Content uploaded by Andreas Kaplan
Author content
All content in this area was uploaded by Andreas Kaplan on Dec 09, 2017
Content may be subject to copyright.
A Model to Determine
Customer Lifetime
Value in a Retail
Banking Context
MICHAEL HAENLEIN, ESCP-EAP European School of Management, Paris
ANDREAS M. KAPLAN, ESSEC Business School, Paris
ANEMONE J. BEESER, McKinsey & Company Inc., Frankfurt am Main
During the past decade, the retail banking industry
started to face a set of radically new challenges that
had an overall negative impact on industry margin
and profitability. In response to these challenges,
more and more retail banks have focused on
increasing the scale of their operations, which has
led to a rising importance of mergers and acquisi-
tions (M&A). From a Marketing perspective,
M&A transactions are nothing other than the acqui-
sition of the customer base of one company by
another one, usually based on the assumption that
the acquiring bank can manage this customer base
more profitably than the selling bank was able to. It
is therefore not surprising that questions about the
valuation of customers have become more impor-
tant than ever in the retail banking industry.
Our article provides a contribution in this area by
presenting a customer valuation model that we
developed in cooperation with a leading German
retail bank, which takes account of the specific
requirements of this industry. Our model is based
on a combination of first-order Markov chain mod-
eling and CART (classification and regression tree)
and can deal equally well with discrete one-time
transactions as with continuous revenue streams.
Furthermore, it is based on the analysis of homoge-
neous groups instead of individual customers and
is easy to understand and parsimonious in nature.
In our article we provide proof of the practical
value of our approach by validating our model
using 6.2 million datasets. This validation shows
how our model can be applied in day-to-day busi-
ness life.
Ó2007 Elsevier Ltd. All rights reserved.
Keywords: Customer Relationship Management,
Customer Lifetime Value, Retail Banking, Markov
Chains, CART Analysis
Introduction
During the past decade, the retail banking industry
started to face a set of radically new challenges that
had an overall negative impact on industry margins
and profitability. For the major part, these challenges
have been caused by advances in modern informa-
tion and telecommunication technologies, which ulti-
mately have resulted in higher cost transparency and
brand switching behavior. The resulting increase in
competitive intensity has led to a commoditization
of basic banking products, such as deposit taking,
mortgages and credit extensions. This has further
been fueled by an ever rising number of new entrants
in the retail banking sector coming from industries as
diverse as insurance and automobile production. The
acquisition of the German Allbank by General Elec-
tric to create the GE Money Bank in 2004 as well as
the increasing number of cars sold in a credit vs. cash
mode are omnipresent witnesses of this evolution.
In response to these challenges, more and more retail
banks have focused on increasing the scale of their
operations, which has led to a rising importance of
mergers and acquisitions (M&A). Particularly in the
European retail banking environment, which has his-
torically always been more fragmented than, for
European Management Journal Vol. 25, No. 3, pp. 221–234, June 2007 221
doi:10.1016/j.emj.2007.01.004
European Management Journal Vol. 25, No. 3, pp. 221–234, 2007
Ó2007 Elsevier Ltd. All rights reserved.
0263-2373 $32.00
example, its North American counterpart, this trend
is obvious. Prominent examples include the 1999
merger of BNP and Paribas in France, the acquisition
of the German HypoVereinsbank by the Italian Uni-
Credit Group in 2005 and the merger between Banca
Intesa and Sanpaolo in 2006. There are at least four
reasons that give room for the assumption that this
trend towards increasing M&A in the European
retail banking industry is likely to continue in the
future. First, in many European countries retail bank-
ing is still very fragmented. For example in Italy, the
five largest banks own less than half of the banking
market. Second, the introduction of the Euro has
led to further margin pressure due to an increase in
cross-border competition among retail banks. This
has resulted, among others, in a decreasing cost of
borrowing and the loss of certain revenue streams
such as commission fees on currency exchanges.
Third, the Basel II framework, which sets new stan-
dards in risk management and capital adequacy, is
associated with various changes in operations, the
implementation of which create a significant cost
burden that causes additional worry in this already
tortured industry. Finally, mergers and acquisitions
can be an appropriate strategy for banks to hedge
macroeconomic risks, especially in Europe where
loan portfolios are often severely home-biased.
From a Marketing perspective, M&A transactions are
nothing other than the acquisition of the customer
base of one company by another one, usually based
on the assumption that the acquiring bank can man-
age this customer base more profitably than the sell-
ing bank was able to (Selden and Colvin, 2003). It is,
therefore, not surprising that questions around the
valuation of customers have become more important
than ever in the retail banking industry. At the center
of this interest lies the concept of customer lifetime
value (CLV), which was defined more than 30 years
ago by Kotler as ‘‘the present value of the future
profit stream expected over a given time horizon of
transacting with the customer’’ (Kotler, 1974, p. 24).
The interest the Marketing discipline has recently
been paying to CLV and the related subject of cus-
tomer relationship management (CRM, e.g. Payne
and Frow, 2005) has its roots in an evolution that
started in the mid 1980 s. During that time, Dwyer
et al. (1987) were among the first to highlight that
Marketing, which has historically focused on the
analysis of single transactions, should start paying
attention to the relationship aspect of buyer–seller
behavior. Only three years later, Reichheld and
Sasser (1990) were able to show empirically that such
a relationship-focus can lead to significant advanta-
ges since customers tend to generate higher profits
the longer they stay with the company. Although
Reinartz and Kumar showed that the relationship
between lifetime and profitability may be more com-
plex than Reichheld and Sasser assumed, especially
in non-contractual relationships (Reinartz and
Kumar, 2002; Reinartz and Kumar, 2000), it has nour-
ished the idea that market-based assets, such as cus-
tomer relationships, can lead to superior market
performance and shareholder value (Srivastava
et al., 1998; Srivastava et al., 1999).
Central to the idea of CRM is the assumption that
customers differ in their needs and the value they
generate for the firm, and that the way customers
are managed should reflect these differences. CRM
is therefore not about offering every single customer
the best possible service, but about treating custom-
ers differently depending on their CLV. Such appro-
priate treatment can have many faces, starting with
offering loyalty programs to retain the most profit-
able customers (Shugan, 2005) through to the aban-
donment of unprofitable customer relationships
(Haenlein et al., 2006). Yet, selecting between these
strategies requires that the company knows the
value its different customers generate. Consequently
many papers have been published dealing with cus-
tomer valuation as well as conceptual and practical
challenges associated with it (e.g. Berger and Nasr,
1998; Jain and Singh, 2002; Rust et al., 2004). How-
ever, only a few of them (e.g. Berger et al., 2003;
Keane and Wang, 1995) are tailored to specific
industries and, hence, take account of sector-specific
challenges associated with the implementation of
those approaches. This is rather surprising as it
has long been highlighted that customers may differ
substantially across industries and that such differ-
ences should translate to the models used to value
them. For example Jackson (1985) stressed that cus-
tomers can be grouped into different categories,
depending on the level of commitment they show
to a particular seller. On one end of the spectrum
is the ‘‘always-a-share’’ model, which assumes that
customers can easily switch part or all of their
spending from one vendor to another. The opposite
end of the behavior spectrum assumes that, due to
high switching costs, the buyer is committed to only
one vendor to satisfy his or her needs. Once the
customer stops purchasing from this vendor and
changes to another one s/he is ‘‘lost-for-good’’
and cannot return to the vendor easily. Although
the category every customer can be allocated to
depends to a certain extent on this specific cus-
tomer’s preferences, it is also heavily influenced
by the type of product sold and, hence, the indus-
try. Building on this categorization, Dwyer (1989)
showed that models used to value customers in
these two settings differ substantially and proposed
two approaches to determining CLV: a customer
migration model and a customer retention model.
