of retaining an existing customer is far less than acquiring a new one." Companies from these sectors often have customer service branches which attempt to win back defecting clients, because recovered long-term customers can be worth much more to a company than newly recruited clients.Companies usually make a distinction between voluntary churn and involuntary churn. Voluntary churn occurs due to a decision by the customer to switch to another company or service provider, involuntary churn occurs due to circumstances such as a customer's relocation to a long-term care facility, death, or the relocation to a distant location. In most applications, involuntary reasons for churn are excluded from the analytical models. Analysts tend to concentrate voluntary churn, because it typically occurs due to factors of the company-customer relationship which companies control, such as how billing interactions are handled or how after-sales help is provided.When companies are measuring their customer turnover, they typically make the distinction between gross attrition and net attrition. Gross attrition is the loss of existing customers and their associated recurring revenue for contracted goods or services during a particular period. Net attrition is gross attrition plus the addition or recruitment of similar customers at the original location. Financial institutions often track and measure attrition using a weighted calculation called Recurring Monthly Revenue (or RMR). In the 2000s, there are also a number of business intelligence software programs which can mine databases of customer information and analyze the factors that are associated with customer attrition, such as dissatisfaction with service or technical support, billing disputes, or a disagreement over company policies.http://en.wikipedia.org/wiki/Customer_attrition

19

HK & AS20Customer Lifetime Value (CLV)The net profit a company accrues from transactions with a given customer during the time that the customer has a relationship with the company.RT Rust & KN Lemon, HBR (Sept 04); pg 112-113This implies FIRST having a consolidated/unified view of our customersCurrently Im viewed as 5 independent subs for my FLP, FWP, VCC, BB etc.Then we must put in metrics to track the costs of servicing EACH customerNumber of calls, emails, visits to retail stores/shops etc.

20

HK & AS21CLV (cont.)Net Profit = Total Revenues generated over a customers lifetime Total Cost (direct & indirect) for that customerTotal Cost = Service Costs + Retention Costs + Defaulted AmountsService Costs = Contact Costs + Repair CostsContact Costs = Email, Fax, Phone, visits to RWW/WWERetention Costs = Discounts + Upgrades + Loyalty BonusesCLV = LTV = EP Customer Lifetime Value = Life-Time Value = Economic Profit

21

HK & AS22CLV (cont.)In marketing, customer lifetime value (CLV), lifetime customer value (LCV), or lifetime value (LTV) is a metric that projects the value of a customer over the entire history of that customer's relationship with a company. Use of customer lifetime value as a marketing metric tends to place greater emphasis on customer service and long-term customer satisfaction, rather than on maximizing short-term sales.Customer lifetime value has intuitive appeal as a marketing metric, because in theory it allows companies to know exactly how much each customer is worth in dollar terms, and therefore exactly how much a marketing department should be willing to spend to acquire/retain each customer. In reality, it is often difficult to make such calculations due to the complexity of the calculations, lack of reliable input data, or both.The specific calculation depends on the nature of the customer relationship. For example, companies with a monthly billing cycle, such as mobile phone operators, can count on a reasonably reliable stream of recurring revenue from each customer. Car manufacturers, on the other hand, have less insight into when or whether a customer will make a repeat purchase. http://en.wikipedia.org/wiki/Customer_lifetime_value

22

HK & AS23CLV (cont)

23

HK & AS24CLV (Net Profit Total Rev)

24

HK & AS25CLV (NetP TRev CapEx)

DISGUISED CDMA TELCO EXAMPLE

25

HK & AS26CLV (NetP TRev OpEx)

26

HK & AS27NP TR OpEx Variable

DISGUISED CDMA TELCO EXAMPLE

27

HK & AS28NP TR OpEx Var Voice

28

HK & AS29NPTROpExVarVoiceTariffs

29

HK & AS30CLV (NP Total Costs)

30

HK & AS31CLV (NP TC Service Costs)

