What to do with Unprofitable Customers? Customer Lifetime Value, Customer Metrics of Adverse Behavior, and Feasible Strategies for Managing Unprofitable Customers

Table of Contents

List of Figures

List of Abbreviations

1 Introduction

2 Customer lifetime value and its measurement models
2.1 Different perspectives on CLV
2.2 Measurements models of CLV in contractual context
2.2.1 Basic structural model
2.2.2 Recency, frequency and monetary model
2.3 Measurement models of CLV in noncontractual context
2.3.1 Pareto/NBD model and its variations
2.3.2 Markov chain model

3 Customer metrics of adverse behavior
3.1 Share of wallet
3.2 Demand for customer service.
3.3 Partial churn
3.4 Deal-proneness

4 Analysis of feasible strategies for managing unprofitable customers and their linkage to customer metrics and CLV
4.1 Retention strategies
4.2 Explanatory framework for retaining unprofitable customers
4.3 Abandonment strategies
4.4 Explanatory framework for abandoning unprofitable customer

5 Conclusion.

References

List of Figures

Figure 1: Linkage of customer metrics and strategies

List of Abbreviations

illustration not visible in this excerpt

1 Introduction

An increasing value of the customer base contributes more and more to the overall value of the firm (Wiesel, Skiera, and Villanueva 2008, p. 4). Thus, valuing customers or their behavior, respectively, has become an indispensable issue for any commercial activity. When determining causes and reasons of the customers’ contribution to firm value or performance, the customer base usually is analyzed and evaluated, whereas profitable and unprofitable customers are identified. Especially the subject of unprofitable customers, the methods to single them out and their input on the firm’s financial performance were thoroughly discussed in the literature (Abe 2009; Fader, Hardie, and Shang 2009; Haenlein, Kaplan, and Schoder 2006; Holm, Kumar, and Rohde 2012). Unprofitable customers have a strong negative effect on the profitability of the firm, e.g. 10%-35% of the clients, who cross-buy and are unprofitable, cause 39%-88% of total loss of the company from its customers. (Shah et al. 2012, p. 79).

It has been argued that scarce marketing resources should be allocated according to the ranking of the customers to reduce costs and increase revenues (Haenlein, Kaplan, and Schoder 2006, p. 5). Because regular financial metrics have restricted diagnostic potential, relying on customer metrics appears more suitable for determining customer’s profitability (Gupta et al. 2006, p. 140). There are diverse methods for evaluating customers, such as previous period customer revenue, past customer value, customer lifetime duration and customer lifetime value (CLV) (Venkatesan and Kumar 2004, p. 119). CLV examines customer profitability from a prospective perspective, foreseeing future customer behavior and discounting future cash flows (Holm, Kumar and Rohde 2012, p. 388). CLV and its measurement models, depending on the kind of customers and products obtained by the company, provide a basis for strategic and tactical decisions (Jain and Singh 2002, p. 36). Customer’s persistent adverse behavior can lead to unprofitable outcome and should be considered by determining profitability on the base of CLV. (Shah et al. 2012, p. 80). There are several strategies for handling unprofitable customers. Before applying one of these, it is necessary to measure potential benefits and losses, as the chosen strategy can have a long-run effect on the firm’s clientele.

There are some interconnections between various CLV measurement models, other customer metrics and strategies applied to unprofitable customers. In this paper, I analyze how customer metrics like CLV are linked to strategies for managing unprofitable customers.

This paper is organized as follows. In chapter 2 different perspectives on CLV and its different measurement models are described in contractual as well as in noncontractual context. In the third chapter, the customer metrics of adverse behavior, such as demand for customer service, partial churn, share of wallet and deal-proneness are characterized, regarding their contribution for determining customer unprofitability. In the final part strategies of retention and abandonment are contrasted and the interconnections between different customer metrics are examined to link them to the successful strategies, which should be applied for managing unprofitable customers.

2 Customer lifetime value and its measurement models

2.1 Different perspectives on CLV

The concept of customer lifetime value is fundamental for this research. Essentially CLV is defined with the help of net present value (Gupta at al. 2006). The definitions of CLV differ across the literature, although the expected cash flows, discounted to the present value, are the common components (Dwyer 1997, p. 7; Kumar, Lemon, and Parasuraman 2006, p. 88).

