**Publication: Academy of Marketing Studies Journal***Author: Asllani, Arben; Halstead, Diane*

*Date published: July 1, 2011*

*Language: English*

*PMID: 78756*

*ISSN: 10956298*

*Journal code: DMSJ*

(ProQuest: ... denotes formulae omitted.)

INTRODUCTION

An organization's long-term viability requires a focus on the profitability of each customer within that organization (Forbes, 2007). Customer lifetime value (CLV), the net present value of cash flows expected during a customer's tenure with a firm, can therefore be a valuable marketing metric to evaluate (Blattberg, Malthouse, & Neslin, 2009; Pfeifer & Carraway, 2000; Venkatesan, Kumar, & Bohling 2007). CLV is also often used as the basis of customer relationship management (CRM) decisions, including service level delivery (Zeithaml, Bitner, & Gremler, 2009; Zeithaml, Rust, & Lemon, 2001; Jackson, 2007). For example, customer profitability might be used to determine whether a service policy exception is made for a key account or whether a credit card customer's credit limit or interest rates are increased (Aeron, Bhaskar, Sundararajan, Kumar, & Moorthy, 2008). At its core, CLV guides a firm's acquisition and retention strategies (Blattberg et al., 2009).

Estimating CLV accurately can be difficult, however, and is sometimes beyond the ability of many firms (Stahl, Matzler, & Hinterhuber, 2003; Vogel, Evanschitzky, & Ramaseshan, 2008). Even the predictions of the winning model from a recent CLV modeling competition were inaccurate by more than 500 percent, for example (Blattberg et al., 2009). An approach is needed, therefore, that allows fairly simple prediction of a customer's long-term profitability potential while simultaneously providing marketers with effective CRM decision input.

One variant of the CLV estimation models is the RFM framework used in direct marketing in which the probability of customers' future purchases is based on the recency (R), frequency (F), and monetary value (M) of their previous transactions. These RFM probabilities are then used to categorize customers according to their profit potential. Customers with the highest profit potential would then be the possible targets of a company's direct marketing campaign. The RFM approach offers a potential solution to the problems associated with predicting CLV and gives direct marketers input on customer profitability and relationship management issues.

The research presented here offers a linear programming (LP) approach that combines data provided by RFM analysis alongside budgeting data for a given campaign. The model can help direct marketers determine whether to continue or curtail their relationship with a given RFM customer segment. A novel characteristic of this model is the budget constraints. Theoretically, when a company has an unlimited marketing budget, managers can afford to reach all their customers, even those who have low RFM scores. This approach would minimize Type I error, which occurs when a company does not contact a customer who could have potentially provided additional revenue and profits. Such a strategy is clearly not practical, however, because organizations typically operate under annual marketing budget constraints.

Such a strategy maximizes Type II error as well. A Type II error occurs when the company reaches a customer who is not yet ready to purchase (Venkatesan & Kumar, 2004). The LP model proposed here establishes a balance between these two errors by identifying both those RFM segments that should be reached and those RFM segments which are not worthy of pursuing because they are not profitable or because of marketing budget constraints. When faced with budget limitations, marketing managers are forced to prioritize their promotional spending strategies toward customers who will provide the highest growth in cash flows and profits. A contribution of this research is that RFM data is incorporated into an LP approach into a single model for all customers who are potential targets of a direct marketing campaign.

The paper is organized as follows. First, a brief overview of CLV and RFM models is provided. The next section discusses the modeling framework and presents the mathematical formulation of customer relationships. The empirical application of the model is then presented in which the objective is to maximize profits from potential customer purchases without exceeding the budget constraints. Three model variations are illustrated using purchasing data from over 2,300 customers from a sample from a CDNOW dataset: a recency only model, a recency and frequency model, and a full RFM model. Finally, conclusions and implications of the model are discussed.

THEORETICAL BACKGROUND: CLV, MCMS, AND RFM MODELS

CLV is a central concept in the CRM literature and is generally determined to be a function of four elements: 1) the expected duration of the customer relationship with the firm, 2) the expected revenues generated by the customer during the duration of the relationship, 3) the expected costs of marketing to the customer during that timeframe, and 4) the discount rate (Blattberg et al., 2009). Several marketing variables have been shown to have a positive effect on CLV, including customer satisfaction, marketing efforts, cross-buying, and multichannel purchasing (Blattberg et al., 2009).