DISGUISED CDMA TELCO EXAMPLE

31

HK & AS32Customer Equity (CE)The sum of the lifetime values of all the firms customers, across all the firms brandsRust & Zeithaml, HBR (Sept 04)Customer Equity is the Net Present Value of a customer from the perspective of a supplier. It can - and should - also include customer goodwill that is normally not expressed in financial terms, eg a customer's level of loyalty and advocacy. http://en.wikipedia.org/wiki/Customer_equity Maximising Customer Equity should be the PRIMARY goal of ALL firms to ensure long term successALL other measure including Brand Equity are secondary

32

Source: McKinsey (2004 Q3)33Calculating Economic Profit

Economic Profit (EP) is a measure of the current profitability of individual customers and is stated as monthly figure

Revenue

ARPU

Costs

Interconnect (outgoing minutes)

Network usage (total minutes)

Collections cost (by credit category)

= Economic Profit

Interconnect (incoming minutes)

Cost of store (FSD interactions)

Call center cost (call center interactions)

Bad debt (amount overdue)

Billing and IT (fixed per sub)

G&A and others (fixed per sub)

33

Source: McKinsey (2004 Q3)34Lifetime value drivers and associated revenues & costs Cumulative customer lifetime value in dollars

11987654321

10

ConsiderationAcquisition costRecurring revenueCash cost to serveCross-sell/up-sellCredits & adjustmentsRenewal promosMigrationChurnBad debtWin-backMonths of subscription life

123412133637383940

CostRevenuesCustomer joins (rejoins) serviceCustomer leaves service

Differentiated cust. treatment should be applied to the 11 value drivers across the cust. lifecycle

34

Source: McKinsey (2004 Q3)35

Insight:20% receivers-only seem as inactive subscribers, yet generate incoming revenues and can potentially be switched to outgoing usage

SubscriberdistributionMonthly minutes of usePercent of subscribers, I/B and O/B MoU by quotient of inbound vs. outbound MoUPercent of total postpaid subscriber base, MoU

Quotient of inbound MoU / outbound MoU

Fraction of subscribers who use phone as receiver only much larger than in postpaid

I/B and O/B develop in exact opposite way to postpaid - O/B constant, I/B rising in prepaid20% of prepaid subscribers use phone as receiver ONLY & do not generate outgoing callsDISGUISED CLIENT EXAMPLE

35

Source: McKinsey (2004 Q3)36

Number of recharges2)

Significant portion of prepaid acquisitions have never recharged

Subscribers recharging at least once, recharge an average of 4.1x in the following 6 months

Number of recharges done by customers of one cohort1)Percent of the cohort customers

All prepaid customers acquired in specific calendar month Number of recharges per subscriber

Insights:Trial offer of prepaid recharge card directly with bundle at special priceSpecific recharge offers for infrequent recharge customersAlternative recharge methods based on customer locationBreaking the 1st time recharge barrier presents an opportunity for increasing prepaid revenuesDISGUISED CLIENT EXAMPLE

36

Source: McKinsey (2004 Q3)37

First recharge stimulation provides ARPU upliftTest results from standardized ROI-reporting

5% bonus on recharge to new customer two months after activation contacted with SMS(ARPU in EUR)

Launch of campaignAverage revenue lift of 0.42 (Rs. 23) in months 0-3*

Standardized post-campaign reporting available in June 2004

Target group

Control group

Test designOffer: 5%-10% on next recharge within 30 daysCommunication: Target group contacted with SMS or mailingTarget group: New customers with 2-6 months since activation without first rechargeCampaign design: 9 different sub-campaigns defined

Submitted to Test Environment and tested in Feb / March 2004

Months relative to campaign drop-dateDISGUISED CLIENT EXAMPLE

37

Source: McKinsey (2004 Q3)38

Clusters constructed per network age-group, according to:- Existing segment - Handset model- Current rate plan- Current usage- Economic profitCalculating CLVCompile migration matrix

The second step documents, with the help of a "migration matrix", the cluster to which cust from a certain network age-group will migrate the following year, and how many cust migrate overall (a migration occurs when one of the parameter changes Migration frequency derivation

On the basis of the calculation of migration frequency, an assertion can be made about how probable it is that a cust in the next network age-group will belong to a particular cluster (e.g., probability that a cust in cluster B this year, will be in cluster K next year)Calculate CLV (per cluster)