It therefore makes intuitive sense that models to
determine CLV should, at least to a certain extent,
be adapted to specific industry characteristics.
Thinking about the retail banking environment, a
model to determine CLV should satisfy at least three
conditions: First, it needs to be able to handle discrete
one-off transactions, which occur either only once in
a lifetime or in very long purchasing cycles (e.g. mort-
gages), and continuous revenue streams (e.g. regular
A MODEL TO DETERMINE CUSTOMER LIFETIME VALUE IN A RETAIL BANKING CONTEXT
222 European Management Journal Vol. 25, No. 3, pp. 221–234, June 2007
account maintenance charges) equally well. This is
due to the fact that retail banks generate revenue in
two main ways: by gaining a margin on lending
and investment activities and by receiving transac-
tion fees for transactions, credit cards, etc. (Garland,
2002). Second, in order to be easily implementable,
it should focus on the valuation of homogeneous
segments of customers instead of individual clients
(Libai et al., 2002). This requires a trade-off between
reflecting individual client characteristics, such as
product usage or lifetime phase, while at the same
time taking account of the sheer size of an average
retail bank’s customer base, where individual valua-
tion would result in disproportionate effort and
unmanageable complexity. Finally, it needs to be easy
to understand and parsimonious in nature to ensure
its applicability in many business contexts. This
specifically implies limiting data requirements to
the information available in an average bank’s infor-
mation system.
In this article we present a customer valuation
model, which we developed in cooperation with a
leading German retail bank and which takes account
of these specific requirements. This model is based
on a combination of first-order Markov chain model-
ing and CART (classification and regression tree)
analysis and has been validated using a sample of
roughly 6.2 million datasets. In the next section we
will discuss in more detail which type of data is
needed as an input for our model. We will then
develop our model conceptually and subsequently
validate it to show how it has been implemented at
our cooperating retail bank. The article finishes with
a discussion of the limitations of our approach and
areas of future research.
Data Requirements
Our model is based on four different groups of prof-
itability drivers: age, demographics/ lifestyle data,
product ownership (type and intensity) and activity
level. These profitability drivers have been defined
in collaboration with the management of the collabo-
rating retail bank to ensure that all of them fulfill the
aforementioned condition of being easily operation-
alizable using data available in the bank’s informa-
tion system.
Age
The work of Garland gives an indication that cus-
tomer contribution (defined as relationship revenue
minus relationship cost) is significantly influenced
by the customer’s age. Garland (2002, 2004) analyzed
1,100 personal retail customers of a New Zealand
regional bank. Using a stepwise regression analysis,
he showed that out of 26 non-financial profitability
drivers representing perceived service quality, cus-
tomer satisfaction, customer loyalty and customer
demographics, only four had significant explanatory
power for customer contribution. These were age,
share-of-wallet, household income and joint
accounts, with age being the most important one.
Also Campbell and Frei (2004) stress that in a retail
banking context age can be assumed to influence
profitability by its impact on consumption patterns.
For example middle-aged customers tend to be more
profitable than younger ones because they tend to
maintain higher balances and are more likely to have
mortgages.
Demographics/Lifestyle
Before and immediately after acquisition, a company
has only very limited data at its disposal which can
be used to determine customer value. Hence, it is a
common approach to use census and overlay data
on demographics and lifestyle as a proxy for other
unknown customer characteristics in early relation-
ship stages. As noted by Campbell and Frei (2004)
a typical retail bank spends between $1 million and
$2 million annually to procure demographic data
from outside vendors. This highlights the strong
importance this type of data has in the day-to-day
reality of many retail banks. Although demographic
variables often tend to be only weak predictors of
future behavior, we opted for their inclusion as prof-
itability drivers due to their high relevance in busi-
ness life.
Type and Intensity of Product Ownership
Several studies in the area of Direct Marketing and
Customer Relationship Management have shown a
strong relationship between future and past pur-
chase behavior. For example, Venkatesan and Kumar
(2004) provided an indication that past interpurchase
time (i.e. the time period between two consecutive
purchases) is a good predictor of future interpur-
chase time. This implies that the number of transac-
tions in the previous period (which is, essentially,
nothing other than the reverse of the interpurchase
time) can serve as a predictor of the transaction vol-
ume in the current period. Similarly, Fader et al.
(2005b) predicted future customer value based on
information on past customer recency, frequency
and monetary value with high accuracy. Among oth-
ers, such findings can be explained by the existence
of consumer inertia and switching cost which often
result in customers sticking with a certain choice
although changing would be preferable from a util-
ity-maximization perspective. Many models used in
Marketing literature, such as the NBD-Pareto (Sch-
mittlein et al., 1987) and BG-Pareto model (Fader
et al., 2005a) rely on this assumption and have proven
to be very successful in modeling future transaction
A MODEL TO DETERMINE CUSTOMER LIFETIME VALUE IN A RETAIL BANKING CONTEXT
European Management Journal Vol. 25, No. 3, pp. 221–234, June 2007 223
behavior. Since past and current purchase behavior
are reflected by (current) type and intensity of prod-
uct ownership, we decided on the inclusion of these
variables as potential profitability drivers.
Activity Level
In every type of database analysis it is crucial to dif-
ferentiate between contractual and non-contractual
settings (Schmittlein et al., 1987). In a contractual set-
ting, such as magazine subscriptions or health club
memberships, the company can easily observe
whether a customer is still active, i.e. whether the
customer is still doing business with the company,
or not. In a non-contractual setting, such as catalog
retailing or retail banking, where there may not be
a steady revenue stream to be expected from the cus-
tomer, the question of distinguishing between active
and inactive clients is far from trivial. In the retail
banking industry, for example, a client may no
longer be active, but still own an account. One poten-
tial reason for this could be that in the absence of reg-
ular account maintenance charges, this ownership is
not associated with any costs. Hence, there is only
limited motivation for the customer to formally end
the business relationship with his/her bank. How-
ever, inactive customers carry a higher risk of being
unprofitable, because they no longer generate any
revenue, while the client relationship may still lead
to costs, for example due to direct marketing cam-
paigns or regular mailing of account statements. It
can therefore be assumed that customer value and
profitability are heavily influenced by the activity
level of the customer, speaking for the inclusion of
this variable in our model.