Due to varying definitions, differences in the CLV definitions should be considered. Gupta defines CLV as “present value of all future profits obtained from the customer over his or her life of relationship with a firm” (Gupta et al. 2006, p. 141). This approach allows calculating CLV of any specific customer and distinguishing between different ranks of profitability, furthermore it considers the prospect of customer switching to a competitor, but it ignores the corresponding switching costs (Gupta et al. 2006, p. 145).

Principally, costs are important constituents for building measurement models for CLV. Therefore, Dwyer’s definition of CLV as “the present value of the expected benefits (e.g., gross margin) less the burdens (e.g., direct costs of servicing and communicating) from customers” is chosen for this paper (Dwyer 1997, p. 7). This definition does not consider the costs of customer acquisition from the past, but regards both direct monetary profit and indirect benefits from the positive Word-of-Mouth (WoM) (Jain and Singh 2013, p. 4). Based on Dwyer’s definition of CLV, I describe different measurement models of CLV in the next section.

2.2 Measurements models of CLV in contractual context

2.2.1 Basic structural model

Depending on the context, distinctive data about customers is available. A basic structural model is used in the contractual settings. In the contractual context, the expected revenues can be predicted rather precisely, since customers use services regularly and it is assumed that respective cumulative profits increase over the customers’ lifetime (Reinartz and Kumar 2000, p. 17). In the contractual settings, the research focus lies on the forecasting whether and for how long customers stay with the firm (Venkatesan and Kumar 2004, p. 108). Customers, which are satisfied with the services, stay longer with the company than those, which are dissatisfied (Bolton 1998, p. 59).

In the basic structural model, CLV is determined by deducting specific costs from the revenues and discounting it. These costs are promotional cost and cost of sales, which incorporates cost of goods sold and costs of handling the purchase order (Berger and Nasr 1998, p. 20).

The basic structural model of CLV is based on the concept of net present value with the assumption that the cash flows are generated at the end of the time period (Jain and Singh 2002, p. 38). CLV differentiates among customers and acknowledges the possibility of a customer terminating the relationship (Gupta and Zeithaml 2006, p. 724). In this model CLV is a function of anticipated contribution margin, intention of the customer to remain in the relationship with the firm and marketing resources allocated to the customer (here and in the following Venkatesan and Kumar 2004, p. 108). Purchase frequency shows the propensity of the customer towards keeping the relationship with the firm or defecting it. It is assumed that customers tend to purchase with decreasing frequency before relationship dissolution.

The basic structural model considers only actual customers and excludes former and potential customers as well as their acquisition costs. Therefore, it might fall short when searching for a comprehensive explanation of the underlying customer behavior (Jain and Singh 2002, p. 38).

2.2.2 Recency, frequency and monetary model

The recency, frequency and monetary (RFM) model is based on the past customers’ purchase behavior (Gupta et al. 2006, p. 142). The fundamental terms for describing the customers’ former behavior are recency (time of the latest purchase), frequency (number of past purchases) and monetary value (average purchase quantity per transaction) (Fader, Hardie, and Lee 2005a, p. 415). The model generally rates and groups customers according to each of these variables and assigns a score for each group, to adjust marketing communication programs respectively (Zhang, Bradlow, and Small 2015, p. 195). RFM is broadly used to predict future customer behavior, thus to value customers’ future number of purchases, lifetime duration or CLV (Zhang, Bradlow, and Small 2015, p. 195). Marketing instruments, such as different modes of communication across various channels, relationship benefits of upgrades and loyalty programs, affect purchase frequency as well as contribution margin, which both have an impact on CLV (Venkatesan and Kumar 2004, p. 121). The amount of marketing contacts has a nonlinear relationship with the purchase frequency, although a too intense contact can lead to adverse consequences (Venkatesan and Kumar 2004, p. 110).

