Customer Segmentation in Retail Banking Industry Why Retail Banks should implement Data Mining Techniques?

Eduarda Montalvão Baptista

Professor Rute Xavier Advisor

Dissertation submitted in partial fulfilment of requirements for the degree of MSc in Business Administration, at the Universidade Católica Portuguesa, September 2014

ABSTRACT One of the most methodologies tool companies have to assure customer satisfaction and loyalty is customer segmentation. The purpose of this research is to explain what may be the most appropriate method to segment retail banking customers and how it can be implemented. Although demographic data is still the most commonly characteristics analysed by retail banks, behavioural segmentation as proved to achieve better results as it takes into account the heterogeneous patterns of services usage and consumption among individuals with the same age and income. Nevertheless, retail banks are the organizations by excellence mostly capable to apply this method, as it operates in the sector of activity with more organized and complete information about customers. Not only are known the consumption patterns as well as banks have access to information about customers’ income and attitudes towards investment and saving products. A comprehensive analysis was perform to assess the current state of the art of transaction processing system in Portugal, its operational model and how financial information could be collected. Given the extensive portions of data available, it was identified a technological solution to assist the information treatment. Data mining strategic implementation plan was drawn and this dissertation also assesses all the advantages of its integration by retail banks.

ABSTRACT (Portuguese) Uma das melhores ferramentas que as empresas possuem para garantir a satisfação e fidelização dos clientes é a segmentação. Esta dissertação tem como principal objetivo explicar qual o método mais apropriado para segmentar os clientes bancários e a como este poderá ser implementado. Embora os dados demográficos sejam ainda os mais utilizados pelos bancos, a segmentação comportamental tem demonstrado melhores resultados, uma vez que tem em consideração a heterogeneidade nos padrões utilização de serviços e de consumo entre indivíduos com a mesma idade e rendimento. Ainda assim, os bancos são as organizações por excelência com maior capacidade para aplicar este método, uma vez que atuam no setor de atividade com maior quantidade e mais completa informação sobre os seus clientes. Os bancos não só conhecem os padrões de consumo, bem como têm acesso a informações sobre rendimento e atitudes em relação a produtos de investimento e poupança. Foi realizada uma análise abrangente sobre a situação atual do sistema de processamento de transações em Portugal, o seu modelo operacional e de como a informação financeira pode se encontra disponível. Dado o grande volume de informação financeira, foi identificada uma solução tecnológica para auxiliar no tratamento da informação. Foi estabelecido um plano estratégico de implementação das técnicas de data mining e esta dissertação avalia também as vantagens da sua integração pelos bancos.

I

ACKNOWLEDGEMENTS I want to use this opportunity to express my gratitude to everyone I had the pleasure to work and interact with during the Master degree. From the University staff to the students I met, I am very grateful for all the inspiration and friendly advices. This journey would not have been the same without you all. Professor Rute Xavier, thank you for your guidance throughout the thesis and the consulting project. All the meetings, conversations, different perspectives and wise thoughts shared had an extremely positive impact on all the work developed and particularly in this thesis. I will always remember all the advices you gave me relating to my professional and even personal life. This dissertation would not have been the same without the knowledge and expertise transmitted to me by the professionals I had the chance to work with during the consulting project. Furthermore, I am very thankful for the patience and constructive criticism that made the delivered project better and also enabled me to grow as a professional. Raquel, my consulting project partner and friend, you always inspired me with your positive attitude. It was an absolute enjoyable experience to work with you. Furthermore, it would not be possible to conclude the master and this thesis without my family support. Thank you for allowing me to live this experience, motivating me every single day, in order to pursue my dreams. Finally, thanks Pedro for your creative contributions and critical analysis. I will never forget your dedication and assertive answers that allowed me to surpass the obstacles ahead. Your perseverance and positive attitude were undoubtedly determinant for me to continue to work every day.

II

CONTENTS ABSTRACT

I

ACKNOWLEDGEMENTS

II

CONTENTS

III

LIST OF FIGURES

V

1. INTRODUCTION

1

2. LITERATURE REVIEW

2

2.1.

Market Segmentation

2

2.1.1.

Traditional Methods for Market Segmentation

3

2.1.2.

Modern Methods to Market Segmentation

4

2.1.3.

Previous Retail Banking Segmentation – Examples

6

2.2.

Technological solution for customer segmentation

2.2.1.

Data Mining Techniques

3. METHODOLOGY

7 7 9

3.1.

Interviews

10

3.2.

Consulting Company’s Reports

10

3.3.

Consulting Project

11

4. TECHNICAL ANALYSIS: PAYMENT TRANSACTION PROCESSING AND DATA MINING 4.1.

Portugal Payment Transaction’s Processing Overview

12 12

4.1.1.

Four Corner Model

13

4.1.2.

Transaction Message’s Anatomy and Information Transmitted

15

4.2.

Data Mining – Implementation in Retail Banking Industry

18

5. STRATEGIC FACTORS FOR CUSTOMER SEGMENTATION AND DATA MINING IMPLEMENTATION

20

5.1.

Advantages of Customer Segmentation and Data Mining Techniques

20

5.2.

Limitations and Solutions Proposal

22

6. CONCLUSIONS

23 III

6.1.