The migration probability is now applied to project the expected lifetime path of the customer. CLV is calculated by multiplying the average realized EP for each cluster with the path/migration probabilityBCD

Firstly, on basis of historical data, all cust are allocated to network age-groups then, the cust in each network age-group, on the basis of their existing segment, handset model, and probability to churn, are allocated to clusters having similar characteristics Create clusters having similar characteristics 10%

5%

85%AbwanderungCluster KCluster B10%

5%

85%AbwanderungCluster JCluster K10%

5%

85%Abwand-erungCluster KCluster BCluster BA

Assuming no dis-continuity in market forces

38

Source: McKinsey (2004 Q3)39

Calculating CLV

CLV for network age group N (the next to last age group with none zero EPs):CLVN/R = EPN+1/R

CLV for network age group N-1:CLV(N-1)/R = [ fRA* (EPN/A + CLVN/A ) + fRB * (EPN/B + CLVN/B) + ... ] / (1 + i) Key:fRA = Probability, that the customer in following year (i.e., at N) will be in clusterN/Ai = Discount rate Calculation of CLV per cluster for different network age groups

39

Source: McKinsey (2004 Q3)40

First step:Define subject areas in data martCreate tables within each subject areaDefine all data fields including code, name, data type, length and other attributesIdentify primary keys** for each tableSecond step:Create references or links/dependencies among tablesDefine foreign keys for applicable tables and check consistencyThird step:Generate scripts automatically for physical database creationModify the model accordingly if syntax errors detected

DBA executes the scripts to create the database*E.g., Sybase Power Design**A unique identifier of an entityCLM datamart is generatedusing a case tool*

40

Exploiting Social Networks

(via Data Mining of Telco Call Data Records)

41

HK & AS42OverviewWhat is a Social Network (SN)?Network ClassificationAugmented Social Networks (ASN)Mapping SN (Visualisation)Pattern Recognition (Interpretation)Social Network Analysis (SNA)What is Data Mining (DM)?How do we mine CDRs?What types of customer behavior are we interested in?What are the applications in DM of CDR?Customer Lifetime Value (CLV)Residual Customer Value (RCV)

42

HK & AS43Definition of SNA social network (SN) consists of any group of people connected through various social familiarities ranging from casual acquaintance to close familial bonds. Members of a social network may not have any real awareness of the network as a whole.

The rule of 150 states that the size of a genuine/functional social network is limited to about 100-150 members.

Social networks are often the basis of cross-cultural studies in sociology and anthropology. The rule of 150, mentioned above, arises from cross-cultural studies in sociology and especially anthropology of the maximum size of a village (in modern parlance most reasonably understood as an ecovillage). It is theorized in evolutionary psychology that the number may be some kind of limit of average human ability to recognize members and track emotional facts about all members of a group. However, it may be due to economics and the need to track "free riders", as larger groups tend to be easier for cheats and liars to prosper in. Either way, it would seem that social capital is maximized by this size. http://en.wikipedia.org/wiki/Social_network

43

HK & AS44Network ClassificationBroadcast (e.g., TV, radio)Linear networks (one to many)Sarnoffs Law: the value of the network is proportional to the number of actorsPaired connections (e.g., phones, fax, email)One to oneMetcalfes Law: the value grows with the square of the number of actors (nodes)Social NetworksMany to many (eBay, Amazon)Reeds Law:when the network allows communities to form then the value grows exponentially with the number of actorsAllen E. (Sept 03) http://www.cybaea.net/Publications/Business%20Platforms.html

44

HK & AS45Profiting from NetworksPlatforms win because of network effects3 developments enable business platformsAn understanding of the potential value of networksThe ability to connect different networksThe business practices to turn potential network value into actual profitsInterconnecting two networks creates value greatly exceeding the combined values of the original two unconnected networks (synergy)Network value chain (NVC)Broadcast Paired SocialLower value Higher value

45

HK & AS46NVCThe higher up the Network Value Chain you can place your business the better The value of your network increases vastlyThe value of business opportunities for joining disconnected networks increases even fasterLeads to a winner-take-all situationThe company benefiting from a larger network can afford to pay more to grow that network (ROI is greater as you scale)The stickiness' of your network increasesSome of the most stressful events in life like shifting homes or changing jobs involve the disruption of social networks