CLV and Markov Chain Models (MCMs)

CLV can be estimated on the individual customer level or at the aggregate level, i.e., the total value gained from all the firm's customers (Berger & Nasr, 1998; Rust, Lemon, & Zeithaml, 2004; Kumar, Ramani, & Bohling, 2004). Using an individual CLV estimation approach involves predicting customer retention and migration (Ching, Ng, & So, 2004; Pfeifer & Carraway, 2000). Customer retention models assume that once a customer is lost, the firm's relationship with the customer is over, i.e., the customer is lost permanently. Customer migration refers to situations in which customer non-response may not necessarily indicate relationship termination (Pfeifer & Carraway, 2000). MCM models are flexible mathematical models in that they can address both customer retention and migration. That is, they are probabilistic models in that they explicitly account for the uncertainty of customers' future behavior, and they can apply to both current customers and future prospective customers as well.

The theory of MCMs is used to help direct marketing managers optimize their relationships with individual customers. Pfeifer and Carraway (2000) illustrate the relationship between MCMs and the RFM framework and note that one property of MCMs requires that future prospects for the customer relationship depend only on the current state of the relationship. This represents a challenge in constructing an RFM-based MCM since monetary value categories may violate such a property. That is, monetary value categories may be nonMarkovian by nature. Pfeifer and Carraway (2000) suggest that the monetary value category should be based either on the single last purchase amount or the cumulative total of all previous purchase amounts so that it better suits Markov chain modeling.

RFM Models

As noted earlier, one of the most widely developed MCM models is the RFM (recency, frequency, and monetary value) framework used in direct marketing in which the probability of customers' future purchases is estimated to be a function of the recency, frequency, and monetary value of their previous transactions. These RFM probabilities are used to categorize customers according to profitability potential, and they are subsequently selected (or not selected) for future direct marketing investments based on the aforementioned profit potential. Recency refers to the time of a customer's most recent purchase. Frequency is defined as the number of a customer's past purchases, and monetary value is the average purchase amount per customer transaction (Fader, Hardie, & Lee, 2005). It should be noted that some inconsistency exists in the literature regarding the conceptualization of monetary value in that some define it as an average spend per transaction, essentially equivalent to M/F, while others view it as the total amount spent by a customer on all purchases over a specified time period (Fader et al., 2005; Blattberg et al., 2009; Rhee & Mclntyre, 2009). Recency is considered especially important because a relatively long period of purchase inactivity can signal to the firm that the customer has ended the relationship (Dwyer, 1989).

The RFM approach is often used as a promotional decision-making tool in which "promotional spending is allocated on the basis of people's amount of purchases and only to a lesser degree on the basis of their lifetime of duration" (Reinartz & Kumar, 2000). The level of promotional spending is high (low) for high (low) revenue customers. Thus, companies use RFM analysis to determine whether and how to invest in their direct marketing customers (Venkatesan et al., 2007). At the same time, the three attributes are not always equally weighted. Firms tend to assign maximum importance to recency, with lesser importance attached to monetary value, and the least to frequency (Venkatesan et al., 2007; Reinartz & Kumar, 2000).

RFM has been available for many years as an analytical technique for marketing campaigns. Although more sophisticated methods have been developed recently, RFM continues to be used because of its simplicity (McCarty & Hastak, 2007). Many data mining algorithms are based on the RFM approach. Direct marketing campaigns, in particular, have become more efficient because of the use of such data mining techniques that allow marketers to better segment and manage their customer databases and to generate more effective and cost efficient promotional strategies that maximize profits derived from customers' responses. For example, marketing managers may launch a new discount pricing campaign to reach those customers who have low recency values but relatively high frequency and monetary values. Similarly, an organization could launch an up-selling campaign to reach those customers who have high recency and frequency values but low monetary values. Alternatively, the department might launch a cross-selling campaign to reach those customers who have high recency and monetary values but low frequency values.