Further Research

25

REFERENCES

26

APPENDIX

28

Appendix 1- Classification Framework of Data Mining Techniques Applied to CRM

28

Appendix 2 - Bank of Portugal Four Corner Model Representation

28

Appendix 3 - MasterCard Anatomy of the Transaction Representation including Clearing and Settlement Process

29

IV

LIST OF FIGURES Figure 1 – Four Corner Model Representation Adapted

13

Figure 2 – Adapted Representation of Transaction Messages

16

Figure 3 – Overview on the Steps that Compose the KDD Process

18

Figure 4 - % of Articles Publish By Subcategory from January to July 2014

22

V

1. INTRODUCTION Retail banking industry has an extremely competitive market. With many different players operating at a national and cross-border level, it is important to pursue strategic and dynamic plan of operations, in order to differentiate and better satisfy customers’ needs. This objective will only be assured if retail banks follow and implement the ultimate trends and innovations within the market. One of the major concerns of every financial institution is to improve their customers’ experience, as it will lead to successful customer retention and loyalty. And how can this be probably achieved? The answer is simple: through customer segmentation. The aim of this research is to understand which is the best way to accomplish a significant segmentation of retail banking customers. Therefore, a comprehensive research of current and traditional methods used and promissory evolution to modern and technologically supported segmentation methods was developed in the literature review chapter. It tries to represents all the valuable research that has been done so far by the Academy on this subject. Unlike the demographical segmentation that is still widely used in retail banking industry, the behavioural segmentation presents better results as it directly addresses customers’ service usage patterns and consumption habits. Nevertheless, financial institutions in general are the organizations by excellence mostly capable to apply this method. Banks operate in the sector of activity with more organized and complete information about customers than any other industry, comprehending not only the consumption patterns through cards transactions but also income patterns and attitudes towards investment and saving products. Simplifying the explanation of this fact, banks are able to see the whole picture. To properly evaluate how banks have access to financial statements and which type of data is available, it was studied the transaction processing system in Portugal and the operative model of the network. To complement the analysis, it was also proposed an implementation strategy of data mining techniques, becoming at this moment able to respond to the following research questions: What is the payment transactions processing stage of evolution today?

1

What customer financial1 data is available for retail banking industry? How financial data can be analysed to improve customer segmentation? The answer to the former questions lead to the conclusion that the main research question is trying to address: Why retail banks should implement data mining techniques? In the chapter 6, the principal advantages of segmenting retail banking customers through the application of data mining techniques are disclosed. This dissertation is aimed to add value to the Academy, as well as to the financial institution the author worked for developing a consulting project. The present research was certainly enriched by the contribution of the experts of the financial sector interviewed, in which the client’s professionals are included. The former methodology consisted in a review of reports and papers developed by a renowned financial consulting company revealing the best practices and trends on transaction processing systems. The last chapter containing the conclusion assesses what should be the strategic path to follow, as it is believed is the future for the financial sector in Portugal and correspondent customer segmentation.

2. LITERATURE REVIEW 2.1.

Market Segmentation

At the beginning of the industrialized world, goods were produced according to manufacturing oriented strategies, taking advantage from economies of scale and the reduction of production costs. As the technology evolved, production processes became more flexible and companies gained some awareness about the diversification of demand, creating products specifically designed for each submarket, thereby obtaining a competitive advantage (Wedel and Kamakura, 2000). The market segmentation concept was then introduced along with marketing research in 1956 by Smith, when it assumed an important role on the identification of customers’ needs. Taking into account consumers’ characteristics and crossing variables, sellers started to identify broad classes of buyers and adopted more 1

By financial data, the author means purchase/transactions patterns through cards.

2

standardized strategies (Kotler, 1980). Furthermore, market segmentation or customer segmentation can be defined as the marketing strategy that aggregates customers with common needs, characteristics or similar response and behaviour towards a product or service (Lamb and McDaniel, 2003). To identify a group of customers that forms a segment, the group has to be homogeneous in the sense that, individuals within the group have to share one or more common criteria. Individuals within a group have also to present distinctive characteristics from individuals belonging to other groups, exhibiting explicit heterogeneity between groups (Smith, 1956). An unaccountable number of variables can be used to segment the market, as marketers may use whatever variables they consider appropriated and properly measurable for the market in analysis. Therefore, all characteristics available for studying markets are categorized in four proposed segmentation basis: 1) geographic segmentation, which includes geographic location and population density; 2) demographic information, like age, gender, income and educational level; 3) psychographic characteristics, such as lifestyle, interests and values; 4) and behavioural segmentation, which comprehends every element that explain customer’s behaviours from expected benefits to usage rates (Kotler and Armstrong, 2009). 2.1.1. Traditional Methods for Market Segmentation In the 1950’s and 60’s, the market segmentation was predominantly based on demographic characteristics, as there were no advanced resources to collect other type of data and this demographic information could be easily collected and be readily available. The most commonly demographic information collect was age, gender and geographic location. In terms of financial services, banks traditionally distinguish at a first stage, corporate customers from retail customers. The corporate customers were then 3

segmented according the sector of activity and to the extent of its operations (regional, national or international). Personal retail customers were mostly classified using information such as age, profession/income and wealth (Meidan, 1984; Harrison 1994). The segmentation method applied in this case is called a priori segmentation, as segments are already known and its size and characteristics are estimated within a certain population (Green, 1977). Although it still is a valid procedure widely used by financial institutions, consisting in a base to the segmentation method, it does not comprise the customers’ different personalities and attitudes towards the services provided. A practical example will demonstrate it. Applying demographic segmentation, banks could implement customer segmentation by profession and income. As a result, tracing the profile of a private doctor with considerable buying power, it will possibly present different outcomes. In one hand, the individual could have a large appetence to use his computer and internet, being perfectly informed about internet security standards, having access to his online banking account, where he performs almost all banking related operations. He could also present an urge for high investment risk, having no interest in other low risk and secure types of investment. Nevertheless, the same doctor could also be a personal interaction seeker, who always goes to the branch to perform all operations, being sceptical towards technologies and internet and opting to invest in low risk products and retirement saving products (Machauer and Morgner, 2001). As proved by the example and many studies on consumer goods, market segmentation based just on demographic characteristics poorly addresses the problem that companies try to solve and complementary type of information is needed. 2.1.2. Modern Methods to Market Segmentation Modifications in terms of regulatory restrictions in the financial sector during the 1980’s lead to the expansion and proliferation of products provided by financial institutions. Marketing assumed an important role with some institutions recruiting marketers with experience outside the financial services in order to strength their marketing departments (Ennew et al, 2000). In contrast, since the early 1990’s the financial sector have been experiencing a recession cycle, with reduced customer 4

