Journal of Marketing Article Postprint

© 2010, American Marketing Association

All rights reserved. Cannot be reprinted without the express
permission of the American Marketing Association.
Balancing Risk and Return in a Customer Portfolio

Crina O. Tarasi
Assistant Professor of Marketing
Central Michigan University
100 Smith Hall,
Mount Pleasant, MI 48859
(989) 774-3092

T
[Link]@[Link]

IN
Ruth N. Bolton
2009-11 Executive Director
Marketing Science Institute
1000 Massachusetts Avenue

PR
Cambridge, MA 02138
rbolton@[Link]
(617) 491-2060

Michael D. Hutt
Ford Motor Company Distinguished Professor of Marketing
W. P. Carey School of Business
T
P. O. Box 874106
Arizona State University
O

Tempe, AZ 85287-4106
(480) 965-6205
[Link]@[Link]
N

Beth A. Walker
State Farm Professor of Marketing
W. P. Carey School of Business
O

P. O. Box 874106
Arizona State University
Tempe, AZ 85287-4106
(480 965-3621
D

[Link]@[Link]

Acknowledgments:
The authors acknowledge the support of the Marketing Science Institute and are grateful to the managers
at the sponsor firm for their responsive support and high level of cooperation. The authors thank Don
Lehmann, Kay Lemon, and Michael Mokwa for their valuable comments. In addition, the authors thank
the three anonymous JM reviewers and the editor for their incisive comments and useful suggestions.

October 26, 2009

 2

Balancing Risk and Return in a Customer Portfolio

ABSTRACT

Marketing managers can increase shareholder value by structuring a customer portfolio to reduce
the vulnerability and volatility of cash flows. This article demonstrates how financial portfolio
theory provides an organizing framework for (1) diagnosing the variability in a customer
portfolio, (2) assessing the complementarity/similarity of market segments, (3) exploring market
segment weights in an optimized portfolio, and (4) isolating the reward-on-variability that

T
individual customers or segments provide. Using a 7-year series of customer data from a large
business-to-business firm, the authors demonstrate how market segments can be characterized in

IN
terms of risk as well as return. Next, they identify the firm’s efficient portfolio and test it against
(1) its current portfolio and (2) a hypothetical profit-maximization portfolio. Then, using
forward- and back-testing, the authors show that the efficient portfolio has consistently lower

PR
variability than the current customer mix or the profit-maximization portfolio. Guidelines are
provided for incorporating a risk overlay into established customer management frameworks.
The approach is especially well-suited for business-to-business firms that serve market segments
drawn from diverse sectors of the economy.
T
Keywords
O

customer portfolio management; market-based assets; financial portfolio theory; return-on-

marketing; market segmentation
N
O
D
 3

“The advantage of knowing about risks is that we can change our behavior to
avoid them. . . . Optimal behavior takes risks that are worthwhile.”—Robert F.
Engle III, Nobel Prize Acceptance Lecture, December 8, 2003, p. 326

While risk management is central to financial portfolio theory and occupies the attention

of CFOs (chief financial officers), sparse attention has been given to risk in the theory and

practice of market segmentation and customer portfolio management. The existing portfolio of

most firms reflects incremental and uncoordinated decisions from the past where little attention

T
was given to how newly-acquired customers contribute to the profitability and risk of the entire

IN
portfolio. For example, Homburg, Steiner, and Totzek (2009) find that firms tend to overestimate

the value of top-tier customers and underestimate the value of bottom-tier customers. In a similar

PR
vein, Dhar and Glazer (2003, p. 88) observe that few companies bother to consider “whether all

of their individually desirable customers are, from the standpoint of risk, desirable collectively.”

This practice is at odds with financial portfolio theory that posits that, even though assets are
T
selected individually, performance is measured on the entire portfolio, where there is a tradeoff
O

between risk and return (Markowitz 1952). We theorize that, like a financial portfolio, a
N

customer portfolio is formed by making choices among market-based assets (i.e., customers) that

present different risk-reward characteristics and by allocating resources to optimize performance

O

(Gupta and Lehmann 2005; Srivastava, Shervani, and Fahey 1998).

D

The purpose of our research is to explore how financial principles of diversification and

the tenets of financial portfolio theory can be effectively applied to manage a firm’s customer

portfolio. We demonstrate how fundamental tools of analysis used by professional investors in

constructing and managing a stock portfolio can be adapted and used to enrich the market

segmentation and customer portfolio decisions of a firm. First, we aim to identify risk that can

(and should) be divested away because firms do not reap higher returns for assuming it and
 4

(instead) suffer losses when market conditions change. Second, we seek to identify ways to

construct efficient customer portfolios. Third, we build on these components to develop an

actionable approach that, looking beyond the returns from individual customers, exploits the

synergies of a diverse customer base characterized by heterogeneous risk-return profiles, and

provides a new approach for managing the market-based assets of a firm.

An initial application of financial portfolio theory in marketing to product portfolio

decisions (Cardozo and Smith 1983) spawned criticism from Devinney, Stewart, and Shocker

T
(1985) who identified key differences between financial and product investment decisions,

IN
arguing that crucial assumptions of the theory were violated (see reply by Cardozo and Smith

PR
1985). Recent research, however, demonstrates the potential insights that financial portfolio

theory may contribute to customer portfolio management. Dhar and Glazer (2003) describe the

importance of measuring the riskiness of customers (i.e., customer beta) and illustrate how a firm

can maximize returns by acquiring or retaining particular customers or market segments on the
T
basis of how their spending patterns contribute to the diversification of the cash flow of the
O

overall customer portfolio.

N

Ryals (2002; 2003) also adopts a financial theory perspective to examine the risk and

return characteristics of a customer portfolio and describe how a customer relationship scorecard
O

can be used to assess customer risk. Likewise, Buhl and Heinrich (2008) offer a quantitative
D

model based on financial portfolio theory that (1) considers the customer lifetime value (CLV) as

well as the associated risks of customer segments and (2) provides a method for adding or

subtracting market segments. Using a case study from the financial services industry, they test

the model by using the average annual incomes of key customer segments (e.g., lawyers,

physicians) as an indicator of cash flows and demonstrate how the optimal portfolio provides

both higher utility and better risk diversification than the existing portfolio.
 5

This study makes the following contributions to customer portfolio theory and practice.

First, we build on past research to show—theoretically and empirically—how to make a nuanced

assessment of customer value by calculating the customer beta (Dhar and Glazer 2003; Buhl and

Heinrich 2008) and contribute a new metric for customer portfolio management—the customer

reward ratio. Customer beta provides a relative measure of the sensitivity of an individual

customer’s cash flow return to the return of the firm’s current customer portfolio. By adjusting

for variability, the customer reward ratio, drawn from work by Sharpe (1994), takes into account

T
the risk/reward tradeoff associated with the customer. Second, the study evaluates the extent to

IN
which classic market segmentation variables (e.g., demographics or firmographics) can be used

PR
to predict cash flow characteristics (i.e., risk-return profiles) of customers, so that managers can

assess how potential customers (as well as existing customers) might contribute to the customer

portfolio. Note that unlike some prior research that controls for customer heterogeneity (Niraj,

Gupta, and Narasimhan 2001; Venkatesan and Kumar 2004), our approach evaluates and
T
exploits customer heterogeneity to improve business performance.
O

Third, the current study is responsive to calls for research that examines the financial
N

impact of customer portfolio management decisions (e.g., Rust et al. 2004). We show—

conceptually and empirically—how a firm can identify synergies among customers and assemble
O

the optimal mix of customers by constructing an efficient frontier for customer portfolios. The
D

efficient frontier is the set of optimal customer portfolios characterized by minimum risk for a

certain level of return, or maximum return for a certain level of risk, so it describes alternative

customer portfolios for the firm. Finally, Tuli, Bharadwaj, and Kohli (2010) provide evidence

that the number and types of ties a company builds with its best customers ensures not only

higher revenue, but also reduces the variability of their purchases. We extend their work by

demonstrating that the firm can manage its portfolio of customer relationships to control the
 6

overall variability of the firm’s cash flow.

We test the feasibility of applying key financial concepts to customer portfolios by

implementing them using a 7-year series of customer data from a large business-to-business

company. We begin by exploring whether we can segment the firm’s customer base in ways that

are comparable with the classification of financial assets. Then, we identify the firm’s efficient

customer portfolio and test it against (1) its current portfolio and (2) a hypothetical profit-

optimization portfolio. Our results show that customers exhibit substantial differences in their

T
risk-return profiles and that clustering techniques can be used to identify market segments for

IN
building efficient portfolios. Most importantly, we demonstrate that the firm’s efficient portfolio

PR
has constantly lower variability than the current customer mix or the profit-maximization

portfolio, while the profit performance is superior in the long run.

Conceptual Framework
T
This section reviews financial portfolio theory and conceptualizes how key financial constructs
O

can be applied to customer portfolios. Then we describe how these financial constructs can be

calculated from customer purchase history data. Next, our attention turns to how firms can
N

identify the most desirable customers by assessing the rate of reward on risk for each customer.
O

Last, we address how firms can use these constructs and measures to segment the firm’s

customer base in ways that are comparable with the classification of financial assets.
D

Financial Portfolio Theory

Financial portfolio theory describes how investors can construct portfolios to maximize return

based on a given level of market risk, emphasizing that risk is an inherent part of higher reward

(Markowitz 1952). In a stock portfolio, the lower the total correlation of a stock with the total

return, the more desirable the particular stock is to the portfolio. For example, stocks drawn from

different industries, different countries, and different-size companies are affected by

 7

environmental and economic changes in specific ways (Niemira and Klein 1994). Since many

market changes cannot be anticipated, diversification ensures that the portfolio includes positive

cash flow opportunities and smoothes out potentially negative cash flows. Based on the

variability and return of each of the assets, the optimal (efficient) portfolio is considered to be the

one that has the least risk for a desired level of return or the highest level of return for a certain

level of risk. Any other portfolio would be suboptimal. The set of efficient portfolios form the

efficient frontier, which borders the set of all possible portfolios (Markowitz 1987).