With regard to the operationalization of these four
potential profitability drivers, we measured the first
one (age) by one indicator, the second one (demo-
graphics/lifestyle) by four indicators (marital status,
sex, income, region type) and the third one (type and
intensity of product ownership) using 11 and 15
items respectively (see details in Table 1). Concern-
ing the last profitability driver (activity level), we
defined two conditions under which the client was
to be considered as active (vs. inactive), based on dis-
cussions with the management of the collaborating
retail bank: First, all clients owning either a savings
product, a home financing product, a loan or an
insurance product were defined as being active due
to the regular revenue streams (savings, interest pay-
ments, insurance fees) resulting from any of these
products. Second, all clients owning transaction
accounts, custody accounts and savings deposits
were defined as being active when these accounts
either showed a positive balance of at least 100 Euros
and/or at least one transaction had been carried out
during the last three months (for transaction
accounts) or twelve months (for custody accounts
and savings deposits).
Model Development
Our modeling approach consists of three steps: First,
we use the aforementioned profitability drivers as
predictor variables together with the target variable
‘‘profit contribution’’ in a CART analysis to build a
regression tree. This tree helps us to cluster the cus-
tomer base into a set of homogeneous sub-groups.
Second, using these sub-groups as discrete states,
we subsequently estimate a transition matrix which
describes movements between them. In a final step
this transition matrix is then used to calculate CLV
for each of the homogeneous sub-groups using back-
ward induction. We will now describe each of these
three steps in more detail.
In the first step, the four potential profitability driv-
ers and their associated items were used as predictor
variables in a CART analysis with contribution mar-
gin as the target variable. CART analysis (classifica-
tion and regression trees), first introduced by
Breiman et al. (1984), is a technique for determining
membership of a set of classes as a function of certain
predictor variables. In this sense, it is comparable to
traditional discriminant or regression analysis (see
Armstrong and Andress, 1970; Armstrong, 1971;
Crocker, 1971 for a discussion of similarities and dif-
ferences). CART analysis assumes that the researcher
identifies a target (response) variable, which is used
to define the set of classes, and one or more predictor
variables, both of which can be either discrete or con-
tinuous. For our analysis, we relied on ‘contribution
margin’ as the target variable, which we defined as
revenue resulting from interest payments and com-
mission fees less liquidity cost, equity cost, risk cost
and transaction cost covering the bank’s cost of hold-
ing cash, maintaining a certain credit risk-dependent
equity ratio, accepting the risk of credit loss and car-
rying out customer-related transactions respectively.
With respect to predictor variables, using variables
with different scales (nominal for type of product
ownership vs. continuous for intensity of product
ownership and contribution margin) was possible
without running the risk of inconsistent results, due
to the soft statistical assumptions underlying CART
analysis. However, we faced the potential problem
that client age can be assumed to heavily influence
the other potential profitability drivers. With regard
to demographics/lifestyle, marital status and income
can be expected to be life cycle- and, therefore, age-
dependent. Also type and intensity of product
ownership are likely to depend on age, as discussed
previously. Although correlations (even high ones)
between different predictor variables are not a prob-
lem in CART analysis, we decided to carry out sepa-
rate CART analyses for different age groups to
appropriately take account of these effects.
In the second step, each of the resulting homoge-
neous groups was considered as one state of nature,
between which we allowed the customers to flow
A MODEL TO DETERMINE CUSTOMER LIFETIME VALUE IN A RETAIL BANKING CONTEXT
224 European Management Journal Vol. 25, No. 3, pp. 221–234, June 2007
following a first-order Markov process. A first-order
Markov process is a stochastic process in which the
transition probability between two discrete states of
nature depends only on the properties of the imme-
diate preceding state, independent of the path by
which this state was reached – a condition also
referred to as Markov property. The combination of
different Markov processes in a row is called a Mar-
kov chain and the respective transition probabilities
are usually summarized in a transition matrix. Mar-
kov chains have a long history in marketing in gen-
eral (Styan and Smith, 1964; Thompson and
McNeal, 1967), as well as customer lifetime valuation
in particular (Morrison et al., 1982; Pfeifer and
Carraway, 2000; Rust et al., 2004). Due to the fact that
the different groups resulting from the CART analy-
sis were defined as being age-dependent (although,
as will be seen later, there is no one-to-one relation-
ship between states of nature and age groups), the
state each customer belongs to can change due to
increasing age, even if all other factors remain con-
stant. To estimate the corresponding transition prob-
abilities, we determined the state of nature each
customer belonged to at the beginning and end of a
predefined time interval Tby using the decision
rules resulting from the aforementioned CART anal-
ysis. By counting the number of customers who
moved between two states and dividing this number
by the total number of customers, we were able to
estimate transition frequencies that served as proxies
for the underlying transition probabilities.
In the final step, we determined the CLV for each
customer group as the discounted sum of state-
dependent contribution margins, weighted with their
corresponding transition probabilities. This calcula-
tion was carried out using backward induction. We
started with calculating the value of a group of cus-
tomers in the second-to-last age group G1 and
state of nature k. This value can be determined by
multiplying the contribution margin to be expected
from a client in age group Gand state of nature jwith
the probability that a customer will transit from state
kin G1 into state jin G. Summing these products
over all potential states of nature for age group
G(j= 1,2, ... n) and discounting them back one per-
iod using a pre-defined discount rate, results in an
estimate for the value of a group of customers in
the second-to-last age group G1 and state of nat-
ure k. Based on the same logic we then determined
customer values for all periods G1, G2, ... ,2
and associated states of nature, which finally helped
us to calculate the CLV of a client in state of nature k
at the beginning of the analysis period.
1
Model Validation
To carry out validation, we took two random samples
from the customer base of our cooperating retail bank,
the first one consisting of 687,000 and the second one
of 5.5 million datasets. To ensure confidentiality of
Table 1 Operationalization of Potential Profitability Drivers: Type and Intensity of Product Ownership
Type of product ownership
TP-01 Client owns transaction account (0 = no, 1 = yes)
TP-02 Client owns custody account (0 = no, 1 = yes)
TP-03 Client owns savings deposits (0 = no, 1 = yes)
TP-04 Client owns savings plan (0 = no, 1 = yes)
TP-05 Client owns home savings agreement (0 = no, 1 = yes)
TP-06 Client owns own home financing (0 = no, 1 = yes)
TP-07 Client owns complex home financing (0 = no, 1 = yes)
TP-08 Client owns personal loan (0 = no, 1 = yes)
TP-09 Client owns arranged credit product (0 = no, 1 = yes)
TP-10 Client owns life insurance (0 = no, 1 = yes)
TP-11 Client owns other insurance products (0 = no, 1 = yes)
Intensity of product ownership
IP-01 Positive balance transaction account (in EURO)
IP-02 Negative balance transaction account (in EURO)
IP-03 Positive balance custody account (in EURO)
IP-04 Negative balance custody account (in EURO)
IP-05 Market value custody account (in EURO)
IP-06 Turnover custody account during last 12 months (in EURO)
IP-07 Balance savings deposits (in EURO)
IP-08 Balance of savings plans (in EURO)
IP-09 Balance home savings agreement (in EURO)
IP-10 Balance own home financing (in EURO)
IP-11 Balance complex home financing (in EURO)
IP-12 Balance personal loan (in EURO)
IP-13 # cash payments and withdrawals during last 3 months (number of transactions)
IP-14 # form-based fund transfers during last 3 months (number of transactions)
IP-15 # transactions custody account during last 12 months (number of transactions)
A MODEL TO DETERMINE CUSTOMER LIFETIME VALUE IN A RETAIL BANKING CONTEXT
European Management Journal Vol. 25, No. 3, pp. 221–234, June 2007 225
this data we multiplied all monetary values (such as,
for example, contribution margin) in the following
output with an arbitrary factor, which we chose small
enough not to distort key patterns in the data. They
are, therefore, no longer expressed in Euros, but in
currency units (CU).