There have been studies that use methods other than RFM analysis to evaluate the success of customer selection during a marketing campaign and to estimate the future value of customers (Venkatesan et al, 2007). For example, techniques exist for evaluating the financial return from particular marketing expenditures such as advertising, direct mail, and sales promotion (e.g., return on equity) (Rust et al., 2004). There are also studies that take RFM beyond its "traditional" direct marketing approach. For example, Eisner, Krafft, and Huchzermeier (2003) provide a dynamic multilevel model heuristic which combines recency, frequency, and monetary value segmentation with a chi-square automatic detection interaction algorithm to determine the optimal frequency of catalog mailings for a company in the mail order business, helping marketers to predict when customers should receive reactivation packages.

When presenting the top ten strategies that will be implemented by companies during 2010, a leading information technology research and advisory company predicted that "optimization and other analytical tools will be empowered by the right information delivered at the right time through customer relationship management (CRM) or other applications" (Gartner, 2009). Enabled by the data recorded in CRM systems, RFM analysis will likely continue to be at the center of this "analytical" trend. The following section proposes the framework for a direct marketing decision model that incorporates linear programming and RFM analysis.

The proposed model uses linear programming and combines data from an RFM analysis alongside marketing budgeting data for a given direct marketing campaign. The model can therefore help marketing decision makers determine whether to continue spending on certain customer segments or whether to curtail its marketing investment in a given RFM segment. A novel characteristic of this RFM approach is the marketing budget constraints that are built into the model. This approach is a realistic one in that direct marketing campaigns, like most marketing programs, typically have annual spending limits that managers cannot exceed.

LP and RFM analysis were previously utilized in Bhaskar et al., (2009). Their study used RFM analysis for personalized promotions for multiplex customers, incorporated business constraints, and provided useful insights that helped the multiplex implement an effective loyalty program. However, the algorithm in the Bhaskar et al., (2009) research separates RFM from LP: RFM is used for non-recent customers, and LP is used for current customers. A contribution of this research is that RFM data is incorporated into an LP approach into a single model for all customers who are potential targets of a direct marketing campaign.

MATHMATICAL FORMULATION OF CUSTOMER RELATIONSHIPS

In this section, a 0-1 LP formulation of the problem is provided. As stated before, the objective is to maximize profits from potential customer purchases while not exceeding the budget constraints. There is a large number of software programs designed to solve LP models. Considering the availability of the software to marketing practitioners, Microsoft Excel's Solver Add- in was selected as the tool to solve and analyze the problem.

Notations for the Optimization Models

i = 1... R index used to identify the group of customers in a given recency category;

j = 1...F index used to identify the group of customers in a given frequency category;

k = 1...M index used to identify the group of customers in a given monetary category;

V = Expected revenue from a customer's purchase if frequency/monetary value are unknown;

V^sub j^ = Expected revenue from a customer's purchase if the customer belongs to frequency group y and monetary value is unknown;

V^sub jk^ = Expected revenue from a customer's purchase if the customer belongs to frequency group y and monetary group k;

P^sub i^ = probability that a customer of recency i makes a purchases;

P^sub ij^ = probability that a customer of recency i and frequency y makes a purchase;

P^sub ijk^ = probability that a customer of recency i, frequency 7, monetary group k makes a purchase;

N^sub i^ = Number of customers who are presently in recency i;

N^sub ij^ = Number of customers who are presently in recency i and frequency j;

N^sub ijk^ = Number of customers who are presently in recency i, frequency j, and monetary group k;

C = Average cost to reach a customer during the marketing campaign;

B = Available budget for the marketing campaign;

Integer LP Model Formulation for the Recency Case

Let the decision variable for this case be a 0-1 unknown variable as follows:

X^sub i^ = 1 if customers in recency i are reached through the marketing campaign; 0, otherwise;

Using the above notations, a 0-1 mixed integer LP formulation is presented:

Maximize:

... (1)

subject to:

... (2)

x^sub i^ = {0,1} i =1 ...R (3)

Equation (1) is the objective function. It maximizes the expected profit (Z^sub r^) of the marketing campaign. As previously stated, a customer in a state of recency i has a pi chance of purchasing and a (l-p^sub i^) chance of not purchasing. When purchasing, the profit from a customer is calculated as (V - C). When not purchasing, the expected profit is simply (-C). As such, the expected value of the profit from a single customer in state i is:

P^sub i^(V -C) + (1- P^sub i^)(-C) (4)

This can be simplified as:

P^sub l^V-C (5)

Since there are Ni customers in the recency i, the expected profit from this group of customers is:

N^sub l^(p^sub i^V-C) (6)

As such, (1) indicates the sum of profits for all groups of customers for which a marketing decision to advertise to them (x^sub i^=l) is made. Equation (2) assures that the budget B for this marketing campaign is not exceeded. The left side of the equation represents the actual cost of the campaign, which is calculated as the sum of campaign costs for each group i of customers. Equation (3) represents the binary constraints for the decision variables x^sub i^.