The RFM has some significant problems. The scoring model makes assumptions for the next period, but not further into the future (Fader, Hardie and Lee 2005a, p. 416). Data from at least two periods is required to make the prediction about future customer behavior using regression, therefore as a preliminary result it can be stated that RFM variables are rather inexact indicators for the underlying behavior (Fader, Hardie, and Lee 2005a, p. 416). The RFM model overlooks the possibility that prior behavior could have been influenced by the firm’s prior marketing activities (Gupta et al. 2006, p. 142).

Iso-CLV curves help to integrate the RFM model in basic CLV model. They represent different values of recency, frequency and monetary value with the same CLV (Gupta et al. 2006, p. 142). In this integrated model the distribution of monetary values is assumed to be independent of the purchasing process (Fader, Hardie, and Lee 2005a, p. 419). The advantage of the model is including well-known behavioral assumptions, rather than considering only prior purchase data (Fader, Hardie, and Lee 2005a, p. 426). Bayes’ theorem for evaluating person’s latent traits helps to determine a function of these traits and to foresee customer’s future behavior (Fader, Hardie, and Lee 2005a, p. 416).

2.3 Measurement models of CLV in noncontractual context

2.3.1 Pareto/NBD model and its variations

The models, which will be described below, are relevant for the noncontractual settings. Noncontractual relationships are relationships between a firm and a customer without contract or membership (Reinartz and Kumar 2003, p. 78). In noncontractual settings the company should secure the relationship with the customer to stay alive, as the customer has no or very low switching costs and, assumedly, the same product can be bought from different firms (Reinartz and Kumar 2000, p. 17). It is difficult to comprehend which customer has already defected from the company and which one will buy again after a long pause (Fader, Hardie, and Shang 2010, p. 1086). Compared to contractual settings, where the duration of a customer’s relationship with the firm is the key driver of CLV, revenues and contribution margins provide the basis for determining CLV here (Reinartz and Kumar 2000, p. 32).

The Pareto/NBD (negative binomial distribution) model is the combined purchase transaction and duration model for determining the probability of a customer being still active (Schmittlein, Morrison, and Colombo 1987, p. 3). The point in time, when a customer becomes inactive and will not make a further purchase, is modelled with the help of a Pareto (exponential-gamma mixture) timing model and repeat-buying behavior is analyzed applying the NBD (Poisson-gamma mixture) counting model. Customer lifetime, interpurchase times and purchase amounts are assumed to be independent drivers in the model. Furthermore, it is assumed that the customer can defect at any time (Singh, Borle, and Dipak 2009, p. 183). The model can be applied to obtain the information about the extension of the customer base, as well as estimate the amount of transactions in the future, and it is very effective for companies with few long-term customers (Jain and Singh 2002, p. 40).

More convenient for calculations is the variation of the Pareto/NBD called beta-geometric/NBD (BG/NBD). The assumptions about the interpurchase time are identical with those in Pareto/NBD model, but the number of purchases follows a Beta-Geometric distribution (Singh, Borle, and Dipak 2009, p. 183). The difference from the former model is the point in time of the customer’s dropout, that follows directly after a purchase and cannot happen independent to the actual purchase at any time (Fader, Hardie, and Lee 2005b). These models are often used as they predict future customer purchase behavior using customer-level recency and frequency information, which is easy for firms to access, but the assumptions of the model are quite restrictive (Singh, Borle, and Dipak 2009, p. 183).

Another variation of the basic Pareto/NBD model is the model including customer satisfaction in the calculation of CLV (here and in the following Ho, Park, and Zhou 2006, p. 261). Satisfied customers purchase more often, spend more during one transaction and are less likely to defect to the competitors. The potential benefits can be calculated and compared with the significant costs of improving customer service and therefore increasing customer satisfaction.

2.3.2 Markov chain model

The Markov chain model (MCM) is very flexible (Pfeifer and Carraway 2000, p. 44). It is suitable for describing retention situation, where a nonresponse of the customer indicates the end of the relationship, as well as migration situation, where the customer is believed to be active even after a long period without making any purchase (Dwyer 1997, p. 9). The MCM is a probabilistic model, which considers the uncertainty in the relationship with individual customers, thus the forecast for the future relationship with the customer is the function of the present state of the relationship, it is called the Markov property (here and in the following Pfeifer and Carraway 2000, p. 45). If the models show the Markov property with constant probabilities, they can be represented as Markov chains. Recency of purchase is incorporated into the model and determines the probability of prospective relationship.