In order to reflect the latent nature of the predictor
variables’ type and intensity of product ownership
without loosing the detailed information provided
by their respective items, we calculated additional
predictor variables for the CART analysis by sum-
ming up all or sub-sets of their formative indicators.
This approach is in line with the basic philosophy
behind this type of measurement (see Jarvis et al.,
2003 for more details). Regarding type of product
ownership, we defined the variable ‘‘usage intensity’’
as measuring the number of products owned and cal-
culated it by summing up all items TP-01 to TP-11.
With regard to intensity of product ownership, we
calculated three additional variables: First, ‘‘total sav-
ings’’ were defined as the sum of positive balances of
transaction (IP-01) and custody accounts (IP-03) plus
the balances of savings deposits (IP-07) and savings
plans (IP-08). Second, ‘‘total liabilities’’ were deter-
mined as the sum of negative balances of transaction
(IP-02) and custody account (IP-04) plus the balances
from home savings agreements (IP-09), as well as
own (IP-10) and complex home financing (IP-11)
and personal loans (IP-12). Finally, ‘‘client’s net
wealth’’ (CNW) was calculated as the sum of total
savings and total liabilities.
We then defined the indicators of the variables
demographics/lifestyle, type and intensity of prod-
uct ownership, as well as the activity level and the
four newly created composite variables (i.e. usage
intensity, total savings, total liabilities, client’s net
wealth) as predictor variables. Together with contri-
bution margin as target variable, we carried out a
CART analysis based on the first sample consisting
of 687,000 datasets. Hereby, we split the sample into
a training sample (172,000 datasets, 25%) and a test-
ing sample (515,000 datasets, 75%). This analysis
was carried out using the ALICE software tool, Ver-
sion 6.5. To reduce the complexity of our analysis, we
started with the hypothesis of profitability drivers
being equal across different age groups and used
the training sample to estimate a regression tree for
all age groups simultaneously. In a second step we
split the testing sample into six different sub-samples
according to client’s age (<19 years, 19–29 years, 30–
39 years, 40–49 years, 50–65 years, >65 years) and
tested whether this (pooled) tree was able to classify
datasets in each of these sub-samples correctly.
While this was the case for five sub-samples (signif-
icance of all nodes above 0.9981), the tree was not
able to classify the <19 years age-group correctly
(three out of eleven p-values insignificant). We there-
fore carried out a second CART analysis using the
training sample to estimate a separate regression tree
for clients <19 years. These two regression trees (one
for clients <19 years and one for clients of age 19 and
above) then helped to split customers in the different
age-bands into 10 and 22 different homogeneous
sub-groups respectively. Finally, we defined one
additional segment to cover clients who terminated
the relationship with the bank, whether due to brand
switching or death.
Figure 1 shows the resulting regression tree for cli-
ents of age 19 and above.
2
As can be seen, contribu-
tion margin is primarily influenced by client net
wealth with higher net wealth leading to higher aver-
age contribution margins. For clients with negative
net wealth (debtors), profitability is subsequently
dependent on the size of personal loans while for
high net wealth clients (CNW > 7,500) it is a function
of custody account turnover (IP-06). For customers
with medium net wealth (350 < CNW 67,500),
which represent roughly 45% of the total client base
(see Table 2), the main driver of higher contribution
margins is type and intensity of product ownership
(personal loans, savings products, custody accounts),
as well as their activity status and usage intensity.
Finally, clients with low net wealth (0 < CNW6
350), accounting for about 16% of the client base,
are primarily characterized by negative contribution
margins due to low usage intensity.
Subsequently, we used the second sample consisting
of 5.5 million datasets to estimate the transition prob-
abilities between these different states by assuming
an arbitrarily chosen time interval Tof 2 years. We
first split the client base into 39 different age-bands
covering two years each, from clients 1–2 years old
up to clients aged 77–78 years. Using the decision
rules resulting rules from the aforementioned CART
analysis (see Figure 1 and Table 2), we then deter-
mined the segment each client belonged to at the
beginning and end of this 2-year timeframe. The
transition probabilities were subsequently approxi-
mated by the transition frequencies calculated as
the number of clients either staying in one segment
or moving between two divided by the number of
all clients in the respective segment.
3
Table 3 shows one exemplary transition matrix for
the 19th age-band (clients aged 37/38 years). As
can be seen in Table 2,Segment 1 primarily consists
of customers holding small- to medium-sized per-
sonal loans. Since these loans are usually not paid
back within one period, the majority of customers
stay within this segment in the next period (41%).
However, some customers also increase the size of
their loans and move to Segment 2 (12%), while oth-
ers pay their loans back and decide to terminate their
relationship with the bank (12%). Users of large per-
sonal loans and mortgages are within Segment 2 (total
liabilities > 5,250 CU). Due to the long life of such
loans, 66% of these customers stay within Segment
2 and only a small fraction (2%) pays them back
within the same period and leaves the bank. Segment
3is also composed of debtors. However, unlike
A MODEL TO DETERMINE CUSTOMER LIFETIME VALUE IN A RETAIL BANKING CONTEXT
226 European Management Journal Vol. 25, No. 3, pp. 221–234, June 2007
Segments 1 and 2, these customers are not subject to
contractual credit periods but consume bank over-
drafts. Due to the high interest rate associated with
these products, the majority decide to switch to a
(cheaper) personal loan in the next period and move
to Segment 2 (80%). Segment 4 consists of inactive
customer relationships with small net wealth, the
majority of which (87%) are not activated within
the next period. In most cases, these clients are not
interested in actively managing their account. In
cases where they decide to do so, 7% terminate their
relationship with the bank. Although active, clients
in Segment 5 are also characterized by low net wealth.
Except for the rare cases in which these clients decide
either to increase their net wealth (8% move to Seg-
ment 11) or to raise a personal loan (5% move to Seg-
ment 1), the majority remain within the same
segment (46%), become inactive (22% move to Seg-
ment 4) or terminate the client relationship (10%).
Clients in Segment 6 have a higher usage intensity
than those in Segment 5. Consequently, the share that
decides to increase its net wealth (14% move to Seg-
ment 12) or to raise a personal loan (23% move to
Segment 1 or 2) is substantially higher and only 3%
close their account.