46

HK & AS47Four Cs to increase your networks valueContent (e.g. CNN, Yahoo, Amazon)Content is kingEither info or transactionsCurrent, frequently updated, & relevantSarnoffs Law (publishing)Connectivity (e.g. mobile networks, dating sites)Metcalfes Law (connecting people)CollaborationScalabilityAttracting new networksSarnoff X Metcalfe (cube power)Communities (e.g. Usenet)Reeds Law (social networks)Allen E. (Sept 03) www.cybaea.net/Journal/FourC.html

47

HK & AS48The Laws of Network ValueSarnoffLinear ContentMetcalfSquare ConnectivitySar x MetCubeCollaborationReedExponentialCommunities

48

HK & AS49Augmented Social Nets (ASN)S-Nets with Identification and TrustObjectivesTo create a system that enables more efficient and effective knowledge sharing between people across institutional, geographic, and social boundariesTo establish a form of persistent identity that supports the public commons and the values of civil societyTo enhance the ability of citizens to form relationships and self-organize around shared interests in communities of practice in order to better engage in the process of democratic governance it is a model for a next-generation online community that could be implemented in a number of ways, using technology that largely exists todayIt is a system that would enhance the power of social networks (SN) by using interactive digital media to exploit the transitive nature of trust through the principle of six degrees of connectionSN = who do you knowASN = who do you trust

49

HK & AS50ASN (cont.)The Augmented Social Network (ASN) was proposed in a June 2003 paper presented at the PlaNetwork Conference by Ken Jordan, Jan Hauser, and Steven Foster. The paper makes the case for a civil society vision of digital identity that treats Internet users as citizens rather than consumers. The ASN is described as an Internet-wide system that enables users to find others who have relevant interests or expertise, in a context that engenders trust, so that they can form a social network more effectively. At its core is a form of digital identity that supports appropriate introductions between people who share affinities through the recommendations of trusted third parties. It also supports the distribution of media using the same Internet-wide recommendation system.

http://en.wikipedia.org/wiki/Augmented_Social_Network

50

HK & AS51Mapping Social NetworksNetwork visualisation toolshttp://www.touchgraph.com/TGGoogleBrowser.php?start=gloworld.comYour telcos links to other sites online

51

HK & AS52Social Network Analysis (SNA)SNA is the mapping and measuring of relationships and flows between people, groups, organizations, computers or other information/knowledge processing entitiesThe nodes in the network are the people and groups while the links show relationships or flows between the nodesSNA provides both a visual and a mathematical analysis of human relationshipsManagement consultants use this methodology with their business clients and call it Organizational Network Analysis [ONA].Valdis Krebs, 2004 http://www.orgnet.com/sna.html

52

HK & AS53SNASocial network analysis (related to network theory) has emerged as a key technique in modern sociology, anthropology, sociolinguistics, geography, social psychology, information science and organizational studies, as well as a popular topic of speculation and study.

People have used the social network metaphor for over a century to connote complex sets of relationships between members of social systems at all scales, from interpersonal to international. Yet not until J. A. Barnes in 1954 did social scientists start using the term systematically to denote patterns of ties that cut across the concepts traditionally used by the public and social scientists: bounded groups (e.g., tribes, families) and social categories (e.g., gender, ethnicity). Social network analysis developed with the kinship studies of Elizabeth Bott in England in the 1950s and the urbanization studies "Manchester School" (centered around Max Gluckman and later J. Clyde Mitchell), done mainly in Zambia during the 1960s. It joined with the field of sociometry (begun by J.L. Moreno in the 1930s, an attempt to quantify social relationships. Scholars such as Mark Granovetter, Barry Wellman and Harrison White expanded the use of social networks.