T
Among the criticisms of financial portfolio theory is the assumption that asset returns are

IN
normally distributed whereas large swings in the market occur far more frequently than the

PR
normal distribution would predict. [E.g., the S&P 500 stock index has experienced a 3-standard-

deviation negative monthly return event 10 times since 1926, while a normal distribution would

predict such extreme returns perhaps 1 or 2 times (Kaplan 2009).] Another criticism centers on

the assumption that correlations between assets are stable. However, during periods of market
T
stress, assets that were previously found to be uncorrelated can suddenly move in lockstep (e.g.,
O

Hubbard 2009).
N

Criticism has also been levied against the efficient portfolio concept. For example,

DeMiguel, Garlappi, and Uppal (2007) evaluated 14 different optimal portfolio models that have
O

been advanced in the finance literature, based largely on the Capital Asset Pricing Model
D

(CAPM) (Sharpe 1999), and found that none is better than a naive approach where an investor

allocates a fraction of wealth to each of the assets available for investment. However, by drawing

on financial portfolio theory and defining the market portfolio as the existing customer base of a

firm, our focus differs from the CAPM employed in finance that assumes that the market

portfolio includes all available assets where each asset is weighted by its market capitalization

(see Buhl and Heinrich [2008] for a critique in the context of customer portfolios).
 8

Despite these criticisms, financial portfolio theory “plays a role in almost every area of

financial practice and can be a useful tool for many important managerial decisions.” (Grinblatt

and Titman 2002, p. 97) In turn, the theory has been used to inform economic development

strategies at multiple levels of analysis. By applying financial portfolio theory to regional

economics, Conroy (1974) introduced a method for measuring economic diversification that

spawned a rich research tradition in the regional science literature (see Dissart 2003 for a

review). In this context, a region represents a portfolio of assets (industry sectors) that make up

T
the local economy whereby each industry yields a return (employment) but also entails a risk

IN
(employment volatility).

PR
This portfolio management framework has been used to study the growth-instability

tradeoffs of metropolitan areas (e.g., Conroy 1974), individual states and the U. S. economy

(e.g., Lande 1994), and international regions, including Western Europe (Chandra 2003). For

example, Lande (1994) examines the economic structure of selected states, identifying those
T
industry sectors that contribute to employment growth and stability in an optimal portfolio.
O

Collectively, studies from this research tradition lend strong support to our view that financial
N

portfolio theory may provide a valuable framework for evaluating and managing a customer

portfolio, particularly for business-to-business firms that serve customers drawn from diverse
O

industry sectors.
D

Cash Flow Stability and Firm Value

Extending the work of Srivastava, Shervani, and Fahey (1998), we theorize that the market-based

assets of a firm include distinct customer asset classes that are characterized by differing degrees

of cash flow variability and vulnerability. Customer asset classes represent the market segments

that comprise the existing customer base and embody the outcomes of relationships between the

firm and its customers. While investment portfolio decisions involve choices within and among
 9

various asset classes of stocks and bonds, customer portfolio decisions involve choices within

and among distinct customer asset classes (e.g., governance type, size, industry) that encompass

both new and existing customers in the served market and present different risk-return profiles

for the firm. In support, Gupta, Lehman, and Stuart (2004) assert that “customers are indeed

assets, and therefore customer-related expenditures should be treated as investments rather than

expenses.” They demonstrate how the value of the customer base provides a strong guideline for

firm value.

T
In choosing among customers to add to a portfolio, the less a customer’s purchasing

IN
behavior promises to be like that of the current portfolio, the stronger its contribution to the

PR
stability and predictability of the portfolio; the more the behavior is like that of the existing

portfolio, the weaker its contribution. Therefore, the attractiveness of a customer hinges not only

on the size and frequency of purchases but also on the degree to which the customer’s pattern of

purchases co-varies with those of other customers in the portfolio. The declining cash flow from
T
one customer may be offset by increased returns from another. During a recession, a
O

transportation company may, for example, experience a decline in revenue from discretionary
N

retailers that is offset by an increase from discount retailers or declining revenue from auto

producers is partially offset by a growing revenue stream from after-market auto parts retailers.
O

By developing a risk-adjusted customer portfolio to achieve profit targets, marketing

D

managers can contribute to firm value. Since investors favor stable earnings over volatile

earnings (Ang, Chen, and Xing 2006; Srinivasan and Hanssens 2009) and cash flows that are

more stable and predictable reduce working capital needs (Rao and Bharadwaj 2008; Srivastava,

Shervani, and Fahey 1998), firms can enhance shareholder value by reducing the vulnerability

and volatility of cash flows from the customer portfolio.

Customer Portfolios versus Financial Portfolios

 10

While customers, like stocks, represent risky assets and the cost of acquiring them should reflect

the cash flow they are expected to generate over time, key differences exist between a financial

portfolio and a customer portfolio with respect to the nature of the assets, returns, and

uncertainty.1

Assets. By representing but one of many levels of marketing investment made by a firm,

customer portfolio decisions are embedded in a far more complex investment management

framework than financial portfolio decisions. Marketing investments are made to enhance the

T
value of brand assets through, for example, product research and development, channel support,

IN
and advertising. To enhance the value of customer assets, the firm gives special emphasis to

PR
investments in customer relationship-building by using elements of the marketing promotion mix

(e.g., communications, sales force, customer-firm interactions). (Ambler et al 2002)

Financial assets can be identified and readily purchased while particular customers can be

targeted but there is no assurance that the firm will be successful in attracting them to the
T
portfolio. Likewise, individual customer relationships take time to develop and usually require
O

continuing investments (Johnson and Selnes 2004; Kumar 2008). Therefore, the price of an
N

existing customer asset is the retention costs represented by these continuing expenditures while

the price of a new customer asset is the associated acquisition costs. Compared to a customer
O

portfolio, investors can also readily make portfolio adjustments by selling assets at a market
D

price and by changing the proportion (weight) assigned to particular asset classes. By contrast,

there is no liquid market for customer assets (Kundisch, Sackmann, and Ruch 2008) and

customer divestment may be costly and represents a strategic option that must be exercised

sparingly (Mittal, Sarkees, and Murshed 2008).

Financial assets can be purchased in parcels of any size but customer assets are not

infinitely divisible and major portfolio adjustments may be costly and difficult to implement in a
 11

timely manner. An investor who wishes to increase the portfolio weighting of a particular

industry sector can readily implement this change by selling stocks from one sector (e.g., energy)

and buying stocks in another (e.g., technology). To make corresponding changes in the

weighting of market segments within the customer portfolio, a manager faces a longer time

horizon, new strategy priorities, and a host of rigidities that the current strategy imposes. To

illustrate, reorganizing the customer portfolio may require a realignment of sales and marketing

communication strategies, highlighting the higher transaction costs associated with customer

T
versus financial portfolios.

IN
Return. For an investor, return is the change in value of the investment, which includes

PR
capital appreciation (or loss) plus the cash yield. Unlike financial assets, customers may be

interconnected and contribute to the return of a market segment through social processes, such as

positive word-of-mouth (Ryals 2003). Return for the customer portfolio is the cash flow and

profit (revenue minus cost-to-serve) that accrue to the firm from investments made in individual
T
customers and market segments. Clearly there are a host of other marketing investments made by
O

a firm that enrich the customer relationship strategy but that are not directly captured in the cost-
N

to-serve calculation. To illustrate, the decision to increase the weighting of particular market

segments within a customer portfolio may require corresponding investments in new product
O

development or service support that go beyond the direct customer costs that we consider.
D

In a financial portfolio, the rate of return is independent of the amount invested. In

contrast, a distinct difference between a customer portfolio and a financial portfolio is that the

returns from investing in customers are likely to be nonlinear. Specifically, the amount of

investment has a nonlinear relationship with the “return on customer,” which means, for

example, that small investments might be insufficient to attract or retain an individual customer

or market segment.
 12

Managerial control is among the unique characteristics that distinguish customer

investments from financial investments (Devinney and Stewart 1988). For example, a firm can

increase sales and reduce sales variability by forging multiple types of relationship ties with a

customer organization (Tuli, Bharadwaj, and Kohli 2010) or enhance returns by identifying

elements of its customer management effort that provide the greatest marginal return on

additional investments (Bowman and Narayandas 2004). When an investor chooses an optimal

weight for a particular asset class and purchases the associated securities, there is no impact on

T
the risk and return for that asset class. In contrast, managers can exercise a rather significant

IN
degree of control over the risk and return characteristics of the customer portfolio. For example,

PR
the weight assigned to a market segment may affect the performance of that segment because of

increasing or decreasing returns to scale. Some market segments complement the economies of

the seller’s business better than others and some customers within these segments are less costly

to serve than others.

T
Uncertainty. The difference between the expected return and the actual realized return of
O

an asset constitutes uncertainty in financial portfolio theory. Investment uncertainty is

N

characterized by the variability, or risk, in the return of a security, namely the deviation of the

return from expected value during the holding period. Variations in the returns of securities are
O

then used to estimate the covariance among the array of assets that comprise a portfolio. In the
D

customer portfolio context, the deviation of customer cash flow and profit from their expected

values provides a measure of risk. However, there are other sources of uncertainty that are

unique to a customer portfolio. Unlike financial assets that can be retained as long as the investor

desires, customers can take independent actions and defect or shift a share of their total

purchases to a competitor. Therefore, customer cash flow stability provides a rather narrow

measure of the strength of a customer relationship.

 13

Customer Portfolio Applications. To capitalize on the strength of financial portfolio

analysis while managing the associated constraints and limitations, a firm can examine the risk-

return characteristics and structure of the current customer portfolio. Specifically, we will

demonstrate how portfolio theory provides an organizing framework and supporting

methodology for (1) diagnosing the variability in the overall customer portfolio, (2) assessing the

complementarity/similarity of market segments, (3) exploring the weights of market segments in

an optimized portfolio, and (4) gauging the reward-on-variability that individual customers or

T
segments provide.