Clients who at the same time hold savings with the
bank and consume a personal loan, are within Seg-
ment 7. While 21% of these clients remain within
the same segment during the next period, an equally
large group pays their loans back (20% move to Seg-
ment 12 or 14) or increases total liabilities (17% move
to Segment 1). Customers in Segment 8 have small to
medium net wealth (67,500 CU) and small custody
account turnover that is often below break-even
point. Nearly half of these clients (48%) stay within
the same segment, while the others either reduce
their net wealth even further (20% move to Segments
4, 5 and 6) or close their account (9%). Clients in Seg-
ment 9 also have small to medium net wealth and
limited custody account turnover (yet above that of
Segment 8). During the next period 10% reduce their
net wealth and move to Segments 5 or 6. Roughly
one third remain in the same segment (34%) and
another third increase net wealth and custody
account turnover (30% move to Segment 17). Clients
in Segment 10 are comparable to the ones in Segments
8 and 9 with respect to net wealth, but do not per-
form any share trading. In the next period 24% do
not change this, 10% start using a custody account,
but with limited turnover (Segment 8), while another
Segment 4
Usage intensity = 0
Mean: - 19
Segment 5
Usage intensity = 1
Mean: - 12
Segment 6
Usage intensity > 1
Mean: 51
CNW < 0
0 < CNW 350
Mean: 96
Segment 1
Total liabilities = 5,250
Mean: 105
Segment 2
Total liabilities > 5,250
Mean: 451
TP-08 = 1
350 < CNW = 7,500
7,500 < CNW
CNW > 22,500
Segment 8
0 < IP-06 = 1,500
Mean: -95
Segment 9
IP-06 > 1,500
Mean: 66
Segment 12
Usage intensity > 1
Mean: 38
Segment 10
IP-06 = 0
Mean: 2
TP-02 = 1
Segment 11
Usage intensity = 1
Mean: 9
0 < Total savings 2,000
TP-02 0
Total savings > 2,000
TP-08 = 0
Segment 15
Total savings = 0
Mean: - 14
Segment 22
CNW = 0
Mean: - 16
Segment 3
TP-08 = 0
Mean: 28
Segment 7
TP-08 = 1
Mean: 109
Segment 13
Inactive client
Mean: 31
Segment 14
Active client
Mean: 96
Segment 16
0 < IP-06 = 3,000
Mean: 54
Segment 17
IP-06 > 3,000
Mean: 271
Segment 18
IP-06 = 0
Mean: 144
Segment 19
0 < IP-06 = 4,500
Mean: 343
Segment 20
IP-06 > 4,500
Mean: 890
Segment 21
IP-06 = 0
Mean: 390
Segment 4
Usage intensity = 0
Mean: - 19
Segment 5
Usage intensity = 1
Mean: - 12
Segment 6
Usage intensity > 1
Mean: 51
CNW < 0
0 < CNW 350
Mean: 96
Segment 1
Total liabilities 5,250
Mean: 105
Segment 2
Total liabilities > 5,250
Mean: 451
TP-08 = 1
350 < CNW = 7,500
7,500 < CNW 22,500
CNW > 22,500
Segment 8
0 < IP-06 1,500
Mean: -95
Segment 9
IP-06 > 1,500
Mean: 66
Segment 12
Usage intensity > 1
Mean: 38
Segment 10
IP-06 = 0
Mean: 2
TP-02 = 1
Segment 11
Usage intensity = 1
Mean: 9
0 < Total savings
TP-02 0
Total savings > 2,000
TP-08 = 0
Segment 15
Total savings = 0
Mean: - 14
Segment 22
CNW = 0
Mean: - 16
Segment 3
TP-08 = 0
Mean: 28
Segment 7
TP-08 = 1
Mean: 109
Segment 13
Inactive client
Mean: 31
Segment 14
Active client
Mean: 96
Segment 16
0 < IP-06 3,000
Mean: 54
Segment 17
IP-06 > 3,000
Mean: 271
Segment 18
IP-06 = 0
Mean: 144
Segment 19
0 < IP-06 4,500
Mean: 343
Segment 20
IP-06 > 4,500
Mean: 890
Segment 21
IP-06 = 0
Mean: 390
Figure 1 Regression Tree for Clients of Age 19 and Above (Mean = Average Contribution Margin of Segment in CU)
A MODEL TO DETERMINE CUSTOMER LIFETIME VALUE IN A RETAIL BANKING CONTEXT
European Management Journal Vol. 25, No. 3, pp. 221–234, June 2007 227
Table 2 Overview of Segmentation Criteria and Segment Description
Segment
#
Average Contribution
Margin (Segment size)
Segmentation
Criteria
Description
20 890 (2.1%) CNW > 22,500 Clients with high net wealth and high custody account turnover
IP-06 > 4,500
2 451 (4.7%) CNW < 0 Clients with mortgages and large personal loans
Liabilities > 5,250
TP-08 = 1
21 390 (2.2%) CNW > 22,500 Clients with high net wealth, focused on traditional investment
strategiesIP-06 = 0
19 343 (2.2%) CNW > 22,500 Clients with high net wealth and low to medium custody account
turnover0 < IP-06 64,500
17 271 (4.2%) CNW > 7,500 Clients with medium net wealth and medium to high custody
account turnoverCNW 622,500
IP-06 > 3,000
18 144 (4.8%) CNW > 7,500 Clients with medium net wealth, focused on traditional investment
strategiesCNW 622,500
IP-06 = 0
7 109 (2.3%) CNW > 350 Clients simultaneously consuming investment and financing
productsCNW 67,500
TP-08 = 1
1 105 (4.1%) CNW < 0 Clients owning small to medium personal loans
Liabilities 65,250
TP-08 = 1
14 96 (10.1%) CNW > 350 Clients with small to medium net wealth, focused on payment
transactions and savings productsCNW 67,500
Savings > 2,000
TP-08 = 0
Active client
9 66 (1.7%) CNW > 350 Clients with small to medium net wealth, focused on using custody
accounts; only limited use of other servicesCNW 67,500
0 < Savings 62,000
TP-08 = 0
TP-02 = 1
IP-06 > 1,500
16 54 (5.3%) CNW > 7,500 Clients with medium net wealth and low custody account turnover
CNW 622,500
0 < IP-06 63,000
6 51 (0.9%) CNW > 0 (Often relatively young) clients with very low net wealth, focused
on payment transactions and short-term savings productsCNW 6350
Usage intensity > 1
12 38 (4.7%) CNW > 350 Clients with low net wealth, focused on payment transactions
and a limited amount of savings productsCNW 67,500
0 < Savings 62,000
TP-08 = 0
TP-02 = 0
Usage intensity > 1
13 31 (5.5%) CNW > 350 Clients with low to medium net wealth, focused on investment;
payment transactions conducted using alternative bank accountsCNW 67,500
Savings > 2,000
TP-08 = 0
Inactive client
3 28 (4.0%) CNW < 0 Clients with overdrafts
TP-08 = 0
11 9 (12.4%) CNW > 350 Clients with low net wealth, focused on savings products
CNW 67,500
0 < Savings 62,000
TP-08 = 0
TP-02 = 0
Usage intensity = 1
A MODEL TO DETERMINE CUSTOMER LIFETIME VALUE IN A RETAIL BANKING CONTEXT
228 European Management Journal Vol. 25, No. 3, pp. 221–234, June 2007
10% increase their net wealth, but decide to invest in
traditional savings products (move to Segment 18).