To evaluate the relationship economically, the expected present value of the cash flows determined by recency and therefore different probable states is used, which corresponds to CLV (Pfeifer and Carraway 2000, p. 46). The MCM is easily modified for estimating changes in the policy towards customers, such as ceasing remarketing earlier, if a customer has negative expected present value (Pfeifer and Carraway 2000, p. 47). In more complicated modifications of the MCM, remarketing expenditures and expected net contribution additionally to purchase probability are dependent on the recency (Pfeifer and Carraway 2000, p. 48).

The CLV without considering customer’s behavior, especially the adverse habits, can misrepresent the real value of the customer for a firm and overestimate customers who are unprofitable and cause losses. Therefore, in the following chapter adverse behavior patterns will be considered.

3 Customer metrics of adverse behavior

3.1 Share of wallet

Customers with adverse behavior are inclined to persist with the same pattern in their purchase behavior in the future and therefore are responsible for unprofitable performance (Shah et al. 2012, p. 80). The share of wallet of a customer for a company is a behavioral customer metric, that shows the part of the entire requirements (wallet size) in all the product categories of the company (Du, Kamakura, and Mela 2007, p. 96). Different levels of share of wallet correspond to distinct levels of profitability as well as to changes in profitability over time (Larivière 2008, p. 13). Share of wallet partly defines the difference between loss-making and profitable customers together with age and household income (Garland 2004, p. 265).

Personal retail banking features young unprofitable customers with the intention to increase their share of wallet i.e. demand for more expensive offers and retaining them as a long-term customers due to the high perceived switching costs (Garland 2004, p. 261).

It is difficult for companies to recognize unprofitable customers with big growth potential, because a company relies only on the information about internal transactions and does not take into account the data of transactions with competitors, so the real size of wallet and share of wallet are unknown (Du, Kamakura, and Mela 2007, p. 97). To determine customers with large total market potential and small share, considering both internal and external databases, a survey and the list augmentation approach can be used (Du, Kamakura, and Mela 2007, p. 109). This approach links data from surveys or secondary sources to actual databases in order to develop predictive models for individual-level estimates (Du, Kamakura and Mela 2007, p. 94).

3.2 Demand for customer service.

Customer service costs are increasing and can define customer with positive CLV as unprofitable (Garland 2004, p. 259). Customers perceive the prepaid of the total cost for services, which is fixed, as more fair comparing to the total cost, paid according to the usage level (Bolton and Lemon 1999, p. 175). In the BSM, only direct marketing costs are considered regardless of all other selling, general and administrative expense. That leads to the following assumptions: firstly, that service capacity is fixed and cannot be modified according to different levels of demand for service; secondly, that service resource requirements are homogeneous among the customers (here and in the following Holm, Kumar, and Rohde 2012, p. 389). It means that the CLV of the customer, who claims excessive service capacity (e.g. because of numerous sales visits, constant time consuming customer service calls, repeated small-size deliveries to remote location) will be overrated, the customer profitability will be distorted and decision making affected. How strongly this affects a customer ranking will depend on the variety of customer service demands and their spread besides adaptability of service capacity resources for predicting future customer service requirements.

The loyal customers, gained through the loyalty programs, are less likely to have an excessive demand for customer service, hence the servicing costs are reduced. (Dowling and Uncles 1997, p. 72). It is necessary to take into account costs of excessive demand for customer services to define unprofitable customer.

3.3 Partial churn

Partial churn stands for cancelation of the service or the product purchase (Larivière 2008, p. 7). A customer’s product return is directly influenced by customer buying behavior and indirectly by marketing communications profits (Petersen and Kumar 2009, p. 37). The amount of product returns depends on the time of purchasing - during holiday season more products are likely to be returned, except for the gifts, hence they have added value besides its practical utility profits (Petersen and Kumar 2009, p. 46). Purchasing products on sale makes customers underestimate the value of the product, so product returns due to negative postpurchase utility are less likely (Petersen and Kumar 2009, p. 46).