spending, increasing competition and market saturation. Marketing strategies gained even more importance, being a way to capture and retain more effectively customers, as profit margins dropped significantly and financial services were rationalized (Kitching, 1982). Modern approach to segmentation suggests the study of the population based on a set of relevant variables, finding similarities between individuals, called post hoc or cluster-based segmentation. In opposition to a priori segmentation, the clusters are identified just after completing the cluster analysis and designed accordingly to the conclusions (Green, 1977; Wind, 1978). Moreover, multivariable profiles are traced involving more complex information, such as attitudes and purchasing behaviour, since technology evolved and now enables the collection of more than demographic characteristics. In this sense, the central problem of a priori segmentation is solved, as there was no proved link between the segments established and customer’s behaviour (Speed and Smith, 1992). Psychographic segmentation has no consensual definition, since it has been developed over the time (Ziff, 1971). It started as a segmentation based on basic personality characteristics, but also involves attitudes, interests, lifestyles, opinions and values (Kotler and Armstrong, 2009). Psychographic variables are important to market segmentation as it supports the understanding about customers’ motivations and perceptions (Harrison, 1994). Life-stage is also considered part of psychographic characteristics, but its use has been criticized as it does not have in consideration the psychological element from age that may differ from the biological age (Speed and Smith, 1992). Nevertheless, behaviours were considered the most appropriate variable to study retail banking customers, in a way that different behaviours and usage patterns suggest different requirements, expected benefits or utility from usage of each customer (Elliott and Glynn, 1998). Behavioural segmentation uses customer’s knowledge and respond towards a product or service as variables. It involves benefits sought, the usage rate and brand loyalty, as well as user status (regular, non-user or potential user, among others), readiness to buy and occasions (situations that lead to a need of use or purchase) (Kotler and Armstrong, 2009).

5

2.1.3. Previous Retail Banking Segmentation – Examples As mentioned before, demographic variables have traditionally been used to segment the retail banking market. Improvements to segmentation methods lead to different studies suggesting alternative segmentation comprising different bases. A psychographic segmentation focused on “individuals’ own perceived knowledge and understanding of financial services, the perceived confidence and ability in dealing with financial matters and the expressed level of interest in financial services”. Four customer segments were identified called “financially confused”, “apathetic minimalists”, “cautious investors” and “capital accumulators”, based on attitudes towards financial services and financial maturity (Harrison, 1994). In a behavioural perspective, a study was developed based on usage patterns and frequency to label four segments as “traditional”, “convenience”, “investment” and “debt” (Burnett and Chonko, 1984). Ten years later, other segmentation is proposed using customer’s needs as one of the variables studied. Two different segments were identified: “convenient”, in which customers appreciated the convenience of the branch locations, its opening hours and the existence of automated teller machines, and “performance”, in which customers appreciated the competence of the bank’s employees and the efficiency in terms of operations (McDougall and Levesque, 1994). Another study characterized by attitudes and expected benefits was developed, this time having into consideration the element technology and internet as a source of information to decision making process. Four new segment were identified: “technology opposed” which characterizes customers that value individual service in traditional branch and are sceptical towards technologies, “service oriented” which characterizes customers that value information and consulting services more than technology and electronic transactions, “transactional oriented” which characterizes customers that have lack of interest towards information services and value technology and online transactions and “generally interested” which characterizes customers that manifest interest for transactions as well as information services.

6

2.2.

Technological solution for customer segmentation

With constant shifts in buying behaviours and more demanding and informed customers, segments are not static and tend to be subject of changes over the time (Fonseca and Cardoso, 2007). Therefore, traditional processes of segmentation are being challenged by the new approach of one-to-one marketing. Moreover, no previous segments proposed for retail banking is based on consumption patterns. The time when card’s purchases were authorized through phone calls has passed. Nowadays, banks have extensive databases full of customer’s consumption habits that are not being used. Even more, banks have all the information necessary to predict these behaviours, from disposable income to credit cards expenses and they are just not capable to analyse all the movements registered. Discovering patterns and purchases behaviours of customers or information that visitors drop on websites and online banking would contribute to improved retail banking customer’s profile (Rygielski et al, 2002). In order to accomplish this type of segmentation, a technological solution for data treatment has to be found. A powerful statistical tool as data mining could be implemented to reach hidden patterns that are invisible to a naked eye. 2.2.1. Data Mining Techniques Data mining definition is not consensual, existing different approaches depending on the author’s research about the subject. One of the more recent researches defines data mining as a software with sophisticated data search capability using statistical algorithms to identify patterns and correlations, in order to extract meaningful knowledge from large volumes of data (Kantardzic, 2011). Years before, it was considered as an automatic and semi-automatic mean to work on large portions of data in order to extract some patterns and rules (Berry and Linnoff, 2000). In another direction, some research points out the support role of data mining models for manager’s decisions making process, contributing to the business competitiveness (Hui and Jha, 1999). The objective of application of data mining techniques could be explained by the necessity of attributing new significance to existing data (Chung and Gray, 1999). 7

Data mining can be seen as a complementary approach of other basic data analysis techniques such as statistics, online analytical processing, spreadsheets and basic data access. It started to become widely accepted by organizations as a mean to enhance organizational performance and gain competitive advantage (Hormozi and Giles, 2004). However, the application of this software does not exclude the need to understand the business and data analysed, as well as implicate a moderated need of awareness of general statistical methods. It is not capable by itself to construct and validate new knowledge. There are multiple data mining models that can be applied depending on the business and purpose of application. It could be used to support the decision making process or forecast the impact of decision. For example, it could increase marketing campaigns responsiveness by segmenting customers in homogeneous groups with different characteristics and needs or predict how likely existing customers could opt for a change to a competitor service (Carrier and Povel, 2003). Data mining models generally comprehend seven different models: 1) association rules; 2) classification; 3) clustering, 4) forecasting, 5) regression, 6) sequence discovery; 7) and visualization(Turban et al., 2007). Association rules establish correlations between given records and are usually applied to cross-selling techniques, when a given company wants to sell additional products to existing customers. Classification implies the construction of a predictive model for future customers’ behaviour, classifying predefined characteristics observed. 8