Integer LP Model Formulation for the Recency and Frequency Case

In this section, we add frequency as a new dimension in our 0-1 LP model. The objective, again, is to maximize the profits from potential customer purchases while not exceeding the budget constraints.

Let the decision variable for this case be a 0-1 unknown variable as follows:

X^sub ij^ = 1 if customers in recency I; frequency y are reached in the marketing campaign; 0 otherwise;

The 0-1 mixed integer LP formulation is presented for the Recency and Frequency Case:

Maximize:

... (7)

subject to:

... (8)

x^sub ij^ = {0,1} i = 7...? j=1...F (9)

Equation (7) is the objective function which maximizes the expected profit (Z^sub rf^) of the marketing campaign. As previously stated, a customer in a state of recency i and frequency j has a p^sub ij^ chance of purchasing and a (1-p^sub ij^) chance of not purchasing. When purchasing, the profit from a customer is calculated as (V^sub j^ - C). When not purchasing, the expected profit is simply (-C). As such, the expected value of the profit from a single customer in state ij is:

P^sub ij^(V^sub i^-C) + (1-P^sub ij^)(-C) (10)

This can be simplified as:

P^sub ij^V^sub j^-C (11)

Since there are N^sub ij^ customers with recency i and frequency j , the expected profit from this group of customers is:

N^sub ij^(P^sub ij^V^sub j^-C) (12)

As such, (7) indicates the sum of profits for all groups of customers for which a marketing decision to reach them (x^sub ij^=l) is made. Equation (8) assures that the budget B for this marketing campaign is not exceeded. The left side of the equation represents the actual cost of the campaign, which is calculated as the sum of campaign costs for each group i of customers. Equation (9) represents the binary constraints for the decision variables X^sub ij^.

Integer LP Model Formulation for the Recency, Frequency, and Monetary Value Case

In this section, the LP model is completed by adding the third dimension of the standard RFM framework: monetary value. The objective remains the same - to maximize the profit from potential customer purchases while not exceeding the budget constraints.

Let the decision variable for this case be a 0-1 unknown variable as follows:

x^sub ijk^ = 1 if customers in recency i, frequency j, and monetary group k are reached; 0, otherwise;

Adding the monetary value category will modify the objective function as follows:

Maximize:

... (13)

subject to:

... (14)

... (15)

Equation (13) is the objective function for the RFM LP model. It maximizes the expected profit (Z^sub rfm^) of the marketing campaign. As previously stated, a customer in a state of recency /, frequency j, and monetary &has a pyk chance of purchasing and (1-p^sub ijk^ chance of not purchasing. When purchasing, the profit from a customer is calculated as (V^sub jk^ - C). When not purchasing, the expected profit is simply (-C). As such, the expected value of the profit from a single customer in state ijk is:

P^sub ijk^(V^sub jk^ - C) + (1 - p^sub ijk^)(-C) (16)

which can be simplified as:

P^sub ijk^ V^sub jk^ - C (17)

Since there are N^sub ijk^ customers with recency i, frequency j, and monetary k, the expected profit from this group of customers is:

N^sub ijk^ (p^sub ijk^V^sub jk^ - C) (18)

Equation (13) indicates the sum of profits for all groups of customers for which a marketing decision to reach them (x^sub ijk^=1) is made. Equation (14) assures that the budget B for this marketing campaign is not exceeded. The left side of the equation represents the actual cost of the campaign which is calculated as the sum of campaign costs for each group i of customers. Equation (15) represents the binary constraints for the decision variables x^sub ijk^.