A customer unsatisfied with some aspects of the product, such as quality, fit or efficiency, returns these products and the firm has to find out the reasons for dissatisfaction in order to eliminate them, to change customer behavior and to make the relationship profitable (Reinartz and Kumar 2003, p. 82). It is particularly relevant for firms using direct marketing, i.e. introducing the products to customers through an impersonal medium, hence the customers can only generally evaluate the product in the catalog before they actually get it (Hess and Mayhew 1997, p. 21; Reinartz and Kumar 2003, p. 82). If customers return products within a certain period, they hope for future benefits, such as better quality of the products, because they are assumed to have confidence in the company. On the other hand, too many returns are the indicators for the trust violation or decrease in the future activity of the customer (Venkatesan and Kumar 2004, p. 121).

It is known that clients who purchase much, return proportionally more merchandise, probably due to their familiarity with the procedure and ability to do it efficiently (here and in the following Reinartz and Kumar 2003, p. 91). Sometimes customers can see the return process as a part of the purchasing process, especially buying from the catalogs or online. If this is a leading motivation for returns, there is a great chance for developing positive relationship with the customer.

Product returns are an important component of the firm-customer-exchange-process. It nevertheless causes high costs in lost sales and reverse logistics and diminishes profits (Petersen and Kumar 2009, p. 35).

3.4 Deal-proneness

For this research, monetary promotions, e.g. price reductions, coupons and rebates, are more relevant than non-monetary (Yi and Yoo 2011, p. 883). For sales promotion, the firms grant deep discounts for their products therefore increasing the perceived value of the product for the customers and attracting deal-prone clientele (Webster 1965, p. 186). Monetary promotions can have a negative impact on the customer’s reference price and weaken brand quality and image (Yi and Yoo 2011, p. 884). After buying at promotional price, customers are likely to reduce the number of repeat purchases because of the forward buying or general low product valuation (Anderson and Simester 2004, p. 4). Products with a deep discount lure customers, who, otherwise, are not likely to purchase the product at a nonpromotional price (Neslin and Shoemaker 1989, p. 206). Deal-prone customers choose promoted items regularly due to their psychological disposition and additional emotional benefits from purchasing on deal. (DelVecchio 2005, p. 377). Both relative and absolute value of a promotion have an impact on the purchase choice of the merchandise (DelVecchio 2005, p. 389).

So called “bad” customers, who purchase solely at deep discounts are often “price butterflies” - they are switching from firm to firm looking for the best deal i.e. the biggest discount (Reichheld and Schefter 2000, p. 110). The deal-prone customers are likely to purchase various brands and commit only small share of their wallet to one and the same firm (Webster 1965, p. 189). Such customers desert the company before it regains the acquisition and additional up-front cost invested at the beginning of the relationship (Cao and Gruca 2005, p. 219). Adverse selection of customers is especially relevant for insurance company and other firms trading in risk products, by avoiding unprofitable customers at the phase of acquisition they can save a lot of money (Cao and Gruca 2005).

After describing the patterns of adverse behavior and the different models for measuring CLV, which are used for determining unprofitable customers, the strategies applicable to them are described and analyzed in the next chapter.

4 Analysis of feasible strategies for managing unprofitable customers and their linkage to customer metrics and CLV

4.1 Retention strategies

The retention of a current customer is cheaper than the acquisition of a new one, and more efficient than gaining market share or reducing costs. That is why, firms prefer retention of unprofitable customers to their abandonment (Zeithaml 2000, p. 76). The acquisition price for customers with the high long-term value is low, the retention price is affected by the cost of retention, retention price sensitivity of the customer and retention marketing investments (Jain and Singh 2002, p. 43). Companies do not want to suffer losses from the futile relationship with an unprofitable customer. They want to turn the customer into profitable and loyal, hence the loyalty of an unprofitable customer is not valuable (Jain and Singh 2002, p. 35). To proceed with this task, the customer lifetime value and the customer behavior should be analyzed thoroughly and the future development of the relationship between the firm and the customer estimated. The firm’s strategies are directed at increasing CLV and reducing adverse behavior of the customer.

[...]