IN
In applying financial theory to customer portfolio management, some key limitations

PR
must be understood and managed. First, during periods of severe economic stress, market

segments that were previously uncorrelated can suddenly move in tandem, limiting the benefits

of diversification. Second, our approach determines how the current customer base might be

reconfigured into an optimal portfolio but some of these adjustments are costly and raise a host
T
of strategic issues beyond the scope of our analysis. The optimal portfolio can best be viewed as
O

an ideal customer base that managers can evaluate, revise, and assemble over time. Therefore,
N

the optimization process should include a qualitative overlay based on managerial judgment to

arrive at recommended resource allocations by segment. In fact, Markowitz, in his seminal paper
O

(1952, p. 91), emphasizes that the statistical results that issue from his approach should be
D

viewed as tentative and then enriched by judgment “on the basis of factors or nuances not taken

into account by the formal computations.”

Third, our analysis examines the cash flow and profit of individual customers but does

not assess other important customer metrics, including customer satisfaction, loyalty, or share of

wallet. Likewise, we do not consider the host of factors that influence individual customer

profitability, such as demand stimulating efforts by the firm or competitive behavior (Bowman
 14

and Narayandas 2004). To that end, our conceptualization of customer portfolio risk

complements, rather than replaces, other approaches from the customer management research

tradition that examine other types of risk such as the risk of defection or the probability of

achieving customer lifetime value outcomes (e.g., Blattberg, Getz, and Thomas 2001; Bolton,

Lemon, and Verhoef 2008; Rust, Lemon, and Zeithaml 2004).

Appropriate Market Contexts. Our approach specifically applies to those situations

where there are meaningful differences in variability across the market segments that comprise a

T
firm’s customer portfolio. Therefore, we believe that the approach is best suited for the business

IN
market where these conditions are often present. Compared with consumer packaged goods

PR
contexts, business marketers tend to allocate greater proportions of their sales and marketing

resources at the level of individual customers. Likewise, many business-to-business firms serve

market segments drawn from diverse sectors of the economy that each demonstrates a distinct

demand function (Dickson and Ginter 1987). The approach may also be appropriate for those
T
business-to-consumer firms that have direct contact with the customer, such as
O

telecommunications and financial services companies. However, the approach will be less
N

suitable in these or other situations if market segments tend to be highly correlated.

Customer Portfolio: Risk and Reward

O

Markowitz (1987) measured risk using the variability of the price of the asset, which represents a
D

good proxy for the probability of encountering an unexpected outcome. The risk and return

associated with the cash flow of each customer can be computed using purchase history data.

Historic analyses are based on the assumption that the future will be like the past (Sharpe,

Alexander, and Bailey 1999) and variance is very difficult to forecast. However, we will assume

that the relationships and correlations of the past are sufficiently stable and that past variability is

a good proxy for future variability (Balagopal and Gilliland 2005; Chan, Karceski, and
 15

Lakonishok 1999).

Cash flow variability and overall customer portfolio risk. Risk is defined as volatility or

variability associated with cash flow, and it is traditionally estimated using standard deviation or


NA
( x Ai  x A ) 2
variance. The formula for computing the variance of customer A, V A , is: V A  i 1
,
N A 1

and standard deviation is  A  V A

1/ 2
, where x Ai is the cash flow for customer A in the ith period

T
in which a cash flow occurred, x A is the average value of cash flow from customer A for the N A

IN
periods, and N A is the number of periods in which a cash flow from customer A occurred.

In order to obtain a standardized measure of variance that corrects for differences in the

PR
average levels of cash flows across customers, we compute the coefficient of variation,

CV   A / x A .

The risk of the entire portfolio V P will be computed using a similar formula, except that the cash
T
flow used will be the average of all customer cash flows (Markowitz 1987).
O


N
j 1
( x j  xP ) 2
VP  ,
N 1
N

where x j is the cash flow from all customers active in period j, x j  1 j x jk (where N j is the
M

number of customers active in period j and x jk is the cash flow from firm k in period j), N is the
O

number of periods considered, and x P is the average value of cash flow from the customer
D


N
j 1
xj
portfolio for the N periods and M firms, x P  . In order to be able to compare the
N 1

performance of portfolios with different levels of performance (e.g., different means), we will

standardize the values by dividing the monthly values by the mean of the portfolio before

computing variability.
 16

Customer Beta and Customer Reward Ratio

Customer beta. To identify the most desirable customers, we need a reliable measure of

the consistency of returns for an individual customer vis-à-vis a reference customer or portfolio.

In finance applications, beta—a measure of the volatility of an investment—is computed relative

to an appropriate asset class, usually the market portfolio. The market portfolio consists of all

assets, with the weight of each held in proportion to the total market value. Since the

determination of a comparable portfolio that includes all customer assets across all firms

T
represents a daunting, if not impossible, task, we define the market portfolio as the firm’s current

IN
customer base in line with Dhar and Glazer (2003), Ryals (2002), and Buhl and Heinrich (2008).

PR
In the financial context, the market is assumed to be efficient, implying that information

is fully and immediately reflected in market prices (Fama 1970; Sharpe, Alexander, and Bailey

1999). By contrast, the customer portfolio is not efficient. For a company, variations in the

customer portfolio might reflect the overall performance of certain industries or sectors of the
T
economy. Therefore, rather than using beta to describe the risk of the overall portfolio, customer
O

beta captures the degree to which an individual customer contributes to the risk of the entire
N

portfolio.

cov( x i , x P )
i  ,
O

VP

where cov(xi, xP) is the covariance between the individual customer cash flow and the cash flow
D

of the overall customer portfolio, and VP is the variance of the cash flow for the overall customer

portfolio.

Customer reward ratio. In measuring the rate of return on risk of a customer, or in other

words, the reward for assuming variability, Sharpe’s pioneering work (1994; see also 1966)

provides the foundation for the customer reward ratio. The reward is measured as the return
 17

above the risk-free rate

Ri  R f
RRi  ,
i

where RRi represents the customer reward ratio, Ri represents the return for customer i and Rf

represents the return for the risk-free customer proxy, and σi represents the standard deviation of

the return. When there is no risk-free asset available, Rf =0 and the equation is simplified to

Ri

T
RRi  .
i

IN
Finding a risk-free proxy for the customer portfolio (the equivalent of treasury bills, the

benchmark for risk-free investments) is often possible. Some companies, for example, may have

PR
a set of low-return customers that they might prefer not to serve, but they choose to do so to fill

spare capacity and achieve a modest return. These customers, while not directly targeted, provide

a benchmark return.
T
If one of the goals in designing a customer portfolio is to minimize the risk for a certain
O

level of return, a key question becomes: What is the level of return that a customer or segment

with a certain level of variability provides? The customer reward ratio provides the means for
N

evaluating the risk-reward tradeoffs of customers in the portfolio. Provided here is a measure for
O

evaluating the relative attractiveness of customers with different levels of return and variability.

When customers possess similar return or variability characteristics, distinguishing the most
D

desirable customer is straightforward (for the same level of risk, the customer with the highest

return will be preferred, while for the same level of return the customer with the lowest

variability will be preferred, all else being equal). However, when both risk and return are

different, the customer reward ratio provides the means to identify the most attractive customer.

Segmenting or Classifying Customers Based on Risk

 18

In financial markets, assets are grouped into categories that share certain risk-return and

variability characteristics (blue chip stocks, bonds, treasury bills). We can group customers into

segments using cluster analysis based on the monthly variability in their cash flows and then

observe whether the resultant segments share other characteristics that are meaningful and

actionable in the marketplace, such as demographics or firmographics. In other words, two key

questions in determining the feasibility for building an efficient customer portfolio include: (1)

Are there significant differences in variability and rate of return across market segments? and (2)

T
Can we identify the differences in variability associated with specific customer characteristics

IN
(e.g., size of the company, industry)? If the answer to both questions is yes, then we can build

PR
efficient portfolios based on the risk-return profiles of clusters, rather than individual customers

(for which cash flows can be somewhat unpredictable). Therefore, we will first test whether there

are significant differences in cash flow variability among different segments that can be

characterized in ways that are normally used for segmentation. Then we will attempt to construct
T
an efficient customer portfolio and evaluate its performance.
O

Research Design
N

Study context
O

We test the applicability of our approach to customer portfolios using purchase history data from

a business-to-business company with a diverse customer base.2 The client company provided
D

monthly sales and profit data (earnings before interest and taxes or EBIT) for all customers for a

seven-year period. The company’s records also contained information for each customer

concerning number of product lines purchased, size of business, geographic locations, and

industry sector. The company had served over 10,000 customers in the seven years. However, we

focused on the top 250 customers from each of the years from 2001 through 2007, which

amounts to 516 unique customers and 98% of all sales. We supplemented the cooperating
 19

company’s purchase records with information from public databases. Specifically, 456 of the 516

business customers were uniquely identified based on Dun & Bradstreet (D&B) codes so that we

could record the number of employees and sales revenues for specific sites and for the entire

company/customer.

Analysis Plan

In Stage 1, before developing the efficient portfolio, we want to assess whether meaningful

differences in variability exist among customer segments.

T
Stage 2 centers on segmenting (i.e., clustering) customers based on purchasing patterns

IN
(using standardized monthly purchases over six years), rather than using an a priori segmentation

PR
scheme. Next, we identify the segments by examining their financial and nonfinancial

characteristics. For the segmentation to be actionable for managers, customers within the same

segments must share common characteristics, which can be used to identify similar (potential)

customers.
T
Stage 3 centers on identifying the efficient frontier and building an efficient customer
O

portfolio using the variability-based segmentation scheme identified in Stage 2 to develop a

N

diversified portfolio of customers, which should outperform value maximization portfolios in the

long run.
O

Stage 4 centers on an evaluation of the diversified portfolio’s performance. The firm’s

D

business performance should be enhanced in two ways: higher returns or reduced risk (or both).

Hence, we evaluate the success of our approach by comparing the scenario reflecting the

outcomes of the efficient frontier with the “actual” risk-return profile for the following year and

a profit-maximization scenario—all calculated using the holdout sample data.