Empirical Example

In order to illustrate the models, a sample of a CDNOW dataset, as used in Fader et al. (2005) is utilized. The sample has purchasing data for 2,357 customers and contains a total of 6,696 records. Each record contains the customer's ID, the transaction date, and the dollar value of the transaction. Fader et al. (2005) used this dataset to illustrate how Excel can be employed to automate the calculation process when grouping customers into RFM segments. The template and the data provided are applied here to illustrate the three different LP models: 1) a Recency model, 2) a combined Recency and Frequency model, and 3) a combined Recency, Frequency, and Monetary Value Model.

Applying Integer LP Model for the Recency Case

This case is relevant for companies which do not have accurate or complete records on the frequency or monetary value of their customers' purchases. This case can also be used for direct marketing campaigns where recency is the only value that matters (e.g., when customers who respond to the campaign will buy only one product or service at a given price, one time).

The model is applied as follows. First, customers are put in five groups where group 1 represents customers with the least recent purchases and group 5 represents those with the most recent transactions. Second, the number of customers that belongs to each group is calculated. This step can be accomplished with a pivot table. Pivot tables can also be used to calculate the number n^sub ip^, which indicates the number of customers in recency i who actually make a purchase during the next six months. Assuming that CDNOW launches a promotional campaign every six months, the probability that a customer in group i will purchase is calculated as:

... (19)

As indicated in Figure 1, given a campaign budget of B= $12,500, a cost to reach a customer of C= $7.50, and the average revenue from the purchasing customer of V=$35, the company should only select customers of recency 3, 4, and 5 for future promotional efforts.

Applying Integer LP Model for the Recency and Frequency Case

This case is relevant to companies where recency and frequency are the only significant values in their direct marketing campaign. In this situation, customers are first organized into 25 groups, with each group G^sub ij^ containing customers who belong to recency value i (1, 2..., 5) and frequency value y (1, 2..., 5). Companies are interested in determining which customer groups should be targeted and which groups should not be reached. As was the case previously, the number of customers in each group is calculated using a pivot table. Figure 2 indicates how the 2,357 customers are distributed in each group. Next is n^sub ijp^; which indicates the number of customers in recency i and frequency j who will actually purchase during the next six months. If the number of customers in group G^sub ij^ is N^sub ij^, then the probability that a customer in this group will purchase is calculated as:

... (20)

Using the same budget, cost to reach a customer, and average revenue from a purchase as in the recency model, the results demonstrated in Figure 2 indicate that the company should only target the following groups of customers: G^sub 31^, G^sub 41^ G^sub 51^, G^sub 32^, G^sub 42^, G^sub 52^, G^sub 43^, G^sub 53^, G^sub 14^, G^sub 24^, G^sub 34^, G^sub 44^, G^sub 54^, G^sub 25^, G^sub 35^, G^sub 45^, and G^sub 55^ (where G^sub ij^ represents customer groups of recency i and frequency j). In other words, the company should target all customers of recency 5 and 4 with its direct marketing campaign, and select groups of customers of a lower recency. The graph inside the Excel spreadsheet shown below indicates visually the profitability of various customer groups as a function of recency and frequency.

These results indicate that recency is a slightly better predictor of segment profitability. In 14 of 15 cases, the direct marketing company would maximize its profits at all frequency levels for recency 4 and 5 (10 cases), and in four out of five cases for recency 3 (all except G^sub 33^ are profitable). When looking at the same scenario for frequency, however, only 1 1 of 15 groups are profitable (i.e., at frequencies 3, 4, and 5). Only two groups are profitable at frequency 3 (recency 4 and 5), and nine out of 10 groups are maximizing profits at frequency 4 and 5 (all except G^sub 15^). These results somewhat supports previous research which argued that recency is the most important variable in the RFM framework (Venkatesan et al., 2007; Reinartz & Kumar, 2000).

Applying Integer LP Model for the Recency, Frequency, and Monetary Value Case

Finally, the case for companies with customer data on the recency, frequency, and monetary value of customers' previous purchases is illustrated. Using the same data subset from CDNOW, customers are in groups G^sub ijk^, which are characterized by recency value i, frequency value j, and monetary value k. The number of customers in each group is calculated using a pivot table. Similarly, the number of customers in recency i, frequency j, and monetary value k who actually makes purchases during the next six months is calculated n^sub ijk^). If the number of customers in group Gyk is Nyk, then the probability that a customer in this group will purchase is calculated as:

... (21)

It is difficult to represent the results visually because the model has four dimensions: recency, frequency, monetary value, and profit. A series of graphs is therefore provided. Figure 3 shows the results of the LP solution when the monetary values are 1, 2, and 3. For monetary values 1 and 2 (M=I, 2), the company should not target any customer segment, so these two solutions are combined into a single graph. The Solver solution indicates that customers with monetary value 3 should be targeted if recency is greater than 4 and frequency is over 3.