Clustering differentiates from classification as the clusters are not known or predefined in the beginning and relies only on data mining process to identify groups with similar characteristics and patterns. Forecasting estimates future values based on past and current numbers and is most commonly used to estimate demand and sales. Regression is a statistical estimation that models relationships between variables and test scientific hypothesis. Sequence discovery implies the identifications of patterns over the time. Finally, visualization is used in conjunction with other data mining models to provide clear understanding of complex data patterns, making use of specific software to display simplified view. The usage of each data mining method has been studied to complement customer relationship management techniques, comprehending specific objectives such as customer identification, customer attraction, customer retention and customer development2. In this way, analysis these four dimensions will result in a comprehensive study about customers, understanding and maximize their value for the company in the long term. For the retail banking industry, the most suitable models are association rules, classification and clustering analysis as the data presents “moderate unitarily amounts, large databases and exchanged files and relatively standard form” (Tufféry, S., 2008). By this, it means that the information involved and the method on which payment transactions occurs is almost the same, with little variability in terms of amounts in which almost all processing systems do not surpass an established limit.

3. METHODOLOGY Different research methods were used accordingly to the research questions proposed. In the following chapter, it will be introduced the methodology applied to provide an answer for the three research questions:

2

See Appendix 1- Classification framework for data mining techniques applied to CRM

9

What is the payment transactions processing stage of evolution today? What customer financial data is available for retail banking industry? How financial data can be analysed to improve customer segmentation? From the combination of the former questions it will be possible to respond to the main research question: Why retail banks should implement data mining techniques? For this reason, firstly will be displayed the current state of transactions processing system in Portugal. This analysis was performed taking into consideration the knowledge transmitted by the experts in the financial sector through extensive interviews conducted. Secondly, throughout the dissertation will be presented some trends revealed in papers and consulting companies’ reports provided by the professionals interviewed. The information collected from these reports will enhance even more the results obtained. Finally, will be worth to mention the consulting project aims and challenges that contributed, even if not directly, to the present research. 3.1.

Interviews

In order to properly address the research questions proposed, interviews were conducted with professionals from different financial institutions operating in Portugal. These interviews were important at a first stage to define the scope of this dissertation, as it was helpful to understand retail banks’ perspective and interests and concerns in terms of customers’ segmentation. Furthermore, the interviews also served the purpose of building a comprehensive analysis of the Portuguese payment transaction processing system. This way, it was possible to construct a framework on how the payments processing system works and which entities are directly and indirectly involved. In addition, independent users of financial services were also interview to test the awareness and sensibility to this subject. 3.2.

Consulting Company’s Reports

Technical reports and previous marketing research in retail banking segmentation were also insights that contributed to this research. Throughout this 10

thesis were also presented some trends on consumer payments, based on reports and articles from Datamonitor. 3.3.

Consulting Project

The purpose of the consulting project developed in partnership with an external entity was to provide a working experience to researchers, through close participation in the research for solutions to company’s problems. Attending a Portuguese financial institution’s need to optimize their online banking service, the project was developed with the aim of present the state of the art of online banking channel. The bank in question, which would not be revealed due to a confidentiality agreement, asked specifically for an analysis of the current stage of evolution of homebankings in Portugal and to address best practices and trends in this field. A comparative research in terms of image/ communication and functionalities resulted in a benchmark of the national financial industry. The same benchmark technique was performed to analyse the best examples at an international level. Moreover, private and corporate segments at the national homebanking market were evaluated by their users through surveys delivered. The sample revealed a wide variety of banks analysed, covering almost all retail banks operating in Portugal. As a result of the inquiries made, it was possible to conclude the existence of a disparity between the reality of the Portuguese online banking market and needs and wants expressed by online banking customers. Moreover, analysing the results from respondents who were customers of the consulting project client was possible to infer and make some recommendations about improvements on the service. A draft of the online banking platform was developed taking into consideration best practices and trends as well. The consulting project served as dissertation background as the project developed had one main purpose: to improve the customers’ experience. One of the main conclusions of the project was that the communication style and functionalities available should respect the customers’ heterogeneous characteristics. As a result of lack of structured data about their customer’s patterns of online 11

banking usage, the solution delivered included a simplified and standardized homebanking website, adapted to everyone. In the end, this project proved that the bank does not make use of the information available of its customers to adjust their offer. In this sense, the bank also contributed to the present research with insights and its perspective as a key player in the payments transaction processing system.

4. TECHNICAL ANALYSIS: PAYMENT TRANSACTION PROCESSING AND DATA MINING 4.1.

Portugal Payment Transaction’s Processing Overview

The main purpose of the following chapter is to briefly introduce and explain the operative model of card payment transactions to be able to address the research questions earlier proposed and materialize the implementation solution. For the technical analysis in this research, the card payment transactions were chosen among all the different types of transactions available, as it directly demonstrates the consumption patterns of cardholders. The first credit cards were introduced in Portugal in the 50’s to respond to pressures from foreigners that started to visit the country, but the usage of debit and credit cards locally was initiated about 30 years ago when the firsts credit unions and the Portuguese financial processor emerged. Furthermore, the Portuguese transaction system was beneficiated by the credit card activity dynamism and earlier experiences of development and implementation of other countries’ systems. As a result of late development of debit and credit cards transaction system, the Portuguese network, MULTIBANCO, is still considered as one of the most advanced network systems in the world. Consequently, the automated teller machines (ATM) still presents a high number of functionalities comparing with other countries. As a matter of fact, even point-of-sale (POS) terminals have a higher number of functionalities available, although they are less well-known by the common users that mostly perform purchase payment transactions at these terminals. POS network was already born using telephone lines and internet to exchange transaction information. Currently, it uses host-to-host protocols to communicate. A host is generally a physical computer or device connected to a network that is capable to automatically communicate through specific software. Figuratively, it is 12