Stage 5 makes necessary adjustments in the composition of the customer portfolio by

reexamining the performance (purchases) of current customers, individually and by segment.

 20

Using customer reward ratios and customer beta indicators, we can asses the riskiness of

individual customers and gauge their impact on the overall portfolio. Using this information—

and maintaining a perspective of company goals and the external environment—marketing

managers can decide on a case-by-case basis whether it is desirable to attract more business from

the specific customer or to identify segments with similar characteristics to pursue in the future.

Stage One: Assessing Differences in Variability among Customer Segments

Our assessment is based on an examination of the sales over time from different market

T
segments defined on an a priori basis. Specifically, we use the first six years of purchase history

IN
data to investigate whether there are significant differences in coefficients of variation across

PR
market segments defined by relationship type (contractual versus noncontractual), size of

business, and industry type. A full discussion of this analysis is provided in Appendix 1.

Compared to noncontractual relationships, the analysis indicates that contractual

customer relationships have lower variability and their introduction into a customer portfolio
T
reduces the overall variability. Likewise, customers from small and medium-sized businesses
O

(SMBs) have lower variability than large business customers. For the industry analysis, we
N

classified the customers using the NAICS (North American Industrial Classification System)

combined with the Standard & Poor’s (S&P) sector classification.3 Appendix 2 provides
O

customer reward ratios, betas, and coefficients of variation of customer purchases over time in
D

different industry sectors. The industry analysis was inconclusive because many of the industries

were represented by only a few large customers; however, the trends were visibly distinctive.

Figure 1 provides a graph of sales revenue over time for the 10 industry categories of

customers that generated the highest average sales. For example, observe that the retail sector

exhibits a pronounced growth pattern over six years, while all others exhibit more modest

growth, with the auto sector and transportation manager customer registering a noticeable
 21

decline after 2004.

(Figure 1 about here)

Based on this analysis, we conclude that there are statistically significant differences in sales

variability for customers with contractual versus noncontractual relationships and between

customers of different sizes. Also there are meaningful differences in sales trends for customers

from different industries. Therefore, the foundation may be in place to identify market segments

characterized by different risk levels (e.g., betas, customer reward ratios) and to build an efficient

T
customer portfolio for the cooperating company.

IN
Stage Two: Transactional Segmentation

PR
Market segments should be characterized by different demand functions and purchase patterns

(e.g., Dickson and Ginter 1987). Market segmentation based on similarities or differences in

purchasing patterns is called transactional segmentation. Transactional segmentation has been

previously used in financial services firms to determine patterns that signal defections (Pearson
T
and Gessner 1999).
O

Each customer has unique and common characteristics, so we utilized a hierarchical

N

clustering analysis of the monthly purchase data for each customer to observe the common

characteristics. The procedure (PROC CLUSTER, in SAS, using the average linkage method)
O

grouped customers based on squared distances, where distance was measured by the monthly
D

cash flow levels (standardized revenue).4 Since there are 72 months of observation in the data,

each customer is characterized by 72 variables. A six-cluster solution was robust to method

changes, providing support for a solution that is useful for managerial action. The six-cluster

solution grouped together customers with similar trend characteristics. Comparisons among

clusters revealed that, even though the statistical techniques were based on cash flow patterns

exclusively, the resulting clusters differed in terms of company size, dominant industries, overall
 22

variability, customer reward ratios, and betas. The results of these comparisons are presented in

Appendix 3 and the patterns of the clusters are presented in Figure 2.

(Figure 2 about here)

Stage Three: Identifying the Efficient Frontier and Building an Efficient Customer Portfolio

Each cluster has a certain level of return, as presented in Figure 2. The return per cluster was

computed using profitability data by customer as provided by the sponsoring firm (return = total

EBIT per cluster divided by the total revenue per cluster). Based on the six clusters, we can now

T
build an efficient portfolio by minimizing the cash flow variability for 2006 given a certain level

IN
of return. Even though we use the data for 2001 through 2006 to build the clusters, we use 2006

PR
as the reference year—being the closest to the holdout period (2007)—to compute the efficient

frontier. We need to identify a set of optimal weights for each of the clusters X’ = [x1 x2 x3 x4 x5

x6] that minimizes the portfolio variance and that, multiplied by the return per cluster, adds up to

the targeted return. By varying the expected return, we can draw the entire efficient frontier
T
(Markowitz 1991).
O

To develop the efficient frontier, the function quadprog was used in Matlab to minimize
N

variance-covariance matrix for various levels of return (in increments of .2%). The quadprog

function is designed to solve quadratic programming problems, in which a covariance matrix is

O

minimized by varying the weights of the parameters, while satisfying certain linear conditions:
D

1 '
min X HX , such that AX ≤ B and Aeq X= Beq,
2

where X is the vector of weights (XT is X transposed) and H is the return covariance matrix or the

covariance matrix computed using the monthly return for all clusters. The inequality AX ≤ B,

where A=-[I6], I being the identity matrix and B’=[0,0,0,0,0,0], insures that all cluster weights

are positive. The equation Aeq X= Beq, where Aeq=[1,1,1,1,1,1; r1, r2, r3, r4, r5, r6], r1,- r6 being
 23

the actual returns for clusters 1 to 6, and Beq’=[1, R], with R being the target return, insures that

the sum of weights for all clusters is 1 and that the sum of the returns for the efficient portfolio

matches the desired return. Quadratic programming is classically used for mean-variance

portfolio selection (Feldstein 1969). The quadprog function uses the medium scale algorithm for

this type of problem and involves a two-stage approach: first it estimates a feasible point, and

then generates a sequence of feasible points until convergence occurs.5

As expected, the efficient portfolios bordered the set of possible portfolios (Markowitz

T
1959). The efficient portfolio with the lowest risk is portfolio E1 (see Figure 3 and Table 1),

IN
which has a relatively equal representation of all clusters, except cluster 6.6 Observe that cluster

PR
3, which is comprised predominantly of small business customers, has the highest representation

in this portfolio (26%). The efficient portfolio with the highest return is portfolio E10, and is

dominated by cluster 5 (92%, see Table 1), which is the cluster with the highest return.

(Figure 3 and Table 1 about here)

T
The weights of cluster 3 (small business customers) in the efficient frontier portfolios
O

vary from 6% in portfolio E9, to 26% in portfolio E1. As the percentage of cluster 3 decreases,
N

the level of risk increases. This pattern shows how diversity increases the stability of a portfolio

because introducing smaller business customers into a portfolio overweighted with large
O

business customers reduces the risk of the portfolio. In order to have a balanced portfolio, the
D

cooperating company requires a certain percentage of small business customers, but no more

than 26%. Above 26%, the variability of small business customers outweighs the benefits of

diversification. From the standpoint of an individual investor buying securities, the stocks of the

companies that comprise a firm’s efficient customer portfolio would not, of course, represent an

efficient stock portfolio for that investor because a host of factors influence stock values,

including company strategy, new product announcements, operating efficiency, among many
 24

others (e.g., Srinivasan and Hanssens 2009).

Not only is it computationally more efficient to build the efficient frontier using clusters

of customers and not individual customers, but it is also more actionable for managers. An

efficient customer portfolio constructed from individual customers might suggest seeking

incremental sales from a given customer that far exceed the customer’s requirements. By

selecting customers from clusters, the role of similar characteristics is emphasized, making the

identification of potential new customers easier and implementation more straightforward. This

T
approach offers managers the choice of either increasing the level of business conducted with

IN
current customers in the cluster (if the opportunity exists) or serving new customers with similar

PR
characteristics that define the cluster. Moreover, from a practical standpoint, managers can even

apply this approach at the group (cluster) level if they find it difficult to determine the specific

return per customer. The risk could be estimated based on the variability of the customer revenue

and used in combination with the return per cluster to estimate the efficient portfolio.
T
Stage Four: Testing the Efficient Portfolio
O

We have constructed an efficient customer portfolio that minimizes variance for the study
N

period. We compare the performance of this portfolio with that of a profit maximization

portfolio, built using the best customers for 2006 and assuming that the company is able to
O

acquire 25% more customers with the same level of profit as its best customers (which the client
D

company would do if it could).

First, we compare the performance of an efficient portfolio (E5), the actual portfolio and

the profit maximization portfolio and “back-test” them for 2001 through 2005.7 A customary

practice in finance is to test strategies under historical market conditions to evaluate their

viability and effectiveness. This method is especially useful to test a portfolio under different

economic conditions, given that testing with future data is not an option. In Figures 4A and 4B,
 25

we compare the results for the three different portfolios; that is, we compare variability (risk) and

actual profits (return).

(Figure 4 about here)

Using back-testing, we notice that the efficient portfolio constantly has much lower

variability than either of the other two portfolios for all six years examined (Figure 4A). In terms

of profit performance, the efficient portfolio outperforms the actual portfolio each of the years

except for 2004 and 2005, which were extremely profitable for the company (Figure 4B). The

T
profit-maximization portfolio outperforms the actual portfolio and the efficient portfolio for just

IN
one of the years, 2005. In years of high growth, riskier portfolios are more likely to outperform

PR
low-risk portfolios. However, for the other years, further out on the horizon from the benchmark

year for which the portfolio has been optimized, the efficient portfolio outperforms both the

actual portfolio and the profit-maximization portfolio, thereby providing supporting evidence for

the stability of our method.

T
Second, we compare the three portfolios using forward-testing, that is, we use the data for
O

2007 that have not been used in any other previous analysis. (To do so, the customers that have
N

entered the top 250 for the first time in 2007 have been matched to clusters using the size of the

business, industry profile, and previous purchase history.) When comparing 2007 performance,
O

the efficient portfolio outperforms the actual portfolio: higher profit and lower variability. The
D

efficient portfolio has lower overall profitability than the profit-maximization portfolio, but it has

a much lower variability. In stable economic conditions, one would expect that in the first year

(short run), the efficient portfolio might not outperform a profit-maximization portfolio.

To further test the robustness of the efficient portfolio concept as applied to a customer

portfolio, we also built efficient portfolios by using the data for each of the years in the 2001-

2005 interval, then utilizing data from the remaining years as the holdout sample. For example,
 26

we built the efficient frontier for 2001 and used data from 2002 to 2007 as the holdout sample.