Segment 11 consists of clients with low usage inten-
sity and strong focus on savings products. These
Table 2 (continued)
Segment # Average Contribution
Margin (Segment size)
Segmentation
Criteria
Description
10 2 (1.7%) CNW > 350 Clients with low net wealth and (probably inactive) custody
accounts without turnoverCNW 67,500
0 < Savings 62,000
TP-08 = 0
IP-06 = 0
TP-02 = 1
512 (7.1%) CNW > 0 Clients with very low net wealth (i.e. often clients without assets
or using the bank as secondary correspondent bank)CNW 6350
Usage intensity = 1
15 14 (4.7%) CNW > 350 Clients focused on stock trading, small custody account turnover
and small to medium custody account volumeCNW 67,500
Savings = 0
TP-08 = 0
22 16 (5.6%) CNW = 0 Clients without assets and liabilities, often using only specific
services (e.g. credit cards, safe-deposit boxes)
419 (7.7%) CNW > 0 Inactive accounts with ‘dead’ savings books or unused
custody accountsCNW 6350
Usage intensity = 0
895 (1.7%) CNW > 350 Clients focused on stock trading with small turnover
CNW 67,500
0 < Savings 62,000
0 < IP-06 61,500
TP-08 = 0
TP-02 = 1
0 0 Terminated client relationships (due to brand switching or death)
Table 3 Transition Matrix for 19th Age-Band: Clients Aged 37/38 Years
State of Nature iState of Nature j(Transition Probability in %)
123456789101112131415161718192021220
1411201523000440300010001112
21766101130001201000100042
3180 1000000000000000000017
400087 2000001000000000037
550022 46 1000081111000000310
61941412135100514 07200200063
7177112421 20129111 111800153
8110710 3 148 211132120200059
90000643034 004320030 036024
10 521124710 024 2511700010 00153
11 61051110100432360002000513
12 10211455000829119100500153
13 00066101010713410118001318
14 6201324201891330111401265
15 0171300200000058004000519
16 120011410600114037 14 011 3021
17 011012205001030042 0039 012
18 2112312801323810042 00758
19 121001112001010
16 904419 011
20 0200020010000100110079 012
21 1211101201211300023004658
22 1120100000000080010005333
A MODEL TO DETERMINE CUSTOMER LIFETIME VALUE IN A RETAIL BANKING CONTEXT
European Management Journal Vol. 25, No. 3, pp. 221–234, June 2007 229
are usually clients that use the bank as a secondary
correspondent bank for short- and medium term
financial investment, leading to a relatively large
churn probability for this segment (13%). Clients in
Segments 12,13,14 and 15 are similar to the ones in
Segment 11 with respect to product usage. Segments
12 and 14 do, however, have a higher usage intensity,
leading to significantly lower churn probabilities (3%
and 5% respectively). Clients in Segment 16 have
medium net wealth (between 7,500 and 22,500 CU),
but relatively small custody account turnover (below
3,000 CU). While the majority of these customers
stays within the same segment (37%), 14% signifi-
cantly increase their custody account turnover and
move to Segment 17 while another 11% also increase
their net wealth and move to Segment 19. Clients in
Segment 17 have larger custody account turnover
than the ones in Segment 16. The main share of these
clients can be maintained in this attractive state dur-
ing the next period (42%) and some (39%) even
increase their net wealth and profitability from the
bank’s perspective.
In contrast to clients in Segments 16 and 17, custom-
ers in Segment 18 do not conduct any share trading
and are unlikely to change this in the immediate
future (42% stay within Segment 18). Of these,
16% decide to reduce their net wealth (moving to
Segments 8 and 14) or to close their account (8%).
Clients in Segment 19 have large net wealth (above
22,500 CU) and low to medium custody account
turnover (below 4,500 CU). In the next period 19%
increase their custody account turnover and move
to Segment 20, while an equally large share (16%)
reduces turnover, transferring to Segment 16. The
bank’s most profitable customers are within Segment
20. These customers do not only have large net
wealth, but also significant custody account turn-
over and stability (79% stay within Segment 2).
The churn probability is only 2% which is remark-
able, given that these customers are also likely to
be attractive for competing retail banks. While cli-
ents in Segment 21 also have substantial net wealth,
they do not conduct any custody account transac-
tions and have a significantly larger probability of
closing their account (8%). Finally, customers in Seg-
ment 22 have neither savings nor liabilities with the
bank. Many of these clients only use specific ser-
vices (e.g. safe-deposit boxes) and others have a
credit card for which the associated transactions
are conducted via an alternative correspondent
bank. Very often these clients have already termi-
nated their client relationship and transferred their
net wealth to another bank, but not yet officially
closed their account. Consequently, 33% are likely
to do so in the next period.
19-20
23-24
27-28
31-32
35-36
39-40
43-44
47-48
51-52
55-56
59-60
63-64
67-68
71-72
75-76
4
11
1
14
3
1
-1.000
0
1.000
2.000
3.000
4.000
5.000
6.000
7.000
Figure 2 CLV (in Currency Units) for Clients of Age 19 and Above by Segment (State of Nature)
A MODEL TO DETERMINE CUSTOMER LIFETIME VALUE IN A RETAIL BANKING CONTEXT
230 European Management Journal Vol. 25, No. 3, pp. 221–234, June 2007
Table 4 CLV (in Currency Units) for Clients of Age 19 and Above by Segment (State of Nature)
A MODEL TO DETERMINE CUSTOMER LIFETIME VALUE IN A RETAIL BANKING CONTEXT
European Management Journal Vol. 25, No. 3, pp. 221–234, June 2007 231
Using these transition probabilities, the customer
lifetime values for each customer segment were
determined using backward induction as described
above. As can be seen in Table 4 and Figure 2, clients
with highest CLV can be found in Segment 20 (net
wealth above 22,500 and custody account turnover
above 4,500), closely followed by Segments 17, 16
and 9. This is especially true for elderly customers,
mainly caused by their higher contribution margins
and higher probability to stay within these attractive
segments. Only for clients of age 60 and above does
CLV start to decline due to shorter overall lifetimes.
Next to customers with high custody account turn-
over, users of account overdrafts (Segment 3) and
large personal loans or mortgages (Segment 2) are
also part of the top quartile with respect to CLV, as
long as these customers are in low or medium age
segments. The high profitability of users of overdraft
facilities can be explained by the high interest rates
charged for these products as well as the fact that
many of these customers subsequently consume per-
sonal loans of significant amounts (transition to Seg-
ment 2). However, retired customers within these
segments prove to have only medium CLV, due to
the fact that the majority of interest payments have
already been conducted before this point in time.