close to the example of computers that are connected through internet network and use Internet Protocol (IP) to share information. 4.1.1. Four Corner Model Therefore, many different types of card transactions and payment channels could be analysed. For demonstration of the operative model, in this research will be assumed a purchase transaction in-store with debit card issued by a Portuguese financial institution under the Portuguese scheme, MULTIBANCO, designated cards on-us, present at a Point of Sale terminal (POS) in Portugal. The Four Corner Model illustrated by figure 1 is the model that describes how a transaction occurs and who are the main intervening parties in the process. Figure 1 – Four Corner Model Representation Adapted

Source: Bank of Portugal Booklet and Professionals Interviewed

The transaction begins with the action of the cardholder who presents the card to the Merchant assistant to conclude a purchase. Since the moment the card is inserted on the POS terminal until the purchase is accepted passes just a few seconds, but some online messages are exchanged automatically in real-time between the parties to process the transaction. The first stage is called authentication and some validations occur with the payment processor. This entity assures the processing operations, exchanging information between the parties to conclude a transaction. It has a technological infrastructure that keeps the record of all the transactions in the network, as well as the description about all the cards issued with the most common national scheme. 13

Additionally, the processor can also include the clearing and settlement services that secure the corresponding money transfer. The authentication stage comprehends five steps: it is checked if the card is listed in the processor database, if the type of transaction is authorized for that card, if the card is not expired, if the security parameters are correct and lastly if the personal code introduced is valid and the number of attempts to insert it correctly have not been exceeded. In the following stage occurs the authorization process where the transaction is approved or declined before a purchase is finalized and cash is disbursed. The first banking intervenient to be consulted is the Issuer, the bank where the cardholder owns and manages a bank account that issued the card used by the client. The Issuer approves or denies the transaction taking into consideration primarily the account balance, but also if there are any errors in the message received and fraud suspect or earlier card capture requests. In exceptional cases when the Issuer does not respond to the message in time, the processor could have instructions to decide by the Issuer, taking into account the parameters established in the processor’s database, at the time the card was issued. Processors also receive an update file every day, disclosing information about last account balances from the end of the day before of all cards-on-us. This way, the processor is capable to approve or denied automatically the transaction, with lower risk levels of fraud. Finally, the Acquirer, which is the financial institution that accepts the payment transactions made at the POS terminal of the Merchant, is consulted. In some cases, the Acquirer provides the Merchant the POS terminals, even though some financial institutions without acquiring services and telecommunications companies started to develop and distribute their own machines. In this case, these companies are called Supporting Terminal Entities. Once again, when the request for authorization arrives, the Acquirer responds validating the Merchant establishment information, the POS terminals information, the terminal capability to accept the transaction and the Bank Identification Number (BIN) of the card used. If every validation confirms that the transaction is legitimate, the payment is 14

authorized and the card is returned by the store assistant to the cardholder. In the end of the day, there are clearing and settlement processes that transfer the sum of all purchase transactions at that specific terminal to the merchant bank account3. 4.1.2. Transaction Message’s Anatomy and Information Transmitted Every processing system has their own method to codify the messages exchange between the parties involved. In this subchapter will be described the anatomy and the information transmitted in the messages used to process the purchase payment transaction. Each of the messages exchanged consists in different numeric and alphanumeric codes depending on its purpose. Each moment of the transaction process has a designated message and proper codification. Even in cases of cancelation of a transaction or a message received containing errors or unknown codes has a specific answer message. Moreover, every message has a relatively standardize structure that is divided into groups and fields. Each group has a designated number of fields. Each field has to be filled with a predetermined sequence and length, corresponding to a specific number of characters. In case of messages exchange with origin on the processor and destined to issuer, the main composition of the message initiates with the header. The following groups identify the POS terminal, the card used in the transaction and the amount of the transaction. Moreover, the message exchanged between the processor and acquire have in addition other groups that provide data about the authorization process, the international standards of the card when it applies and security parameters. In the other hand, when occurs a responsive message from the issuer or acquire to the processor, the message includes the header, answer information and data about the account balance of the cardholder. The following figure 2 presents an adapted representation with main features of both types of messages. This representation also demonstrates the information transmitted to acquirers and, most important for this research purpose, to the issuer. 3

See Appendix 2 – Bank of Portugal Four Corner Model Representation; See Appendix 3 – MasterCard Anatomy of the Transaction Representation including Clearing and Settlement Process.

15

Figure 2 – Adapted Representation of Transaction Messages

From Processor to Issuer Header

From Processor to Acquirer Header

Header

Message Code

Message Code

Massage Version

Massage Version

Time/Date

Time/Date

Currency Global Terminal Data

Currency Global Terminal Data

Type of Terminal

Type of Terminal

Support Bank code

Support Bank code

Terminal Identification

Terminal Identification

Terminal Capabilities

Terminal Capabilities

Time/Date

Time/Date

Address

Address

MCC

MCC

Country Code Specific Terminal Data

Message Code Message Version Answer Data Response Code Response Number Account Balance Account Balance Currency Variable Data

Country Code Specific Terminal Data

Owner Identification

Owner Identification

District and Municipality Cards Data

Answer Message

District and Municipality Cards Data

PAN

PAN

Card Expiration Date

Card Expiration Date

Type of Authentication

Type of Authentication

Transaction Amount

Transaction Amount

Amount

Amount

Sign Variable Data

Sign Authorization Data Type of Card Authorization Entity Number of Authorization Security Data Pinblock Variable Data