Except for 2004, when the company implemented mid-year accounting changes related to the

measurement of customer profitability, the efficient frontier could be fully identified. According

to the simulation results, each of the efficient portfolios had similar benefits: while the

profitability was comparable with that of the current portfolio (Figure 4D), the variability was

constantly much smaller for all of the years examined while controlling for the mean (Figure 4C)

and in absolute value (Figure 5). These results show the stability of the solutions computed for

T
different years; in simulations they all manifested a similar level of profitability and substantially

IN
lower variability.

PR
(Figure 5 about here)

Stage Five: Revising the Current Customer Portfolio toward an Efficient Portfolio

To this point, our analysis has centered on groups of customers that share certain characteristics.

However, inside clusters, some customers might be more desirable than others, and given limited
T
resources, the firm should prioritize its customer retention/acquisition efforts. This issue should
O

be considered when the firm reweights its customer portfolio to move toward an optimal
N

composition.

Recall that the customer reward ratio can be measured as a function of a risk-free asset
O

(in our case, a risk-free customer proxy) or in absolute terms. By incorporating the risk-free asset
D

into the calculation, the customer reward ratio provides a more meaningful measure of the

relative attractiveness of alternative customer investments.8 The manager can consider the return

provided by investing in the risk-free asset, and take into account how a diversified allocation of

resources might be more attractive than the investment in a single asset.

Consider the risk-free asset that a logistics services company might use. Transportation

managers act as brokers for small and medium-size companies. They are often used by logistics
 27

service companies to find customer shipments to fill at least some capacity for return routes from

one-way transports. For performing this helpful service, they are charged less than most other

customers and their purchases are highly variable. Considering them as the “risk-free proxy” in

computing the reward ratio provides a useful benchmark for a logistics company. For firms that

produce maintenance and operating supplies (MRO items), a risk-free proxy might be a segment

of large distributors that desire private label products that a firm could produce to fill excess

capacity. Identifying the risk-free customer proxy for a business requires deep insight into the

T
strategic and daily operations of the business because, ideally, the risk-free customers should be

IN
strategically irrelevant and always available for the right price. For businesses where the

PR
“strategically irrelevant” customers cannot be identified, the return of the risk free proxy is zero.

For the client company, we identified a segment as a risk-free proxy that initially seemed

rather unappealing: a lower return (EBIT) than other customers (2.4% compared to 6.2%, p <

.05), without loyalty, and strategically irrelevant. Importantly, observe that the customer reward
T
ratio does not determine the absolute desirability of a customer. One should also consider the
O

impact of the customer on the overall portfolio (i.e., customer beta) as well as other strategic
N

aspects, like growth potential. However, for customers with similar impact on the portfolio and

no specific strategic consideration, the customer reward ratio provides a clear criterion for
O

choosing the most desirable customer. Appendix 2 provides a summary of the customer reward
D

ratios for key industry sectors.

Observe from Appendix 2 that most of the industries in the top 10 (if ranked using the

customer reward ratio) belong to the discretionary category. Interestingly, customers in lawn and

garden, machinery, and office supply have the highest levels of reward on risk. Upon seeing the

analysis, the client company realized that many of the customers that provide the highest reward

on risk were not receiving adequate attention. Represented here are customers that provide very
 28

attractive margins but also are characterized by very high variability. These clients request

services when they need them and, in order to receive the speed and quality of the services that

they demand, they are willing to pay a premium price.

Data analysis conclusion

Using the efficient frontier applied to customer segments, we were able to identify an optimal

composition of the customer portfolio that outperformed, in terms of variability, both the client

company’s current strategy and a profit-maximization portfolio as demonstrated by back-testing

T
and forward-testing (see Figure 4). By using a diversified, efficient portfolio, companies could

IN
reduce the vulnerability and volatility of cash flow from the customer portfolio, better insulating

PR
the firm during downturns in the economy without sacrificing performance in the long run. We

demonstrated how managers can use customer beta and the customer reward ratio to evaluate

specific individual customers and to make corresponding adjustments in the customer portfolio.
T
Discussion and Managerial Implications
O

Marketing managers face increased pressure to demonstrate the financial impact of marketing

resource allocation decisions (Rust et al. 2004). By demonstrating how financial portfolio theory
N

can be applied to customer portfolio management, our research contributes to marketing theory
O

and practice on several counts. First, we extend the work of Srivastava, Shervani, and Fahey

(1998) by demonstrating, conceptually and empirically, how the market-based assets of a firm
D

include distinct customer asset classes that are characterized by differing degrees of cash flow

variability and vulnerability. We tested whether customers can be categorized into segments that

share similarities with asset classes used in traditional financial investments. We found support

for our belief that financial portfolio theory is relevant in a customer portfolio context. In

particular, we showed that market segments—defined a priori based on classic market

segmentation variables—could be characterized in terms of their risk, as well as their return,

 29

thereby contributing to traditional market segmentation theory.

Second, we contribute to research on customer portfolio management (e.g., Johnson and

Selnes 2004; Dhar and Glazer 2003) by introducing two methods to assess the value of a

customer: customer beta and the customer reward ratio. Responding to the call of Rust et al.

(2004), our approach embraces (rather than controls for) customer heterogeneity as a path to

improved business performance. For example, we demonstrated how the customer reward ratio

can be used to examine a customer portfolio through a new lens that allows managers to isolate

T
desirable customers that receive high scores on the reward-on-variability measure. As we

IN
illustrated, the attractive customers that rise to the top on this measure often present a profile that

PR
may not be detected when using classical criteria such as average level of purchases.

Third, we present an actionable plan to guide marketing managers in creating and

managing a diversified portfolio among existing and new customers. To this end, we constructed

segments based on the variability of standardized revenue. We obtained clusters with a high
T
degree of uniformity in terms of level of revenue, size of the business, and industry. We
O

combined the clusters to form an efficient frontier that describes the portfolio with the lowest
N

variability of returns for a desired level of return. Both back-testing and forward-testing showed

that it is possible to build an efficient customer portfolio. We conclude that if companies want to
O

increase the stability of the customer cash flow, risk management techniques can be implemented
D

to ensure diversity among existing and potential customers/segments. This study demonstrates

that companies can diversify their customer portfolios by developing a thorough understanding

of customers’ purchase patterns and the drivers of these purchasing patterns (e.g., size,

preferences for product lines, and industry sector).

The Efficient Versus the Profit-maximizing Customer Portfolio

The goal of building an “efficient” customer portfolio will be different from the profit-
 30

maximizing objectives first identified in Blattberg and Deighton’s (1996) path-breaking article

and extended in subsequent research (Blattberg, Getz, and Thomas 2001; Reinartz and Kumar

2003; Reinartz, Thomas, and Kumar 2005), which focus on profit maximization for the short and

long run. There will be some similarities between the efficient customer portfolio and the profit-

maximizing customer portfolio, but there will also be differences. For example, observe from

Table 1 that cluster 5, which dominates the current portfolio and contains some of the most

profitable customers, has similar weight in the optimized portfolio (E5).

T
In contrast, in the efficient portfolio, weights for the other clusters are increased

IN
dramatically, especially for cluster 3, which is dominated by SMBs. In other words, SMBs—

PR
often not a priority for some businesses—provide a balancing element when it comes to portfolio

optimization. Cluster 1, which has relatively low profitability and high variability, is the one in

which weight has been decreased most drastically for the efficient portfolio as compared to the

company’s current portfolio, from 41% to 4%. Considering that cluster 1 is one of the clusters in
T
which customers have exhibited the most growth, decisions regarding the customers in this group
O

should be made on a case-by-case basis.

N

Implementing the Portfolio Approach

Compared to corresponding adjustments in a financial portfolio, changes in the composition of a

O

customer portfolio involve higher transaction costs, require a longer time horizon to implement,
D

and may introduce a host of strategic alignment issues to consider. Managerial judgment and the

strategic goals of the firm ultimately guide the selection of the target portfolio. Therefore, the

cluster weights that define the efficient customer portfolio provide a tentative portfolio structure

that managers can then adjust after examining the full range of customer metrics that CLV-based

methods employ. Observe from Table 1 that the portfolios along the efficient frontier vary

widely in cluster weights. While the portfolio with the lowest risk (E1) has a very balanced
 31

composition, the portfolio with the highest return (E10) gives dominant weight (92%) to cluster

5. Represented in cluster 5 are large customers, drawn from several different industries, that

provide higher average revenue and lower variability than other clusters. By isolating the risk-

return characteristics of these customers, managers can make more informed judgments for

targeting customers and estimating future returns.

The key difference between a customer portfolio and a financial portfolio is that

managers can directly influence outcomes. Past studies in the CLV research tradition provide

T
valuable insights into how profitability can be enhanced by selecting the right customers for

IN
targeting and determining the level of resources to be allocated to specific customers (e.g.,

PR
Kumar 2008; Bowman and Narayandas 2004). Also a wealth of other metrics such as customer

loyalty, share of wallet, and strength of the exchange relationship guide customer management.

Our approach can be readily incorporated within established customer management frameworks

to aid managers in balancing risk and return in a customer portfolio.

T
O

Limitations and Future Research

While our approach allows managers to examine the risk-return characteristics of a customer
N

portfolio through a new lens, this study presents some limitations that could spawn further
O

research. First, since the current study is confined to a single firm and industry, further research

is needed to test the viability of our approach in different industry contexts. The data and
D

methods used in this research are available to most companies: purchase transactions over time,

limited demographics or firmographics, and profitability by cluster. Second, the current study

centers squarely on the structure of the existing portfolio and therefore does not consider the

addition of new market segments. Buhl and Heinrich (2008) offer a heuristic method for adding

new segments to the customer portfolio that provides a promising start for future research.