Figure 4 indicates the results for monetary values 4 and 5. As shown, the company should allocate its direct marketing resources toward all the customers in those monetary value segments under the condition that those segments also have relatively high values for recency and frequency.

Figure 5 provides a summary of the optimal solution for the full RFM model. This figure indicates the segments that are profitable for the company. The future promotional campaign must exclude customers with monetary values of M=1 and M=2 as these segments are clearly unprofitable at any recency and/or frequency level. For monetary value M=3, the company should target only customers with a recency score of 5 due to the high probability of response. In this cluster, segments of frequency 2, 3, and 5 are profitable. Customers in the monetary segment M=4 and frequency 1, 2, and 3 must be targeted only if they have recency scores of 4 or 5. Customers in the same monetary segment (M=4) and frequency 4 and 5 can be reached even if their recency score drops to as low as 3. Finally, customers in the monetary segment M=5 should be targeted regardless of frequency score as long as they have a recency scores of 4 or 5. For this group of customers, the recency score may even be dropped to 2 if the frequency score is 5. If the customers in segment M=5 have a recency of 3, then only those with frequencies of 1, 2, 4, or 5 will be profitable (the frequency 3 segment will be unprofitable).

As Figure 5 reveals, the full RFM model indicates that the segments that have monetary values 4 and 5 are consistently the most profitable segments overall in that they are profitable at any frequency and at multiple recency values. Lower monetary values do not lead to profitable segments even when recency and frequency scores are high, however. This appears to somewhat contradict previous research claiming that recency is the most important variable of the RFM framework (Venkatesan et al., 2007; Reinartz & Kumar, 2000). It appears that, when all three purchase history variables are incorporated into the model (recency, frequency, and monetary value), some shifting in emphasis from recency to monetary value occurs. This represents an interesting topic for discussion.

CONCLUSIONS AND RECOMMENDATIONS

Chief marketing officers have been obsessively watching the bottom line and gauging their return on investment for every spending decision, with many being forced to reduce budgets in recent years (Wong, 2009). At the same time, in 2009 the direct marketing industry was responsible for more than half of all U.S. advertising expenditures, spending just over $149.3 billion last year and generating almost $1.8 trillion in incremental sales (Direct Marketing Association [DMA], 2010). Direct marketers also employed 1.4 million people in the U.S. in 2009, with another 8.4 million additional jobs indirectly supported by direct marketing sales (DMA, 2010). Thus, despite recent economic pressures, firms large and small are recognizing the effective and growing role that direct marketing plays in a company's overall marketing arsenal.

No organization has unlimited marketing resources, however, so managers are forced to prioritize promotional spending decisions. Given the traditionally small response rates in many direct marketing campaigns (e.g., 1.65% for direct mail prospect lists to 4.41% for outbound telemarketing house lists), spending scarce resources to reach customers who are not ready to purchase (a Type II error) is clearly inefficient (Farrante, 2009; Venkatesan & Kumar, 2004). The LP model proposed here establishes a balance between Type I (missing customers who are potentially profitable) and Type II errors by identifying RFM segments that should be reached and RFM segments that are not worthy of pursuing because they are unprofitable or because of budget constraints. By indicating which customer segments will be most profitable (given certain marketing costs to reach a customer and total marketing budget constraints), an LP approach applied to RFM data can, in a single model, provide direct marketing companies with optimum decision-making capabilities regarding future promotional investments. Depending on a customer segment's profit maximization potential, a direct marketing firm can determine whether to continue its promotional spending in an attempt to generate future sales, or whether it should curtail spending and allocate those marketing resources to other, more profitable customer targets.