Social network analysis has now moved from being a suggestive metaphor to an analytic approach to a paradigm, with its own theoretical statements, methods and research tribes. Analysts reason from whole to part; from structure to relation to individual; from behavior to attitude. They either study whole networks, all of the ties containing specified relations in a defined population, or personal networks, the ties that specified people have, such as their "personal communities".

http://en.wikipedia.org/wiki/Social_network_analysis

53

HK & AS54

Pattern Recognition (PatRec)PatRec is the art of finding order in often chaotic masses of data.One of the goals of PatRec is to quickly narrow down your set of possibilitiesMacro-filtration (before fine particle analysis)One of the toughest challenges in PatRec is knowing when youve looked at enough info to make a reliable judgment.HBR (Nov 02)

54

HK & AS55PatRec (cont.)Pattern recognition is a sub-topic of machine learning. It can be defined as "the act of taking in raw data and taking an action based on the category of the data".Most research in pattern recognition is about methods for supervised learning and unsupervised learning.Pattern recognition aims to classify data (patterns) based on either a priori knowledge or on statistical information extracted from the patterns. The patterns to be classified are usually groups of measurements or observations, defining points in an appropriate multidimensional space.A complete pattern recognition system consists of a sensor that gathers the observations to be classified or described; a feature extraction mechanism that computes numeric or symbolic information from the observations; and a classification or description scheme that does the actual job of classifying or describing observations, relying on the extracted features.The classification or description scheme is usually based on the availability of a set of patterns that have already been classified or described. This set of patterns is termed the training set and the resulting learning strategy is characterized as supervised learning. Learning can also be unsupervised, in the sense that the system is not given an a priori labeling of patterns, instead it establishes the classes itself based on the statistical regularities of the patterns.The classification or description scheme usually uses one of the following approaches: statistical (or decision theoretic), syntactic (or structural). Statistical pattern recognition is based on statistical characterisations of patterns, assuming that the patterns are generated by a probabilistic system. Structural pattern recognition is based on the structural interrelationships of features. A wide range of algorithms can be applied for pattern recognition, from very simple Bayesian classifiers to much more powerful neural networks.An intriguing problem in pattern recognition yet to be solved is the relationship between the problem to be solved (data to be classified) and the performance of various pattern recognition algorithms (classifiers).http://en.wikipedia.org/wiki/Pattern_recognition

55

HK & AS56Data Mining (DM)an analytical tool that enables business execs to advance from describing historical customer behavior to predicting the future.Martin Morgan, Telecommunications International (May 03)DM enables companies to:Proactively manage business relationshipsDrive growthAnswer complex questions like:Who are your most profitable customers?How can you increase your levels of customer satisfaction, loyalty, lifetime value (CLV or LTV)?Identify business opportunitiesImplement strategies that increase revenueReduce expensesOffer new competitive advantagesSame as Knowledge Discovery (KD) but less sexy and older by a few decades

56

HK & AS57DM (part deux)Its a multi-step process that includes:Defining a business problemExploring and conditioning dataDeveloping the modelDeploying the knowledge gainedTelecom operators must tackle specific business challenges like:Segmenting customers (top-down and bottom-up)Predicting customer propensity to buy (or to churn in the next period)Detecting fraud/abuseIncreasing organisational efficiency

57

HK & AS58What DM is NOT The term "data mining" is often used incorrectly to apply to a variety of other processes besides data mining. In many cases, applications may claim to perform "data mining" by automating the creation of charts or graphs with historic trends and analysis. Although this information may be useful and timesaving, it does not fit the traditional definition of data mining, as the application performs no analysis itself and has no understanding of the underlying data. Instead, it relies on templates or predifined macros (created either by programmers or users) to identify trends, patterns and differences.A key defining factor for true data mining is that the application itself is performing some real analysis. In almost all cases, this analysis is guided by some degree of user interaction, but it must provide the user some insights that are not readily apparent through simple slicing and dicing. Applications that are not to some degree self-guiding are performing data analysis, not data mining.

http://en.wikipedia.org/wiki/Data_mining#Misuse_of_the_term

58

HK & AS59DM (cont)Case studies (just Google these online)Telstra Mobile (Australias largest mobile operator) is reducing customer churn using data mining with SAS Enterprise MinerA european operator calculates revenues and costs for EACH customer so it knows the actual value of each subscriber not just the ARPUA US operator uses DM to ensure that calls are routed effectively by continuous monitoring of performance rules and data analysis of:The history of component & trunk usageCurrent network activity metricsRetention (higher ROI than Acquisition)Cost of keeping an existing customer is 10 times less than acquiring a new one