Third, the customer portfolio measures used in this study center on the variability of cash
 32

flow and profitability but are insensitive to the direction of movement. Therefore, managerial

judgment, informed by established customer management approaches, is needed to discern the

root cause of the variability (i.e., growth or decline of cash flow). Future research might explore

customer resource allocation from a downside-risk perspective (Harlow 1991). Fourth, we

assessed the desirability of customers by analyzing the past volatility of purchases. However, for

estimating future customer worth, the most appropriate measure would be future volatility. In

order to determine the future volatility, Engle (1982) proposes a weighted moving average model

T
that takes into account the long-term behavior of a financial asset. By analyzing customer

IN
purchase information, a similar model of weighted moving averages could be explored in order

PR
to more accurately predict future customer cash flow variability.

Conclusion

Markowitz, in his Nobel Prize acceptance speech, mentioned that “an investor who knows the
T
future returns with certainty will invest in only one security, namely the one with the highest
O

future return” (1991). However, as Bernstein (1999, p.1) says, “[E]ven the most brilliant of

mathematical geniuses will never be able to tell us what the future holds. In the end what matters
N

is the quality of our decisions in conditions of uncertainty.” We propose an approach that

O

customer portfolio managers can follow to cope with uncertain market conditions and to improve

the quality of their resource allocation decisions. This research offers a new perspective on
D

customer portfolio management, acknowledging an aspect that has been virtually ignored: the

risk of the customer. Paraphrasing Engle’s (2003) Nobel Prize acceptance speech, we infer that

acknowledging risks should provide insight about which customers are truly worthwhile.9
 33

Footnotes

1. We thank an anonymous reviewer for suggesting this organizational scheme and focus.

2. In order to respect confidentiality agreements, numbers have been scaled.

3. S&P identifies 10 different industry sectors: Energy, Materials, Industrials, Consumer

Discretionary, Consumer Staples, Health Care, Financials, Information Technology,

Telecommunication Services, and Utilities (Source: S&P Industry Classification Standard,

[Link]/spf/pdf/index/[Link], accessed June 5, 2009).

T
4. Each customer’s revenue was standardized by dividing the monthly value by the mean

IN
revenue for the 72 months. Standardization allows clustering by using variability patterns alone,

PR
without interference from the size of the customer purchases.

5. MathWorks [Link] accessed

10/10/2010

6. Even though cluster 6 was introduced in the analysis, it had zero weight in all the efficient
T
portfolios. This cluster was characterized by low return and high variability.
O

7. The data for 2001–2005 has been used to identify the clusters, but not for the efficient frontier.
N

8. For example, for a return of 15 and a standard deviation of 10, the customer reward ratio

without the risk-free asset is 1.5 (15/10), while with a risk-free asset with a return of 3, it is 1.2
O

(12/10). For an asset with return of 28 and standard deviation of 20, the customer reward ratio
D

without the risk-free asset is 1.4 (28/20, less attractive than the first asset), but taking into

account the risk-free rate, the customer reward ratio is 1.25 (25/20, compared to 1.2 for the first

example), which is more attractive than the first asset considered.

9. “The advantage of knowing about risks is that we can change our behavior to avoid them. [...]

Optimal behavior takes risks that are worthwhile.” See Robert F. Engle III (2003), “Risk and

Volatility: Econometric Models and Financial Practice.” Nobel Lecture (Ed.). New York, p. 326.
 34

References

Acs, Zoltan J., and David B. Audretsch (1990), “The Determinants of Small-Firm Growth in

U.S. Manufacturing,” Applied Economics, 22 (February), 143-53.

Alexander, Gordon J., and William F. Sharpe (1989), Fundamentals of Investments. Englewood

Cliffs, N.J.: Prentice Hall.

Ambler, Tim, C. B. Bhattacharya, Julie Edell, Kevin Lane Keller, Katherine N. Lemon, and

Vikas Mittal (2002), “Relating Brand and Customer Perspectives on Marketing

T
Management,” Journal of Service Research, 5 (August), 13-25.

IN
Ang, Andrew, Joseph Chen, and Yuhang Xing (2006), “Downside Risk,” Review of Financial

Studies, 19 (4), 1191-1239.

PR
Balagopal, Balu, and Guy Gilliland (2005), “Integrating Value and Risk in Portfolio Strategy,”

Boston Consulting Group, Report.

Bernstein, Peter L. (1999), “Wimps and Consequences,” Journal of Portfolio Management, 26

T
(Fall), 1.
O

Blattberg, Robert C., and John Deighton (1996), “Manage Marketing by the Customer Equity
N

Test,” Harvard Business Review, 74 (July-August), 136-44.

———, Gary Getz, and Jacquelyn Thomas (2001), Customer Equity: Building and Managing
O

Relationships as Valuable Assets. Boston: Harvard Business School Press.

D

Bolton, Ruth N., Katherine N. Lemon, and Peter C. Verhoef (2008), “Expanding Business-to-

Business Customer Relationships: Modeling the Customer's Upgrade Decision,” Journal of

Marketing, 72 (January), 46-64.

Bowman, Douglas and Das Narayandas (2004), "Linking Customer Management Efforts to

Customer Profitability in Business Markets," Journal of Marketing Research, 41(4), 433-47.

Buhl, Hans Ulrich and Bernd Heinrich (2008), “Valuing Customer Portfolios under Risk-Return
 35

Aspects: A Model-based Approach and its Application in the Financial Services Industry,”

Academy of Marketing Science Review, 12 (5), 1-32.

Cardozo, Richard N. and David K. Smith, Jr. (1983), "Applying Financial Portfolio Theory to

Product Portfolio Decisions: An Empirical Study," Journal of Marketing, 47 (Spring), 110-

19.

Cardozo, Richard N. and David K. Smith, Jr. (1985), "On the Use of Financial Portfolio Theory

in Marketing Decisions: A Reply to Devinney, Stewart and Shocker," Journal of Marketing,

T
49 (Fall), 113-15.

IN
Chan, Louis K. C., Jason Karceski, and Josef Lakonishok (1999), “On Portfolio Optimization:

(January), 937-74. PR
Forecasting Covariances and Choosing the Risk Model,” The Review of Financial Studies, 12

Chandra, Siddharth (2003), “Regional Economy Size and the Growth-Instability Frontier:

Evidence from Europe,” Journal of Regional Science, 43 (1), 95-122.

T
Conroy, Michael E. (1974), “Alternative Strategies for Regional Industrial Diversification,”
O

Journal of Regional Science, 14 (1), 31-46.

N

DeMiguel, Victor, Lorenzo Garlappi, and Raman Uppal (2007), “Optimal Versus Naïve

Diversification: How Inefficient is the 1/N Portfolio Strategy?” The Review of Financial
O

Studies, 22 (5), 1916-53.

D

Devinney, Timothy M., David W. Stewart, and Allan D. Shocker (1985), "A Note on the

Application of Portfolio Theory: A Comment on Cardozo and Smith," Journal of Marketing,

49 (Fall), 107-12.

——— and ——— (1988), “Rethinking the Product Portfolio: A Generalized Investment

Model,” Management Science, 34 (September), 1080-95.

Dhar, Ravi, and Rashi Glazer (2003), “Hedging Customers,” Harvard Business Review, 81
 36

(May), 86-92.

Dickson, Peter R., and James L. Ginter (1987), “Market Segmentation, Product Differentiation,

and Marketing Strategy,” Journal of Marketing, 51 (April), 1-10.

Dissart, J. C. (2003), “Regional Economic Diversity and Regional Economic Stability: Research

Results and Agenda,” International Regional Science Review, 26 (October), 423-46.

Engle, Robert F. III (1982), “Autoregressive Conditional Heteroskedasticity with Estimates of

the Variance of U.K. Inflation,” Econometrica, 50 (July), 987-1008.

T
——— (2003), “Risk and Volatility: Econometric Models and Financial Practice.” Nobel

IN
Lecture (Ed.) New York.([Link]

PR
[Link], accessed November 11, 2009)

Fama, Eugene F. (1970), “Efficient Capital Markets: A Review of Theory and Empirical Work,”

Journal of Finance, 25 (May), 383-417.

Feldstein, Martin S. (1969), “Mean-Variance Analysis in the Theory of Liquidity Preference and
T
Portfolio Selection,” Review of Economic Studies, 36 (January), 5-12.
O

Grinblatt, Mark and Sheridan Titman (2002), Financial Markets and Corporate Strategy, 2ed.
N

Boston: McGraw-Hill Irwin.

Gundlach, Gregory T., and Patrick E. Murphy (1993), “Ethical and Legal Foundations of
O

Relational Marketing Exchanges,” Journal of Marketing, 57 (October), 35-46.

D

Gupta, Sunil, and Donald R. Lehmann (2005), Managing Customers as Investments: The

Strategic Value of Customers in the Long Run. Upper Saddle River, N.J.: Wharton School

Publishing.

———, ———, and Jennifer A Stuart (2004), “Valuing Customers,” Journal of Marketing

Research, 41 (February), 7-18.

Harlow, W. V. (1991), “Asset Allocation in a Downside-Risk Framework,” Financial Analysis

 37

Journal (September-October), 28-40.

Homburg, Christian, Viviana V. Steiner, and Dirk Totzek (2009), “Managing Dynamics in a

Customer Portfolio,” Journal of Marketing, 73 (September), 70-89.

Hubbard, Douglas W. (2009), The Failure of Risk Management. Hobokon, N.J.: John Wiley &

Sons, Inc.

Johnson, Michael D., and Fred Selnes (2004), “Customer Portfolio Management: Toward a

Dynamic Theory of Exchange Relationships,” Journal of Marketing, 68 (April), 1-17.

T
Kaplan, Paul D. (2009), “Déjà Vu All Over Again,” Morningstar Advisor, Feb/March, 28-33.

IN
Kim, Moon K., and David A. Burnie (2002), “The Firm Size Effect and the Economic Cycle,”

PR
The Journal of Financial Research, 25 (Spring), 111-24.

Kumar, V. (2008), Managing Customers for Profit: Strategies to Increase Profits and Build

Loyalty. Philadelphia, Pa.: Wharton School Publishing.