The high importance of net wealth and custody
account turnover can also be seen when looking at
the bottom quartile with respect to CLV as these cus-
tomers are mainly characterized by low net wealth
and the absence of a custody account. Inactive client
relationships (Segment 4) as well as customers with
very low net wealth and low usage intensity (Seg-
ments 5 and 22) fall into this category. Their low
CLV is hereby less a function of low current usage
but more a consequence of the fact that these custom-
ers primarily use alternative correspondent banks to
conduct payment transactions and are unlikely to
change this in future. With respect to the remaining
50% of clients with medium CLV, customers are
characterized by medium net wealth and the absence
of a custody account (Segments 11, 12, 13, 14) as well
as custody account ownership, but no (Segment 10)
or very limited turnover (Segments 8, 15). Addition-
ally, customers with low net wealth but high usage
intensity (Segment 6) belong to this category, unlike
clients with low net wealth and low usage intensity
(Segment 5) who belong to the bottom quartile. After
client net wealth and custody account turnover,
usage intensity counts among the most important
profitability drivers. For example, the differences
between Segments 11 and 12 in terms of CLV are
only a function of differences in usage intensity.
Limitations and Areas of Further Research
Summarizing our findings, we have highlighted that
M&A transactions have become of increasing impor-
tance for the European retail banking industry in
recent years. Since M&A activities are, at the end,
nothing other than the acquisition of the customer
base of one company by another one, this trend has
resulted in an increasing interest in questions of cus-
tomer valuation. We therefore proposed a model to
value retail banking customers that is based on a
combination of first-order Markov chain modeling
and CART analysis. Using the profitability driver’s
age, demographics/lifestyle, type and intensity of
product ownership and activity level as predictor
variables, we carried out age-dependent CART-anal-
yses to split customers of similar age into homoge-
neous sub-groups concerning the target variable
contribution margin. These groups then served as
states of nature between which we allowed custom-
ers to flow following a first-order Markov process
with corresponding transition probabilities being
approximated by estimated transition frequencies.
Finally, CLV for each customer was determined as
the discounted sum of state-dependent contribu-
tion margins, weighted by their corresponding tran-
sition probabilities. As can be seen our model can
deal equally as well with discrete one-time transac-
tions as with continuous revenue streams, is based
on the analysis of homogeneous groups instead of
individual customers and is easy to understand
and parsimonious in nature. It therefore fulfills the
three general conditions for such a model as stated
in the introduction. Finally, we validated our model
using 6.2 million datasets to show how it can be
applied in day-to-day business life and how it has
been implemented at our cooperating retail bank.
Due to its simple structure and the fact that it only
requires data available in an average bank’s IT sys-
tem, our model can easily be implemented by any
retail bank, where it may serve as a tool for customer
base analysis. For example, customers could be
grouped according to their CLV and, for each of
the resulting customer groups, the company could
then define a specific customer relationship manage-
ment strategy. Our model could also serve as a tool
to optimize the current client base and focus on
unprofitable customers, either by serving them using
special business models (Rosenblum et al., 2003)or
by abandoning them (Haenlein et al., 2006). Finally,
a retail bank could use our model to assign acquisi-
tion allowances for new customers by comparing
prospects with existing customers and, hence, esti-
mating the lifetime value to be expected from an
acquisition prospect beforehand. Beyond the retail
banking industry, companies could build on our gen-
eral approach (i.e. combining CART analysis with
Markov chain modeling) to create similar models
that take account of the specific requirements to be
met in their industries.
Despite its merits and ease-of-use, our approach also
encompasses some limitations with respect to model
building. First, we assumed client behavior to follow
a first-order Markov process, where the transition
probabilities depend only on the behavior during
A MODEL TO DETERMINE CUSTOMER LIFETIME VALUE IN A RETAIL BANKING CONTEXT
232 European Management Journal Vol. 25, No. 3, pp. 221–234, June 2007
the last period. Although this does not mean that
behavior of earlier periods has no influence on cur-
rent behavior, there is no explicit modeling of such
effects. Second, our analysis builds on the assump-
tion that the transition matrix will be stable and con-
stant over time, which seems appropriate for
medium-term forecasts, as long as there are no obvi-
ous and foreseeable reasons for a dramatic shift in
customer behavior. It might not, however, be a sensi-
ble assumption for long-term forecasting. Regarding
areas of further research, we think that combining
our approach with the work on customer equity of
Rust et al. (2004) could be particularly interesting.
In our model, we assumed marketing budgets to be
constant for all customers. Comparable to the
approach of Rust et al. (2004), one could relax this
assumption and estimate the effect of customer-spe-
cific marketing activities on the transition probabili-
ties between the different states of nature.
However, unlike their approach, our transition prob-
abilities do not represent (external) brand switching,
but (internal) changes in customer behavior. Since
the transition probabilities have a direct influence
on the CLV of each customer, it would then be pos-
sible to determine the potential increase or decrease
in individual or aggregated CLV resulting from these
marketing activities.
Notes
1. Details on the mathematical computations and formulas used
can be obtained from the first author on request.
2. In the following discussion and interpretation we only focus on
the results for clients of age 19 and above. The corresponding
results for clients younger than 19 years can be obtained from
the first author on request.
3. Due to space constraints, we do not show each of the resulting
39 transition matrices in detail. However, detailed results can be
obtained from the first author on request.
References
Armstrong, Scott J. (1971) Exploratory analysis of market-
ing data: A reply. Journal of Marketing Research 8(4),
511–513.
Armstrong, J. Scott and Andress, James G. (1970) Explor-
atory analysis of marketing data: Trees vs regression.
Journal of Marketing Research 7(4), 487–492.
Berger, Paul D. and Nasr, Nada I. (1998) Customer lifetime
value: Marketing models and applications. Journal of
Interactive Marketing 12(1), 17–30.
Berger, Paul D., Weinberg, Bruce and Hanna, Richard C.
(2003) Customer lifetime value determination and
strategic implications for a cruise-ship company.
Database Marketing and Customer Strategy Management
11(1), 40–52.
Breiman, Leo, Friedman, Jerome H., Olshen, Richard A.
and Stone, Charles J. (1984) Classification and Regression
Trees. Wadsworth International Group, Belmont.
Campbell, Dennis and Frei, Frances (2004) The persistence
of customer profitability: Empirical evidence and
implications from a financial services firm. Journal of
Service Research 7(2), 107–123.
Crocker, Douglas C. (1971) Comments on: Exploratory
analysis of marketing data: Trees vs. Regression.
Journal of Marketing Research 8(4), 509–513.
Dwyer, F. Robert (1989) Customer lifetime valuation to
support marketing decision making. Journal of Direct
Marketing 3(4), 8–15.
Dwyer, F. Robert, Schurr, Paul H. and Oh, Sejo (1987)
Developing buyer-seller relationships. Journal of Mar-
keting 51(2), 11–27.