Source: Professionals Interviewed

Tanking into considerations the specific functionality of each group, the header has the function of introducing the message, presenting the code of the transaction, the version of the message, the time and date of the transmission of data and the currency in which the transaction is occurring. In every message exchange about the same transaction, the header stays the same, just modifying the version in cases of transaction cancelation. The POS terminal identification is made by mention the correspondent to the type of terminal, the code of the bank that supports the terminal, the time and date of the transaction and the address inscribed for the location of the terminal. To identify the Merchant involved in the transaction, the message states the 16

legal name of the company, the code of the country, the city and municipality and the merchant category code (MCC). This code consists in a common framework design to classify economic activities at national level. It demonstrates the sector of activity in which companies operate. The legal name of the merchant and the MCC are assigned when the company is legally formed and register at the Institute of Registration and Notary Affairs (IRN). The following group identifies the card by the codes correspondent to the primary identification number (PAN), which is a security measure located in the POS terminal that allows the truncation of the card number. On the customer printed receipt, just the last four algorisms are disclosed and the others are replaced by asterisks as a consequence of this code. The other parameters established are the expiration date of the card, the international brand or EMV (Visa, MasterCard or American Express) standard if applied and the type of authentication. In terms of amount, the parameters just include the amount of the transaction and the signal used. In case of devolutions or chargeback it is used the subtraction sign. As mention before, when takes place a message from the processor to the acquirer, it has also an authorization group comprehending the type of card authorized, the authorization number and the code of the entity giving the authorization. After the authorization data, it is defined the security data that involves the Pinblock, which is the standard algorithm that encrypts the personal identification number (PIN) that the cardholder inserts into the terminal pinpad. Furthermore, the answer messages provide the number and code of the response message and additional information if necessary. The information about account balance codify the amount available on the account at the moment of the transaction, the currency and the amount listed in the updating file send by the issuer in the end of the day before. Finally, the variable that is always present at the end of each message. It is used to append additional requirements that do not have a designated field in other groups. Moreover, this field is commonly used to insert the description of the 17

transaction that appears on customers’ bank statements. 4.2.

Data Mining – Implementation in Retail Banking Industry

Therefore, the entities capable and more interested in access their customers’ consumption patterns are the Issuers. They have access through their hosts to transaction information in a codified form. This subchapter will address the model Issuers could adopt in order to structure and extract meaningful knowledge from transactions processed. As explain in the literature review, data mining is a statistical software capable of identify hidden patters in large databases. Data mining techniques are also connected to a more macro level model known as knowledge discovery in databases (KDD). Figure 3 – Overview on the Steps that Compose the KDD Process

Source: Fayaad et al., 1996

KDD is the process that is described on figure 3. It comprehends several steps that are needed to draw conclusions and new knowledge. Only with the correct implementation of the all process is possible to obtain satisfactory results. Firstly, it involves an understanding about the main objective of the analysis, starting from a large set of samples grouped on a database. The extensive files with codifying messages of all the transaction of a bank represent the starting point for customer segmentation. In the following step, a target data must be set according to the variables and 18

samples that the analyst chooses to study. Applying this concept to the processing market, the target market could be set as restrict sample of cards payment transactions in order to study consumption patterns. Other types of transaction such as money transfers, direct debits, investment and savings products subscription, all sorts of loans grated, among others, can constitute different data targets. The pre-processing activity will define rules and strategies to transform information, handling with cases of missing data or statistical specificities that may conduct to a negative impact in data analysis. For example, in the case of treatment of purchase transaction data, the variable data group on transaction messages must be considered and adapted, in order not to distort the final outcomes. Other potential inconsistencies specifically in alphanumeric codes should be verified. Transformation step consists in setting the proper variables that should be used to conduct the study, which depends according to the research objective. In this stage, data is ready to be submitted to data mining analysis. As a result, patterns will be identified. Taking once again the example of purchase transaction as target data, some variables could be set to study a specific subject and obtain the respective data patterns. For example, setting the variables as essential consumer goods purchases in a specific location and specific Merchant would turn possible to extract, through the usage of data mining process, the customer’s consumption patterns. Introducing the proper interpretation is the final step to achieve new knowledge (Fayyad et al., 1996). In this way, it will be possible to segment retail banking customers through the identification of clusters. More specifically, with interpretation of customer’s consumption patterns, it will be possible to extract consumption habits and offer financial products according to this information. Taking a concrete example, data mining identifies a cluster of customers who present high level of expenditures at certain food retailers. According to this cluster, the bank could then offer credit cards with special discounts for payments done on those same food retailers. The likelihood of acceptance of the product in question will presumably be higher within this cluster, than in a broader database. Achieving consumption habits will also turn possible an improved application of association rules. For example, the bank wants to increase the subscription rate of a fuel card. For this particular case, the target data can be set as the group of 19

customers that have a car loan and current expenses within gas stations. Combining these two variables, the cluster identified by data mining will presumably represent the most likely buyers for this type of financial product. In conclusion, using clusters and association rules identified by the application of data mining techniques, based on higher portions of data and more variables, will lead to better segmentation of banking customers.