Kundisch, Dennis, Stefan Sackmann, and Markus Ruch, (2008), “Transferring Portfolio
T
Selection Theory to Customer Portfolio Management—The Case of an e-Tailer,” in
O

Enterprise Applications and Services in the Finance Industry, D. J. Veit et al. eds. Berlin
N

Heidelberg: Springer, 32-49.

Labahn, Douglas W. (1999), “Avoiding Over-Reliance on a Single Large Customer: The Impact
O

of Technical Capability, Product Development and Alternative Key Customers,” Journal of

D

Business to Business Marketing, 5 (4), 5-37.

Lande, Paul S. (1994), “Regional Industrial Structure and Economic Growth and Instability,”

Journal of Regional Science, 34 (3), 343-60.

Markowitz, Harry (1952), “Portfolio Selection,” The Journal of Finance, 7 (March), 77-91.

——— (1987), Mean-Variance Analysis in Portfolio Choice and Capital Markets. New York,

N.Y.: B. Blackwell.
 38

——— (1991), Portfolio Selection: Efficient Diversification of Investments. Malden, Mass.:

Blackwell Publishers Inc.

Mittal, Vikes, Matthew Sarkees, and Feisal Murshed (2008), “The Right Way to Manage

Unprofitable Customers,” Harvard Business Review, 86 (April), 95-102.

Niemira, Michael P., and Philip A. Klein (1994), Forecasting Financial and Economic Cycles.

New York, N.Y.: Wiley.

Niraj, Rakesh, Mahendra Gupta, and Chakravarthi Narasimhan (2001), “Customer Profitability

T
in a Supply Chain,” Journal of Marketing, 65 (July), 1-16.

IN
Pearson, Michael M., and Guy Gessner (1999), “Transactional Segmentation to Slow Customer

PR
Defections,” Marketing Management, 8 (Summer), 16-23.

Rao, Ramesh K. S. and Neeraj Bharadwaj (2008), “Marketing Initiatives, Expected Cash Flows,

and Shareholders’ Wealth,” Journal of Marketing, 72 (January), 16-26.

Reinartz, Werner J., and V. Kumar (2003), “The Impact of Customer Relationship
T
Characteristics on Profitable Lifetime Duration,” Journal of Marketing, 67 (January), 77-99.
O

———, Jacquelyn S. Thomas, and V. Kumar (2005), “Balancing Acquisition and Retention
N

Resources to Maximize Customer Profitability,” Journal of Marketing, 69 (January), 63-79.

Reinganum, Mark R. (1992), “A Revival of the Small-Firm Effect,” Journal of Portfolio

O

Management, 18 (Spring), 55-62.

D

Rust, Roland T., Tim Ambler, Gregory S. Carpenter, V. Kumar, and Rajendra K. Srivastava

(2004), “Measuring Marketing Productivity: Current Knowledge and Future Directions,”

Journal of Marketing, 68 (October), 76-89.

———, Katherine N. Lemon, and Valerie A. Zeithaml (2004), “Return on Marketing: Using

Customer Equity to Focus Marketing Strategy,” Journal of Marketing, 68 (January), 109-27.

Ryals, Lynette (2003), “Making Customers Pay: Measuring and Managing Risk and Returns,”
 39

Journal of Strategic Marketing, 11 (September), 165-75.

——— (2002), “Measuring Risk and Returns in the Customer Portfolio,” Journal of Database

Marketing, 9 (March), 219-27.

Sharpe, William F. (1966), “Mutual Fund Performance,” The Journal of Business, 39 (January),

119-38.

——— (1994), “The Sharpe Ratio,” Journal of Portfolio Management, 21 (1), 49–58.

———, Gordon J. Alexander, and Jeffery V. Bailey (1999), Investments, 6th ed. Upper Saddle

T
River, N.J.: Prentice Hall.

IN
Srinivasan, Shuba, and Dominique Hanssens (2009), “Marketing and Firm Value: Metrics,

PR
Methods, Findings and Future Directions,” Journal of Marketing Research, 46 (3), 293-312.

Srivastava, Rajendra K., Tasadduq A. Shervani, and Liam Fahey (1998), “Market-Based Assets

and Shareholder Value: A Framework for Analysis,” Journal of Marketing, 62 (January), 2-

18.
T
Tuli, Kapil R., Sundar G. Bharadwaj, and Ajay K. Kohli (2010), “Ties that Bind: The Impact of
O

Multiple Types of Ties with a Customer on Sales Growth and Sales Volatility,” Journal of
N

Marketing Research, 47 (1), 36-50.

Venkatesan, Rajkumar, and V. Kumar (2004), “A Customer Lifetime Value Framework for
O

Customer Selection and Resource Allocation Strategy,” Journal of Marketing, 68 (October),

D

106-25.
 40

Table 1

Evolution of Cluster Weights for the Portfolios on the Efficient Frontier

Return Cluster Weights Variance

Portfolio Rate X1 X2 X3 X4 X5 X6 (10-5)
E1 7.20% .20 .22 .26 .16 .16 - 13.30
E2 7.40% .14 .20 .24 .18 .24 - 13.39
E3 7.56% .10 .18 .23 .19 .31 - 13.53
E4 7.60% .09 .17 .22 .19 .32 - 13.58
E5 7.80% .04 .14 .20 .21 .41 - 13.87

T
E6 8.00% - .11 .18 .22 .49 - 14.26
E7 8.20% - .05 .16 .22 .57 - 14.78

IN
E8 8.40% - - .13 .20 .67 - 15.46
E9 8.60% - - .06 .15 .79 - 16.52
E10 8.80% - - - .08 .92 - 18.06
Current 7.56% .36 .06 .04 .04 .43 .07 31.47

PR
T
O
N
O
D
 41

Figure 1
Industry Trends
1,200

Big Retail Paper and Packaging

1,000
Automotive Consumer Goods
Food & Beverage Transportation Manager
Home Improvement Electronics & Appliances
Chemicals Small Retail

800
Revenue (Million $)

T
600

IN
400

200
PR
-
2001 2002 2003 2004 2005 2006
T
YEAR
O
N
O
D
Figure 2
Revenue by Cluster
200.0

Mill. $ Cluster 1: Constant growth (Return = .066)

T
Cluster 2: Rise and decline (Return = .062)
180.0

IN
Cluster 3: Moderate growth followed by decline (Return = .074)

160.0 Cluster 4: Constant decline (Return = .074)

Cluster 5: Best customers slowing down (Return = .089)

PR
140.0 Cluster 6: Low revenue customers (Return = .053)

120.0

T
100.0

80.0 O
N
60.0

40.0
O

20.0
D

Month
-
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 69 71

42
 43

Figure 3
The Efficient Frontier Portfolios and Current Portfolio Risk and Return

10.00%

T
9.50%

IN
9.00%
E9 E10
E8
8.50%
E7

PR
E6
8.00% E5 Current Portfolio
Return (%)

E3, E4 E3 is an efficient portfolio that offers

7.50% E2 identical return with the current portfolio,
7.00%
E1 but has minimum risk (variance)

6.50%

T
6.00%

5.50% O
N
5.00%
12.00 14.00 16.00 18.00 20.00 22.00 24.00 26.00 28.00 30.00 32.00
Variance (*10-5)

Note: Portfolio 3 offers identical return with the current portfolio, for less then half the variance (43%).
O
D
 44

Figure 4:
Back-testing and Forward-testing the Efficient Portfolio (Simulation Results)
A: Profit Variability by Portfolio B: Profit by Portfolio (Million $)

T
0.160 370.0
Current Portfolio Current Portfolio
0.140
Efficient Portfolio 320.0

IN
Efficient Portfolio
0.120 Profit Maximization Portfolio Profit Maximization Portfolio
270.0
0.100

0.080 220.0

PR
0.060
170.0
0.040
120.0
0.020

- 70.0
2001 2002 2003 2004 2005 2006* 2007 2001 2002 2003 2004 2005 2006* 2007

T
C: Variability of the Current Customer Portfolio and D: Profitability of the Current Customer Portfolio
and the Efficient Portfolios Identified Using Annual
0.12
the Efficient Portfolios Identified Using Annual Data
between 2001 and 2006 O 300

280
Data between 2001 and 2006
N
0.1
260

0.08 240

Current portfolio 2001 efficient portfolio 220

Current portfolio
`
0.06 200
2001 efficient portfolio
2002 efficient portfolio 2003 efficient portfolio
O

2002 efficient portfolio

180
0.04 2005 efficient portfolio 2006 efficient portfolio 2003 efficient portfolio
160
2005 efficient portfolio
0.02 140
D

2006 efficient portfolio

120
0
100
2001 2002 2003 2004 2005 2006 2007 2001 2002 2003 2004 2005 2006 2007
 45

Figure 5:
Absolute Variance Reduction in Efficient Portfolios versus the Current Portfolio Results

5.00E-04

T
4.50E-04

IN
4.00E-04 Portfloio Variance by Year
3.50E-04
3.00E-04 Actual Portfolio Variance
Efficient Portfolio Variance

PR
2.50E-04
2.00E-04
1.50E-04
1.00E-04
5.00E-05

T
0.00E+00
2001 2002 2003 2004* 2005 2006

O
N
O
D
 Appendix 1

Cash Flow Variability by Customer Type

Contractual relationships. Customer-firm relationships range from formal contractual to

transactional relationships. Transactional relationships are low involvement, occurring on

an as-needed basis. Contractual relationships are governed by rules that are mutually

agreed on by the contracting parties (Gundlach and Murphy 1993). Due to their explicit

T
nature, contractual relationships are more predictable than transactional ones and

IN
typically yield cash flows with less variability. Hence, when a high proportion of

customers have entered into contractual agreements with the firm, especially long-term

PR
contracts, the overall risk of the firm’s customer portfolio will be low.

We compare the coefficients of variation for customer purchases over time

between contractual and noncontractual relationships for the cooperating company in the
T
following way. The client company offers four different product lines (similar services,
O

but with different delivery characteristics). Product line #1 relies on contractual

relationships, whereby assets are allocated to a specific customer, thus restricting the
N

firm’s flexibility to deploy these assets elsewhere. All other product lines have higher

flexibility and no contracts attached.

