آیا ما می توانیم ارزش طول عمر مشتری را پیش بینی کنیم؟

Relationship marketing assumes that firms can be more profitable if they identify the most profitable customers and invest disproportionate marketing resources in them. While intuitive, such strategies presume that a firm can accurately predict the future profitability of customers. In particular, we argue that the feasibility of such strategies depends on the probabilities and costs of misclassifying customers. This paper presents a detailed empirical evaluation of how accurately the future profitability of customers can be estimated. We evaluate a firm's ability to estimate the future value of customers using four data sets from different industries. Out-of-sample estimates of predictive accuracy are provided. We examine (1) the accuracy of predictions, (2) how accuracy depends on the length of time over which estimates are made, and (3) the predictors of the firm's best customers. We propose the 20–55 and 80–15 rules. Of the top 20%, approximately 55% will be misclassified (and not receive special treatment). Of the future bottom 80%, approximately 15% will be misclassified (and receive special treatment). Thus, a firm cannot assume that high-profit customers in the past will be profitable in the future nor can they assume that historically low-profit will be low-profit customers in the future.

The long-term value (CLV) of a customer “represents the present value of the expected benefits (e.g., gross margin) less the burdens (e.g., direct costs of servicing and communicating) from customers” (Dwyer,1997, p. 7). CLV has become central to relationship marketing (e.g., Sheth, Mittal, & Newman, 1999) and customer equity approaches to marketing (e.g.,Blattberg, Getz, & Thomas, 2001; Rust, Zeithaml, & Lemon, 2000). “In relationship marketing, relationships with single customers are interpreted as capital assets requiring appropriate management and investment (e.g., Hennig-Thurau & Hansen, 2000,p. 16).” Such approaches to marketing contend that a firm can ultimately be more profitable by evaluating the profitability of customers and then designing marketing programs for its best customers.Disproportionate marketing resources should be allocated to retaining best customers and keeping them loyal. This strategy would seem to make obvious sense, since it is common for a small percentage of customers to account for a large percentage of revenues and profits (Mulhern, 1999).Using CLV or predictors of CLV (e.g., historical purchasing behavior) to allocate marketing resources assumes that the future value of a customer can be estimated accurately. This assumption is rarely discussed and there is little empirical evidence evaluating it. The accuracy with which the future value of a customer can be predicted falls along a continuum. One extreme is where future behavior can be predicted perfectly given the customer’s past behavior and the firm’s marketing actions (in regression terms this would correspond to). The other extreme is where the future behavior of customers is independent of their past behavior and the firm’s marketing actions (in regression this would correspond to ). As Mulhern (1999, p. 28)notes, “models incorporating predicted future purchases are subject to a great deal of forecasting error,” but he does not quantify how much forecasting error.The firm considering whether or not to practice such relationship marketing and customer equity strategies must understand where it falls along this continuum.Investing disproportionate resources in specific customers makes unquestionable sense when their future behavior can be predicted perfectly, but no sense when future behavior is unpredictable (R2
0).R2
0 R2
1 In the latter case, an egalitarian strategy where all customers are treated equally or the quid-pro-quo incentives discussed below should be used.Suppose a firm offers two levels of treatment: “bestcustomer” treatment and “normal” treatment.Assuming the firm cannot predict the future behavior of customers perfectly, the firm can misclassify customers in two possible ways. It could misclassify a future normal customer as a future best customer—a false positive using the language of hypothesis testing—or misclassify a future best customer as future normal customer—a false negative. There are costs associated with both types of misclassifications.When a firm makes a false positive misclassification it is spending scarce marketing resources to deliver bestcustomer treatment to a future “normal” customer whose behavior does not justify such treatment. It is more difficult to quantify the costs of a false negative.The customer who deserves best-customer treatment but receives normal treatment could switch part or all of its future expenditures to a competitor, spread negative word of mouth, etc. Whether or not a firm should make disproportionate marketing investments across customers depends on the probabilities and costs of misclassifying customers. The costs of misclassification have not be quantified in either the literature or by business practitioners, to our knowledge.Some examples illustrate our point. An executive who has been using a credit card to spend a large amount of money on expensive clothing, airline tickets, car rentals, hotel rooms, cellular phone service, etc. may retire and spend far less in these categories. This executive goes from being a “best customer” of the companies that provide these products or services to a non-best customer. Showering this executive with discretionary marketing investments after retirement may not be an optimal strategy. This is an example of a false positive. Alternatively, someone who is not so valuable today can, for example, take a new job and become a star customer tomorrow—a false negative.In using historical information to allocate marketing investments a firm may be relying on chance purchases.There will always be a certain level of randomness in a customer’s purchases. Are the customers who receive special treatment really better customers? Or,are they customers who just happened to be “better”during some recent period and will “regress” back to their true, non-best-customer behavior in the future? For example, a consultant who is normally an occasional flyer on some airline may be assigned to a job in the airline’s hub city. The consultant may fly on the airline every week during the job, but resume the occasionalflyer status when the job is completed. Giving this consultant special perks will not be a good strategy.This paper provides a direct evaluation of how accurately the future behavior of customers can be estimated.The focus of this paper is on companies that maintain databases of customer/end-user information on a substantial percentage of customers and that can customize marketing “investments,” at least to some extent, across customers. Such companies include hotels, airlines, credit card companies, banks and financial service providers, companies that sell over the internet, telecommunications companies, catalogers,retail stores with “loyalty/frequent-shopper”programs, publishers, computer companies that sell direct to consumers, and many more. We shall refer to such companies as database marketing companies.The discussion here is not as applicable to organizations that do not know their specific end-users, e.g.,most producers of consumer package goods.