59

HK & AS60Knowledge Discovery (KD)Knowledge Discovery is a concept of the field of computer science that describes the process of automatically searching large volumes of data for patterns that can be considered knowledge about the data. The most well-known application of Knowledge Discovery is data mining also known as Knowledge Discovery in Databases (KDD).Knowledge Discovery is the process of deriving knowledge from the input data. Some forms of Knowledge Discovery create abstractions of the input data. In some scenarios, the knowledge obtained through the process of Knowledge Discovery becomes further data that can be used for continuous discovery.Knowledge Discovery is a complex topic that can be further categorized according to 1) what kind of data is searched; and 2) in what form is the result of the search represented.http://en.wikipedia.org/wiki/Knowledge_discovery

60

HK & AS61DM MythsExpensive, dedicated DB, data marts or analytic servers are neededCostly to purchase & maintainRequire data extraction for each DM projectMajor waste of timeEnterprise-wide Data Warehouse (EDW) is the solutionFunctions as a customer & operational dbTotal cost of investment is considerably loweredNo need to purchase & maintain additional hardwareMinimise the need to move data in & out of the EDW which is labour-intensiveA US operator got consistent info 90% faster after switching to EDW from fragmented data martsOperator decision are based on actual customer behaviour rather than gut instinct

61

HK & AS62CDR DM ApplicationsCustomer Loyalty & retentionCLV & residual customer value (survival time analysis)Fraud & risk managementTesting various marketing plans to determine ROIFormulating new plans based on identified calling patternsOptimising network utilisationOSS analysisTrend forecastingReal-time traffic analysisCredit scoringPost-paid customers (outstanding balance caps/limits)

62

HK & AS63CLV & RCVCustomer Lifetime Value (CLV)the net profit a company accrues from transactions with a given customer during the time that the customer has a relationship with the company.HBR (Sept. 04)Residual Customer Value (RCV)The remaining net profit that can be accrued from a given customer (could also be expressed as a % of CLV) to help determine if its worthwhile to try to retain themCustomer Equity (CE)The sum of the CLV of all the firms customers across all the firms brandsBrand Equity (BE)The sum of customers assessments of a brands intangible qualities, positive or negative

63

HK & AS64CLV (cont)CLV = VE + BE + REVE = Value EquityBE = Brand EquityRE = Relationship EquityVE is the objectively considered quality, price, and convenience of the offeringBE is the customers subjective assessment of a branded offerings worth above and beyond its objectively perceived valueRE is like the switching costs the customers reluctance to go elsewhere because of learning curves, community benefits, relationships with salespeople, etc.

64

HK & AS65

TCCTCR & TCCCLV = NPV (TCR TCC)TCR = Total Customer RevenueTCC = Total Customer CostsAC = Acquisition Costs (initial/capex)RC = Retention Costs (ongoing/opex)TCC = AC + RC = CCTCR = CR

timeCCCR

TCRRs.

65

TCV = CV (area under the CV graph)RCV = area under CV graph AFTER time t = TCVtIf RCV > RC, keep customerValue of retention incentivesmust be < RCV - RCIf RCV 1/4th 1/3rd (25% - 33%)Relationship equityAir travelRental carsAny service that involves loyalty programsIf post-paid / contracts have any form of redeemable reward points then they would fall into this category

67

Chart311122484392784166416525125326362166474934312886451225698172951210100100010241112113312048

SarnoffMetcalfS x MReed

Sheet1YearEvolutionaryRevolutionary2005100%3%0.5200650%6%200725%13%200813%25%20096%50%20103%100%

Sheet1

EvolutionaryRevolutionaryYear

Sheet3SarnoffMetcalfS x MReed11122484392784166416525125326362166474934312886451225698172951210100100010241112113312048121441728409613169219781921419627441638415225337532768

Sheet3

SarnoffMetcalfS x MReed

Sheet2RevenueCost0.110.50.150.50.2250.250.33750.1250.506250.06250.7593750.031250.90.01562510.00781250.950.00390625

Sheet2

RevenueCost