5. STRATEGIC FACTORS FOR CUSTOMER SEGMENTATION AND DATA MINING IMPLEMENTATION 5.1.

Advantages of Customer Segmentation and Data Mining Techniques

The advantages of the application of data mining models are a lot vaster that at the first look. The action of analysing big data concentrated in bank’s data warehouses could mean a step forward to improve banks’ performance. Firstly, the identification of clusters will permit the adjustment of financial products offer to customers with a specific profile. Promoting the right financial products to the right customers within a cluster identified would certainly increase the subscription rates, as they proven higher appetence to accept them. This fact will presumably conduct to higher profitability rates. Moreover, studying the patterns that indicates the combined consumption of two financial products through association model, will improve the application of cross-selling techniques. Secondly, an appropriate segmentation and implementation strategy may result in higher levels of customer satisfaction, more profit opportunities and higher customer retention. Therefore, banks should strive for provide as many positive customer experiences as possible, assuring customer loyalty (Sun, 2009). Actually, identification of consumption patterns will also conduct to cost reduction. The usage of patterns extract of the data mining model was already studied for marketing campaigns alignment purpose. Previous research has tested the response to marketing campaign with and without data mining patterns identification for a Canadian bank. The results showed that the positive responses to mass marketing campaigns were less than 1%. In contrast, direct marketing counted 20

with 3% of positive response rate. In the end, the mass marketing campaign presented losses in profitability due to the large customer database and consequently higher costs. For the direct campaign, just 20% of the previous database was identified as likely buyers by data mining. Even taking into consideration the cost of implementing of data mining, the direct marketing campaign presented net profit results (Ling and Li, 1998). As proven by this research, data mining techniques could also be applied to identify more accurate sample to be addressed by marketing campaigns, reducing the waste and costs by better allocation of resources (Wind, 1978). Finally, analysing the consumer payment’s market, it is perceivable a clear tendency of key international players in developing new technologies and solutions for cards industry. The articles released by Datamonitor Financials, the leading consulting company in UK operating as the bank of financial information and reports on financial services industry, were carefully reviewed. These articles describe innovations and improvements implemented by companies on transaction’s processing market. Keeping a close look on almost all key players operating in this sector all over the world, 367 articles were released since the beginning of 2014 on cards subject. It was possible to conclude that for the firsts seven months of the present year, cards channel dominated the payments market attention with 115 dedicated articles, having mobile channel analysis remain in second place. Regarding cards industry and evaluating new trends scan by the consulting company, it is notable that the majority of articles elaborated focused on loyalty and reward cards. After a deeper analysis on these articles, it is possible to affirm that the main trend currently on loyalty cards is card linked offers (Datamonitor, 2014).

21

Figure 4 - % of Articles Publish By Subcategory from January to July 2014

Source: Datamonitor Financials

Card linked offers are represented by the combination of loyalty programs normally developed by merchants to normal debit or credit cards. The main objective is to cut costs on development of such programs and make a conjoint effort to offer discounts and special offers right in the moment of the purchase payment. The development of such project was only possible with the adoption of data mining techniques. This concept is one step ahead of Portuguese reality, once it can be seen as a new market in payment’s processing system. In conclusion, the implementation of data mining techniques is an opportunity to improve segmentation of retail banking customers. It also represents an opportunity to gain competitive advantage. The first financial institution to adopt this software will be beneficiated by the first mover advantage and it will be recognize as innovative and efficient leading institution in the market. As so,it will reinforce its presence and brand in the financial market. 5.2.

Limitations and Solutions Proposal

As mention before, when a payment transaction is made, the only parameters identifying where the purchase has taken place are the identification of the merchant, which is the legal name of the company and the Merchant Category Code (MCC) associated to the POS terminal. On one hand, the legal name in most cases does not have a connection with the brand or establishment name. On the other hand, the usage of the Merchant Category Code could be revealed as classification method not 22

always efficient. Large retailers with diversified products offer could on main MCC and multiple sub classification codes. This fact could lead to confusion and distortions on databases analyse. To solve this problem, a new classification could be implemented by the acquirers in order to properly address the merchant sector of activity. Another inefficiency of the classification methods could be raised regarding errors and misaligned classification. The merchant category code is associated to a fiscal identification number. Without proper statistics about this issue, it is relevant to refer the situation when a businessman owns more than one establishment. In some occasions, in order to reduce costs, owners tend to request POS terminals linked to one business and use some of them in another store. When the two or more stores are not related to the same type of business, this can induce to mistakes in transaction classification. On the contrary, when the owner decides to abandon a previous business, starting to operate with the same name company and fiscal identity, it causes once more the same mismatch from classification and product or service provided. The solution to solve this problem is the increase of surveillance around the MCC attribution. Furthermore, despite of all the advantages identified in this dissertation, as well as some research identified in the literature review, it was never the purpose of this thesis to perform an analysis/study of the financial benefits of the acquisition of data mining software. By the delivery deadline restrictions in terms of time to develop this dissertation and the work it would involved, it would not be possible to contemplate that sort of analysis in this dissertation. Moreover, the developed of a business case with costs and potential benefits would consist in an interesting subject for further research.

6. CONCLUSIONS This dissertation tries to demonstrate that, given the benefits of behavioural segmentation, retail banking is the most prepared industry to implement it, since financial institutions have extensive databases about the consumption patterns of its customers. As so, the aim of this thesis was to analyse and clarify what is already implemented in Portugal, and propose suggestions for banks that want to adopt 23

behavioural segmentation strategies. From Financial Services Consumer Insight Survey 2012 accessed by Datamonitor, it was concluded that 50% of UK banking customers feel that their banks does not understand their needs. This conclusion can be easily transport to Portuguese retail banking industry, as financial institutions act launching new products and functionalities focussing on increasing their portfolio more than by inferring customer’s needs through their profiles and segments. In order to embrace the constant change in financial market, retail banking industry has to start looking for trends and advanced techniques available. The conservative retail banking industry in Portugal, must to adapt their vision, having this time the main focus on customers and their needs. That being said, Portuguese banks have to maintain their technology development and still be open to other realities and inputs. With a well design transaction processing system and a strong national scheme implemented, it will be a waste not to keep up with current trends and yield to pressures from foreign financial institutions. Data mining application in this context will possibly improve the way retail banking works and establish a closer relationship between customers, merchants and banks on both sides. But it is nothing more than a mean to accomplish in the end of customer satisfaction. Transactions and traditional customer profiles are just one dimensional peek into customers’ complex lives. If banks are seriously engaged with their customers, they must use a wider range of techniques on communication and marketing targeting and technologies available. Customers’ needs and expectations are more than evolving over the time, it is changing constantly and it will change faster in the years yet to come. Having a customer oriented service, from the front office to the senior manager, seems to be the key to succeed in this industry. Finally, it is considered that retail banks would benefit from establishing association rules and cluster segmentation based on actual transaction data, as it will lead to higher profitability and better allocation of resources. The technological solution proposed will also enhance the development of Innovative product solutions and prepare banks to integrate the next trends ahead.