Neils Bohr wrote “prediction is very difficult, especially about the future.” This quote applies to making CLV estimates for the four organizations examined here. Historical value is not a very accurate predictor of future value. In situations where the future cannot be predicted accurately, an organization that invests a disproportionate amount of marketing resources in historically valuable customers may be investing in the wrong customers.Relationship marketing and customer equity strategies suggest that firms should determine the value of customers and invest disproportionately in better customers.These approaches to marketing should emphasize the importance of the accuracy of value estimates. Our empirical work suggests that if a firm offers its alleged best 20% of customers special treatment,it will frequently misclassify customers. Of the actual top 20%, approximately 55% will be misclassified(and not receive special treatment). Of the actual bottom 80%, 15% will be misclassified (and receive special treatment). Misclassifying customers has potential costs. The best customer who is misclassified as normal could defect to a competitor, develop a negative attitude towards the firm, or not consume as much as it would if given best-customer treatment.The now-normal customer who receives perks is not as deserving as others.Should organizations invest discretionary marketing resources in alleged best customers? The answer depends on the probabilities and costs of misclassifying customers, the additional revenue generated as a result of the special treatment, and the cost of the special treatment itself. In some cases this accuracy could be adequate, while in others it could be inadequate.Our point is that these misclassification rates and costs must be considered. This thought process is not currently emphasized—or even mentioned—by writers and speakers on the subject. Offering premium treatment to a select group of customers may improve that group’s CLV, but could it have a negative effect on other customer groups? Does the percentage of true positives increase substantially by offering special treatment?Rust and Oliver (2002, p. 92) ask “What happens if a firm delights the customer in one period and then reverts to the former level of quality?” They label this “hit-and-run delight.” If a customer stops receiving discretionary marketing investments, is the customer more likely to defect to a competitor than if the customer had never received any such investments?These are important research questions that need to be addressed in future research. Future research should also examine how including information about contacts affects predictive accuracy. Management judgment, absent empirical research, may not provide adequate intuition to answer this question.We have provided evidence that firms will have difficulty predicting future behavior of their customers with much accuracy. One might ask why customers’ future behavior is not very predictable? Some reasons have been hypothesized in the relationship marketing literature. As Day (2000, p. 24) notes “a strategy of investing in or building close relationships is neither appropriate nor necessary for every market, customer,or company. Some customers want nothing more than the timely exchange of the product or service with a minimum of hassles. And because close relations are resource intensive, not every customer is worth the effort.” Diller (2000, pp. 39–43) suggests classes of “demotivators of loyalty.” Opportunism means that customers are willing to “take any opportunity to get more value for the money, to be fully flexible when shopping and to only be interested in their own personal benefit (p. 40).” Variety seeking is a second reason. Autonomy “means freedom from others and decision-making independence (p. 42).” Clearly there are many potential explanations.Our discussion so far has focused on discretionary marketing investments. Our position on quid-pro-quo investments is different because the amount of a quid-pro-quo investment depends on actual future behavior, whereas the discretionary investments are made based on predicted future behavior. For example,the customer who actually flies more miles in the future will receive more free flights—the number of free flights is roughly in proportion to future miles flown. The free flight is offered as a carrot to reward desirable future behavior. The important question when deciding to offer carrots is whether the customer would behave in the same way without the carrot.See Humby, Hunt, and Phillips (2003, especially pp. 29 and 215–216) for excellent discussion on this topic and hybrid approaches where better customers are offered proportionally larger carrots than less profitable customers.Based on reading this paper, we expect that firms would be highly circumspect about their targeting and CRM strategies based on predicted customer value. The 20–55 rule means that treating lower valued customers poorly may cause defectors of potentially future high valued customers.