O

The coefficient of variation (CoV) for product line #1, CoVLine 1=.525, is
D

significantly lower (p = .000) than the coefficient of variation for any other product line

(CoVLine2=.879; CoVLine3=1.216; CoVLine4=1.740; CoVAll Lines=.708). In other words, the

contractual product line (#1) has the smoothest, most predictable cash flows, thereby

insulating the firm from troughs (downtimes) and peaks (busy times). By serving

customers that prefer a contractual relationship, the firm reduces the coefficient of

46
 47

variation for its overall customer portfolio.

Size of business. Research regarding financial portfolios has shown that small firms tend

to periodically outperform and underperform large firms, exhibiting a negative

correlation in returns (Reinganum 1992). Small firms outperform large firms during

economic booms, but the effect disappears during recessions (Kim and Burnie 2002).

Large business customers, characterized by financial soundness and greater volume,

T
represent the equivalent of the blue-chip stocks in a financial portfolio—i.e., stocks of

IN
companies with steady earnings and a solid reputation, but slower growth. In contrast,

small and medium-size businesses (SMBs) have high growth potential (Acs and

PR
Audretsch 1990). If large businesses dominate a financial portfolio, variations in their

business cycles will have a substantial impact on their suppliers (LaBahn 1999). SMBs

usually have less influence on the overall financial portfolio individually, but they can be
T
combined to achieve diversification and lower overall variability, provided that their
O

revenue streams are not positively correlated (Markowitz 1987).

We compare the coefficients of variation for customer purchases over time

N

between companies of different sizes using a median split based on the number of

employees. The coefficient of variation for the small companies was .67, statistically
O

different (p = .023) from the value of .77 for large companies.

D

Industry Classification. Industries are affected differently by external economic events.

For example, a downturn in the economy is often accompanied by a decrease in home

construction and an increase in home improvement projects. A price increase for a

commodity, such as silicone, might result in a substantial price increase for automobile

tires, whereas the price of personal grooming products (such as shampoos and liquid
 48

soaps) might increase very little because silicone is not a relevant component. Dhar and

Glazer (2003) show that targeting customers in different segments reduces the risk of a

revenue decline when economic conditions are changing.

We classified customers into major NAICS categories (e.g., transportation, paper

and packaging, automotive) combined with the S&P global industry classification

categories, which are designed to capture sector differences. By combining classification

T
schemes, we obtain a finer granularity that allows for more uniformity within the

IN
identified categories. For example, NAICS identifies retailers, while the S&P standard

makes the distinction between discretionary (e.g. Kohl’s) and staples retailers (Wal-

PR
Mart), which are likely to respond differently to peaks and troughs in the economy.

Customer reward ratios, betas, and coefficients of variation for customer purchases over

time in different industry sectors are presented in Appendix 2.

T
Because there are only a small number of companies in each category (revenue
O

from large discretionary retail companies [line 11 in Appendix 2] is provided by seven

customers), there is insufficient statistical power for t-tests of the differences in the
N

average coefficients of variation across categories. However, when we compare, for

example, the office supply segment (line 9), with the food and beverage segment (line
O

14), we notice important differences. Office supply customers are more attractive
D

compared with the food segment: negative beta signals negative correlation with the

overall portfolio, the customer reward ratio is much higher, and the coefficient of

variation is much smaller (i.e., lower risk). However, the cooperating company has many

more customers in the food sector (42) than in the office supply sector (3), and therefore

the t-tests were inconclusive.

 49

Appendix 2
Coefficient of Variation Classified Using Both NAICS and S&P

Average Six Year

Coef.
Reward Monthly Revenue
S&P Industry N Beta of Var.
Ratio Revenue **
(CoV)
(thou.) (mill.)
1 Beverage* Food & Beverage 14 1.199 1.344 .710 369 328
2 Discretionary Apparel 9 1.261 2.467 .802 245 128
3 Discretionary Automotive 35 1.616 1.400 .604 1,000 2,381

T
4 Discretionary Consumer Goods 4 .883 .609 .720 599 170
5 Discretionary Durables 10 1.874 .181 .656 182 110
6 Discretionary Electronics & Appl. 18 1.533 2.398 .786 368 361

IN
7 Discretionary Home Improvement 27 .067 1.169 .705 313 559
8 Discretionary Lawn and Garden 4 .336 8.521 1.145 167 48
9 Discretionary Office Supply 3 -3.668 2.948 .461 798 162
10
11
12
13
14
Discretionary
Discretionary
Discretionary
Energy
Food
Paper & Packaging
Retail
Sporting Goods
Oil
Food & Beverage
PR
11 1.584
7 1.125
7 -.233
5 1.067
42 8.568
1.976
1.673
.289
2.050
.774
.592
.675
.751
.680
.794
659
5,042
123
355
352
311
2,476
46
64
973
15 Health Medical Supplies 6 .773 .111 .644 138 55
16 Industrials Automotive 8 .909 1.001 .598 237 98
T
17 Industrials Electronics & Appl. 5 1.856 .878 .774 152 47
18 Industrials Machinery 4 .107 3.120 .567 656 181
O

19 Industrials Paper & Packaging 9 3.871 .315 .693 347 223

20 Industrials Transportation 10 -18.483 .557 1.212 243 129
21 Materials Chemicals 18 .834 1.213 .668 274 308
N

22 Materials Metal Manufact. 12 1.569 1.266 .773 87 62

23 Materials Paper & Packaging 30 1.516 .850 .622 770 1,576
24 Materials Wood Manufact. 4 .948 .854 .474 142 33
25 Staples Consumer Goods 18 -3.780 1.058 .792 1,763 2,048
O

26 Staples Paper & Packaging 17 .742 .891 .714 1,687 2,036

27 Staples Pet Supplies 5 16.118 1.789 .631 440 88
28 Staples Retail 5 .366 .674 .629 7,544 2,683
D

29 Transportation Transportation 55 .035 2.498 .702 244 743

*The distinction between food and beverage, though not in the S&P standards, is useful for
distinguishing the different patterns that are likely to characterize foods (e.g., cereals) from
beverages (e.g., beer, soda).
**Cumulative revenue for the years 2001-2006.
 50

Appendix 3
Comparisons among Clusters

Cluster Characteristics Industry Dominance

Cluster 1 (Constant Growth): 84 customers and 24% of the 6-year revenue

Higher beta (more rapid growth) than all other clusters 89% of the discretionary retailers
Higher variability (risk measured using coefficient of (and 53% of all discretionary
variation) than clusters 2, 3, 5, 6 products)
Higher absolute and average revenue per customer 56% of material paper and
than clusters 3 and 6* packaging (and 48% of all materials)
Larger business customers (higher number of 55% of all health products

T
employees) than cluster 2 39% of home improvement

IN
Cluster 2 (Rise and Decline): 74 customers, 12% of the 6-year revenue
Lower beta than clusters 1, 3, 5, and 6, but higher than Discretionary electronics and
cluster 4 appliances (61%) and discretionary
Lower variability (covariance) than clusters 1, 4 consumer goods (42%)

PR
Smaller-size customers (by number of employees)
than cluster 1, but larger than cluster 3
Customers in cluster 2 buy overall and on average
more than the customers in clusters 3 and 6, but
relatively less than the customers in cluster 5*
Manufacturers of metal (47%) and
wood (39%)
Food and beverages (35%)
T
Cluster 3 (Consistency Followed by Decline): 52 customers and 3% of 6-year revenue
Lower revenue per customer than clusters 1, 2, 4* and Health (35%, while 55% is in cluster
O

5, but higher than cluster 6 1)

Lower variability (CoV) than clusters 1 and 4 Energy (30%, while 66% is in
Smaller-size customers (by number of employees) cluster 5)
N

than clusters 2, 4, 5, 6 Machinery (15%, while 80% is in

Average level of beta (more explicitly, lower beta than cluster 5)
clusters 1, and 5, but higher beta than clusters 2 and 4)
O

Cluster 4 (Constant decline): 71 customers and 9% of the 6-year revenue

Lower average revenue and overall revenue than Industrial electronics (60%)
cluster 5, but higher overall revenue than clusters 3* Discretionary automotives (51%)
D

and 6 Durables (45%)

Higher variability (CoV) than clusters 2, and 3, but Beverages (47%)
lower than clusters 5 and 6
The lowest beta among all clusters in terms of
revenue, but highest beta among all clusters in terms
of return
 51

Appendix 3 (continued)
Cluster Characteristics Industry Dominance

Cluster 5 (Best Customers Slowing Down): 61 customers and 47% of 6-year revenue
Higher average revenue and overall revenue than Staples in general (77%)
clusters 2*, 3, 4 and 6 Automotive - staples (74%)
Lower variability (CoV) than clusters 1, 4 and 6* Paper and packaging-staples (92%)
Lower beta than cluster 1, but higher beta than clusters Consumer goods-staples (86%)
2, 3, 4 and 6 Energy (66%)
Larger-size customers than cluster 3 (by number of Industrials (43%)
employees) and cluster 6* (by annual sales). Chemicals (industrial and materials:
36%)

T
Machinery (80%)
Lawn and garden (45%)

IN
Cluster 6 (Low Revenue Customers): 125 customers and 5% of the 6-year revenue
Lower revenue per customer than any other cluster This cluster has (statistically) as
Lower variability (CoV) than clusters 1 and 4, but large of a percentage of the
higher than cluster 5*

PR
Lower beta than 1 and 5, but higher beta than clusters
2 and 4 in terms of revenue, but lowest beta among all
clusters in terms of return
Larger company sizes (based on the number of
transportation industry as cluster 5
(about 23%), and second highest
revenue from the lawn and garden
industry (32%, while 45% is in
cluster 5)
employees) than cluster 3, but smaller customer
T
business size (based on annual sales) than cluster 5*
Customers in this cluster buy significantly less than
the average customer (over 25% of customers account
O

for 5% of the 6-year revenue), and do not have the

absolute majority for any of the categories
N

*The difference is significant at 90% confidence level. All other comparisons are statistically
significant at 95% confidence level or higher.
O
D