24

6.1.

Further Research

Concerning the data mining implementation costs, and as mentioned before, it would be interesting to develop a business case for this matter. It should be explored not only the software costs but also the potential revenues generated by its implementation. In this sense, it will only be possible if the scope of the research would be aim to address defined target markets and variables. As financial markets are not static in terms of trends and product solutions, new ways to use the patterns identified by data mining should be subject of research in order to provide an updated state of the art on customer segmentation.

25

REFERENCES Bank of Portugal (2013), Terminais de Pagamento e Caixas Automáticas, Bank of Portugal Booklet nº 10 Berry, M. and Linoff, G. (2000), Mastering Data Mining, John Wiley & Sons, NY Burnett, J. and Chonko, L. (1984), “A Segmental Approach to “Packaging” Bank Products”, Journal of Retail Banking Carrier, C. G., & Povel, O. (2003), “Characterising data mining software”, Intelligent Data Analysis, pp. 181–192. Chung, H. and Gray, M. (1999), “Special Section: Data Mining”, Journal of Management Information Systems Datamonitor Financials (2014), Consumer Payments Tracker Datamonitor Financials (2014), Payment Card Loyalty: The Future of Loyalty Ennew, C., Wright, M. and Watkins, T. (2000), Marketing Financial Services, Butterworth-Heinemann, Oxford Elliott G. and Glynn, N. (1998), “Segmenting Financial Markets for Customer Relationships: A Portfolio-Based Approach”, Service Industries Journal, pp. 38-54 Fayyad, U., Piatetsky-Shapiro, G. and Smith,P. (1996), “From Data Mining to Knowledge Discovery in Databases”, Artificial Intelligence, vol. 17 Fonseca, J. and Cardoso, M. (2007), “Supermarket customers segments stability”, Journal of Targeting, measurement and Analysis for Marketing, pp. 210-221 Green, P. S. (1977), “A new approach to market segmentation”, Business Horizons, pp.61-66 Harrison, T.S. (1994), “Mapping customer segments for personal financial services”, International Journal of Bank Marketing, pp. 17-25 Hormozi, A. and Giles, S. (2004), “Data Mining: a competitive weapon for banking and retail industries, Information Systems Management, pp. 62 Hui, S. and Jha, G. (2000), “Data Mining for Customer Service Support”, Information & Management, pp.1-13 Kantardzic, M. (2011), Data Mining: Concepts, Models, Methods and Algorithms, John Wiley & Sons, Inc., Hoboken, New Jersey Kitching, D. (1982), “Rationalising Branch Banking”, Long Range Planning, pp. 53-62 Kotler, P. (1980), Principles in Marketing, Prentice-Hall, Englewood Cliffs, NJ Kotler, P. and Armstrong, G. (2009), Principals of Marketing, Prentice Hall Kotler, P. and Keller, K. (2006), Marketing Management, Prentice Hall

26

Lamb, C. and Carl McDaniel, C. (2003), Marketing, Peking University Press, Beijing, pp. 214 Ling, X. and Li, C., 1998. “Data Mining for Direct Marketing: Problems and Solutions”. In Proceedings of the 4th KDD conference, AAAI Press, 73–79 McDougall, G. and Levesque, T. (1994) "Benefit Segmentation Using Service Quality Dimensions: An Investigation in Retail Banking", International Journal of Bank Marketing, pp.15 – 23 Machauer, A. and Morgner, S. (2001), “Segmentation of bank customers by expected benefits and attitudes”, International Journal of Bank Marketing, pp. 6-17 MasterCard (2007), The Anatomy of a Transaction Meidan, A. (1984), Bank Marketing Management, Macmillan, New York, NY Rygielski, C., Wang, J. and Yen, D. (2002), “Datamining Techniques for Customer Relationship Management”, Technology in Society, pp. 483-502 Smith, W. R. (1956), “Product differentiation and market segmentation as alternative marketing strategies”, Journal of Bank Marketing, July, pp. 3-8 Speed, R. and Smith, G. (1992), “Retail Financial Services Segmentation”, The Service Industries Journal”, July, pp. 368-383 Sun, S. (2009), “An Analysis on the Conditions and Methods of Market Segmentation”, International Journal of Business and Management Tufféry, S. (2008), Data Mining and Statistics for Decision Making, John Wiley & Sons, Ltd. Turban, E., Aronson, J., Liang, T. and Sharda, R. (2007), Decision Support and Business Intelligence Systems, 8th edition, Pearson Education Wedel, M. and Kamakura, W. (2000), Market Segmentation: Conceptual and Methodological Foundations, 2nd edition, Kluwer Academic Publishers, Boston, pp. 3 Wind, Y. (1978), “Issues and advances in segmentation research”, Journal of Marketing Research, Vol. 15, pp. 317-37 Ziff, R. (1971), “Psychographics for Market Segmentation”, Journal of Advertising Research, Vol. II No. 2, April, pp. 3-9

27

APPENDIX Appendix 1- Classification Framework of Data Mining Techniques Applied to CRM

Source: Ngai et al., 2009

Appendix 2 - Bank of Portugal Four Corner Model Representation

28

Appendix 3 - MasterCard Anatomy of the Transaction Representation including Clearing and Settlement Process

29