The analysis presented here provides the optimal solutions for three variations of the RFM model: a recency model, a recency and frequency model, and a full RFM model. The optimal solution for the recency model suggests that the company should only select customers of recency 3, 4, and 5 for future promotional efforts. The optimal solution in the recency and frequency model indicates that the company should target all customers of recency 4 and 5, and only select groups of customers of a lower recency. The optimal solution for the full RFM model indicates that any future promotional campaign should exclude all customers with monetary values of M=I and M=2 as these segments are clearly unprofitable at any recency and/or frequency level. For monetary value M=3, the company should target only customers with a recency score of 5 due to the high probability of response. In this cluster, segments of frequency 2, 3, and 5 are profitable. Customers in the monetary segment M=4 and frequency 1, 2, and 3 must be targeted only if they have a recency score of 4 or 5. Customers in the monetary segment M=5 should be targeted regardless of frequency score as long as they have a recency score of 4 or 5. Even with recency scores as low as 3, four of the five segments in the M=5 groups are profitable (where frequency is 1, 2, 4, or 5). Some of the findings supported earlier research on the importance of recency as a direct marketing variable, whereas the full RFM model tested here suggested that greater importance may need to be afforded to monetary value, at least when the reality of budget constraints are considered (Venkatesan et al., 2007; Reinartz & Kumar, 2000). This represents a potentially interesting area of future research.

The study has several limitations, all of which provide avenues for ongoing research. First, some have raised the issue of whether RFM can accurately predict future behavior or profitability given that RFM frameworks represent past or historical behavior (Blattberg et al, 2009; Rhee & Mclntyre, 2009). Of course, uncertainty in predicting behavior is inherent in any consumer decision model, and this example is no exception. Accuracy in prediction will always be a potential limitation when forecasting is based on historical data. In addition, the current model is limited to a six month time frame, whereas Venkatesan et al. (2007) note that three years is generally considered a good horizon for estimates of CLV and for CRM decisions such as customer selection. While this research does not estimate CLV, future applications of an LP approach might go beyond six months. The static nature of this model could be perceived as a potential limitation, although it does have the advantage of simplicity and ease of use for most organizations (as compared to CLV calculations).

This study made no assumptions about the nature of the costs used in the RFM model. Yet ultimately, assumptions regarding costs have an impact on CLV, and therefore may impact any RFM model as well. For example, if only variable costs of serving a customer are considered (i.e., marginal costing) as compared to full costs (with overhead allocation), the calculation of CLV could be quite different. Blattberg, Kim, and Neslin (2008) argue for marginal costing since full costing raises costs and can lead to the rejection of some customers (customers who would increase profits if they were targeted). Blattberg et al. (2009) also support the argument for marginal costing, but note that both full costing and marginal costing applications have been found in the literature. Again, these cost issues relate primarily to the prediction of CLV rather than to RFM analysis, but they do suggest that careful determination of costs is necessary. Future RFM research should take these potential limitations into account in order to continually improve the utility and reliability of this analytical method.

References:

Aeron, H., Bhaskar, T., Sundararajan, R., Kumar, A., and Moorthy, J. (2008). A Metric for Customer Lifetime Value of Credit Card Customers. Journal of Database Marketing & Customer Strategy Management, 15(3), 153-168.

Berger, P. D. and Nasr, N.I. (1998). Customer Lifetime Value: Marketing Models and Applications. Journal of Interactive Marketing, 12(Winter), 17-30.

Bhaskar, T., Subramanian, G., BaI, D., Moorthy, A., Saha, A. and Rajagopalan, S. (2009). An Optimization Model for Personalized Promotions in Multiplexes. Journal of Promotion Management, 15(1/2), 229-246.

Blattberg, R.C., Kim, B., and Neslin, S.A. (2008). Database Marketing: Analyzing and Managing Customers. New York: Springer.

Blattberg, R.C., Malthouse, E.C., and Neslin, S.A. (2009). Customer Lifetime Value: Empirical Generalizations and Some Conceptual Questions. Journal of Interactive Marketing, 23(2), 157-168.

Ching, W.-K, Ng, M. K, and So, M. K. (2004). Customer Migration, Campaign Budgeting, Revenue Estimation: The Elasticity of Markov Decision Process on Customer Lifetime Value. Advanced Modeling and Optimization, 6(2), 65-80.