Fader, Peter S., Hardie, Bruce G.S. and Lok Lee, Ka. (2005a)
Counting your customers the easy way: An alternative
to the Pareto/ NBD model. Marketing Science 24(2),
275–284.
Fader, Peter S., Hardie, Bruce G.S. and Lok Lee, Ka. (2005b)
RFM and CLV: Using iso-value curves for customer
base analysis. Journal of Marketing Research 42(4),
415–430.
Garland, Ron (2002) Non-financial drivers of customer
profitability in personal retail banking. Journal of
Targeting, Measurement and Analysis for Marketing
10(3), 233–248.
Garland, Ron (2004) Share of wallet’s role in customer
profitability. Journal of Financial Services Marketing 8(3),
259–268.
Haenlein, Michael, Kaplan, Andreas M. and Schoder,
Detlef (2006) Valuing the real option of abandoning
unprofitable customers when calculating customer
lifetime value. Journal of Marketing 70(3), 5–20.
Jackson, Barbara Bund (1985) Build customer relationships
that last. Harvard Business Review 63(6), 120–228.
Jain, Dipak and Singh, Siddhartha S. (2002) Customer
lifetime value research in marketing: A review and
future directions. Journal of Interactive Marketing 16(2),
34–46.
Jarvis, Cheryl Burke, MacKenzie, Scott B. and Podsakoff,
Philip M. (2003) A critical review of construct indica-
tors and measurement model misspecification in
marketing and consumer research. Journal of Consumer
Research 30(2), 199–218.
Keane, Timothy J. and Wang, Paul (1995) Applications
for the lifetime value model in modern newspa-
per publishing. Journal of Direct Marketing 9(2), 59–66.
Kotler, Philip (1974) Marketing during periods of shortage.
Journal of Marketing 38(3), 20–29.
Libai, Barak, Narayandas, Das and Humby, Clive (2002)
Toward an individual customer profitability model: A
segment-based approach. Journal of Service Research
5(1), 69–76.
Morrison, Donald G., Chen, Richard D.H., Karpis, Sandra
L. and Britney, Kathryn E.A. (1982) Modeling retail
customer behavior at Merrill Lynch. Marketing Science
1(2), 123–141.
Payne, Adrian and Frow, Pennie (2005) A strategic frame-
work for customer relationship management. Journal
of Marketing 69(4), 167–176.
Pfeifer, Phillip E. and Carraway, Robert L. (2000) Modeling
customer relationships as Markov chains. Journal of
Interactive Marketing 14(2), 43–55.
Reichheld, Frederick F. and Sasser, W. Earl, Jr. (1990) Zero
defections: Quality comes to service. Harvard Business
Review 68(5), 105–111.
Reinartz, Werner J. and Kumar, V. (2000) On the profit-
ability of long-life customers in a noncontractual
setting: An empirical investigation and implications
for marketing. Journal of Marketing 64(4), 17–35.
Reinartz, Werner J. and Kumar, V. (2002) The mismanage-
ment of customer loyalty. Harvard Business Review
80(7), 86–94.
Rosenblum, David, Tomlinson, Doug and Scott, Larry
(2003) Bottom-feeding for blockbuster businesses.
Harvard Business Review 81(3), 52–59.
Rust, Roland T., Lemon, Katherine N. and Zeithaml,
Valarie A. (2004) Return on marketing: Using
A MODEL TO DETERMINE CUSTOMER LIFETIME VALUE IN A RETAIL BANKING CONTEXT
European Management Journal Vol. 25, No. 3, pp. 221–234, June 2007 233
customer equity to focus marketing strategy. Journal of
Marketing 68(1), 109–127.
Schmittlein, David C., Morrison, Donald G. and Colombo,
Richard (1987) Counting your customers: Who are
they and what will they do next?. Management Science
33(1), 1–24.
Selden, Larry and Colvin, Geoffrey (2003) M&A needn’t be
a loser’s game. Harvard Business Review 81(6), 70–
79.
Shugan, Steven M. (2005) Brand loyalty programs: Are they
shams?. Marketing Science 24(2), 185–193.
Srivastava, Rajendra K., Shervani, Tasadduq A. and Fahey,
Liam (1998) Market-based assets and shareholder
value: A framework for analysis. Journal of Marketing
62(1), 2–18.
Srivastava, Rajendra K., Shervani, Tasadduq A. and Fahey,
Liam (1999) Marketing, business processes and share-
holder value: An organizationally embedded view of
marketing activities and the discipline of marketing.
Journal of Marketing 63(4), 168–179.
Styan, George P.H. and Smith, Harry, Jr. (1964) Markov
Chains applied to marketing. Journal of Marketing
Research 1(1), 50–55.
Thompson, William W., Jr. and McNeal, James U. (1967)
Sales planning and control using absorbing Markov
Chains. Journal of Marketing Research 4(1), 55–62.
Venkatesan, Rajkumar and Kumar, V. (2004) A customer
lifetime value framework for customer selection and
resource allocation strategy. Journal of Marketing 68(4),
106–125.
MICHAEL
HAENLEIN,Department
of Marketing, ESCP-EAP
European School of Man-
agement, 79 Avenue de la
Re
´publique, F-75007 Paris,
France, E-mail: haenlein@
escp-eap.net
Michael Haenlein is Pro-
fessor of Marketing at the
ESCP-EAP European
School of Management. He
holds a Ph.D. and an MSc in Business Administration
from the Otto Beisheim Graduate School of Manage-
ment. Before joining ESCP-EAP, he worked five years as
a Strategy Consultant for Bain &Company. His
research interests lie in the area of customer lifetime
valuation, structural equation modelling, and stochastic
marketing models.
ANDREAS M.
KAPLAN,Department of
Marketing, ESSEC Busi-
ness School, Avenue Ber-
nard Hirsch B.P. 50105,
F-95021 Cergy-Pontoise
Cedex, France, E-mail:
mail@andreaskaplan.eu
Andreas M. Kaplan is
Professor of Marketing at
the ESSEC Business
School Paris. He com-
pleted his Ph.D. at the University of Cologne and HEC
School of Management Paris. Andreas holds an MPA
from the E
´cole Nationale d’Administration, an MSc in
Business Administration from the ESCP-EAP Euro-
pean School of Management and a BSc in Business
Administration from the University of Munich.
ANEMONE J. BEESER,
Strategy Consultant at
McKinsey &Company
Inc., Taunustor 2,
D-60311 Frankfurt am
Main, Germany, E-mail:
anemone_beeser@mckinsey.
com
Anemone Beeser is a
Strategy Consultant at
McKinsey and Company.
She holds a Ph.D. in
Business Administration from Frankfurt University
and an MSc in Business Administration from the Otto
Beisheim Graduate School of Management. Addition-
ally, Anemone attended the MBA programme at
Pennsylvania State University.
A MODEL TO DETERMINE CUSTOMER LIFETIME VALUE IN A RETAIL BANKING CONTEXT
234 European Management Journal Vol. 25, No. 3, pp. 221–234, June 2007