Direct Marketing Association. (2010). DMA Releases Revised Quarterly Business Review (QBR) for Q4 2009. Retrieved March 15, 2010, from http://www.the-dma.org/cgi/disppressrelease?article=1393

Dwyer, R. F. (1989). Customer Lifetime Valuation to Support Marketing Decision Making. Journal of Direct Marketing, 3(11), 6-13.

Eisner, R., Krafft, M., and Huchzermeier, A. (2003). Optimizing Rhenania's Mail-Order Business Through Dynamic Multilevel Modeling. (DMLM). Interfaces, 33(1), 50-66.

Fader, P. S., Hardie, B. G. S., and Lee, K. L. (2005). Counting Your Customers the Easy Way: An Alternative to the Pareto/NBD Model. Marketing Science, 24(Spring), 275-284.

Ferrante, A. (2009). New DMA Response Rate Study Shows Email Still Strong for Conversion Rates. DemandGen Report, The Scorecard for Sales & Marketing Automation. Retrieved March 16, 2010, from http://www.demandgemeport.com/home/arcives/feature-articles/183-new-dma-response-rate-study-showsemail-still-strong-for-conversion-rates

Forbes, T. (2007). Valuing Customers. Journal of Database Marketing & Customer Strategy Management, 15(1), 410.

Gartner, Inc. (2009). Gartner Identifies the Top 10 Strategic Technologies for 2010 Press Release. Retrieved October, 24 2010, fromhttp://www.gartner.com/it/page.jsp?id=1210613

Jackson, T. W. (2007). Personalisation and CRM. Journal of Database Marketing & Customer Strategy Management, 15(1), 24-36.

Kumar, V., Ramani, G. and Bohling, T. (2004). Customer Lifetime Value Approaches and Best Practice Applications. Journal of Interactive Marketing, 18(3), 60-72.

McCarty, J. A. and Hastak, M. (2007). Segmentation Approaches in Data-mining: A Comparison of RFM, CHAID, and Logistic Regression. Journal of Business Research, 60(6), 656-662.

Pfeifer, P. E. and Carraway, R.L. (2000). Modeling Customer Relationships as Markov Chains. Journal of Interactive Marketing, 14(2), 43-55.

Reinartz, W. J. and Kumar, V. (2000). On the Profitability of Long-Life Customers in a Noncontractual Setting: An Empirical Investigation and Implications for Marketing. Journal of Marketing, 64(October), 17-35.

Rhee, E. and Mclntyre, S. (2009). How Current Targeting Can Hinder Targeting in the Future and What To Do About It. Journal of Database Marketing & Customer Strategy Management, 16(1), 15-28.

Rust, R. T., Lemon, K. N. and Zeithaml, V. A. (2004). Return on Marketing: Using Customer Equity to Focus Marketing Strategy. Journal of Marketing, 68(January), 109-127.

Stahl, H. K., Matzler, K., and Hinterhuber, H. H. (2003). Linking Customer Lifetime Value with Shareholder Value. Industrial Marketing Management, 32(4), 267-79.

Venkatesan, R. and Kumar, V. (2004). A Customer Lifetime Value Framework for Customer Selection and Resource Allocation Strategy. Journal of Marketing, 68(October), 106-25.

Venkatesan, R., Kumar, V., and Bohling, T. (2007). Optimal Customer Relationship Management Using Bayesian Decision Theory: An Application for Customer Selection. Journal of Marketing Research, 44(November), 579-594.

Vogel, V., Evanschitzky, H., and Ramaseshan, B. (2008). Customer Equity Drivers and Future Sales. Journal of Marketing, 72(November), 98-108.

Wong, E. (2009, March 3) Study: Direct Mail Spending Down 3 Percent. Retrieved March 17, 2010 from, http://www.brandweek.com/bw/content_display/news-andfeatures/direct/e3i39c0981244el31291de24a88207181a7

Zeithaml, V. A., Bitner, M. J., and Gremler, D.D. (2009). Services Marketing: Integrating Customer Focus Across the Firm. New York: McGraw-Hill Publications.

Zeithaml, V. A., Rust, R. T., and Lemon, K., N. (2001). The Customer Pyramid: Creating and Serving Profitable Customers. California Management Review, 43(Summer), 118-142.

Author affiliation:

Arben Asllani, The University of Tennessee at Chattanooga

Diane Halstead, The University of Tennessee at Chattanooga