Lifetime Value: Empirical Generalizations and Some Conceptual Questions

Robert C. Blattberg

Edward C. Malthouse

Scott A. Neslin

Robert C. Blattberg1

Edward C. Malthouse2

Scott Neslin3

August 2007

1Polk Bros. Professor of Retailing; Professor of Marketing; Director of the Center for RetailManagementCenter, Kellogg School of Management,

2Theordore and Annie Sills Associate Professor of Integrated Marketing Communications, Medill School of Journalism, Northwestern University,

3Albert Wesley Frey Professor of Marketing, Tuck School of Business at Dartmouth,

Lifetime Value: Empirical Generalizations and Some Conceptual Questions

Abstract: From the extant literature on LTV we identify four empirical generalizations (well-defined, consistent effects found by at least three different authors): (1) customer satisfaction increases LTV; (2) marketing efforts are associated with higher LTV; (3) cross-buying is associated with higher LTV; and (4) multichannel purchasing is associated with higher LTV. The frequency and monetary value of previous purchases generally have a positive effect on LTV, although there are contradictory findings. We identify additional issues that have received limited attention in the literature, but require further empirical study: the effects of pricing and promotions on LTV, managing a sequence of contacts to maximize response rates and LTV, and whether LTV can be forecasted sufficiently accurately. Eight conceptual or strategic issues are identified and discussed.

Many firms are now focusing on identifying their most profitable customers and nurturing long-term relations, which represents a different way of making and evaluating marketing decisions and the product-centric approach. At the center of the approach is Lifetime Value of a customer (LTV). It is a pivotal concept in the customer-centric approach to marketing that pervades the many customer relationship management approaches that are frequently discussed and practiced by firms such as one-to-one, loyalty, and database marketing. Lifetime value (LTV) is the present value of all the future cash flows attributed to a customer relationship (Pfeifer et al. 2005, p 17). Due to uncertainty in future customer, firm, and competitor behavior, LTV is in reality a random variable and most applications calculate expected LTV, which can bewritten as:

(1)

where is the customer’s net contribution in period t, and d is the discount rate. Net contribution is driven by: (1) customer duration; that is, whether the customer is still active in period t; (2) revenues generated by the customer in period t, given he or she survives to that period; and (3) costs of serving the customer in period t. There are thus four components of LTV: (1) duration, (2) revenues, (3) costs, and (4) discount rate.

LTVcan be used to guide the firm’s acquisition and retention activities, and issometimes aggregated over customers as a measure of firm or segment value. Research on LTV and its components is an active area and there are many research articles that propose methods of estimating LTV or its components under various conditions, study their antecedents, attempt to maximize LTV over some space of marketing actions, or discuss its applications. Given this widespread interest in customer-centric marketing and LTV, it is important to take stock of what is known and needs to be known about LTV.

There are excellent surveys of this subject (e.g., see Jain and Singh 2002 or Blattberg, Kim and Neslin, 2008). As an indication of breadth and interest on this subject, the survey in the latter reference spans multiple chapters.We shall not attempt to duplicate this effort at summarizing the field here. In particular, space will not permit us to review the models used to estimate LTV. The purpose of this article is to (1) discuss empirical generalizations that can be drawn from extant academic research literature, (2) discuss empirical findings that do not yet reach the threshold for generalization but are suggestive and interesting, and (3) discuss conceptual and strategic issues relating to LTV. This is a subjective list of generalizations and conceptual issues based on our review of the literature and what we think is mostimportant.

Blattberg, Briesch and Fox (1995, p G123) defined empirical generalizations as “(1) the topic being analyzed is well defined; (2) there are at least three articles by at least three different authors in which empirical research has been conducted in the specific area; and (3) the empirical evidence is consistent, i.e., the sign of the effect is the same in each of the articles.” It is important to study empirical generalizations for several reasons. From a research viewpoint, multiple studies reporting a similar empirical conclusion can verify theory or identify areas where theoretical work is needed. From a practical viewpoint, empirical generalizations can provide managers with guidance in making marketing decisions.

In the following section, we present issues that satisfy the above definition of an empirical generalization. Next, we discuss issues for which there is some research evidence, but not enough (in our judgment) to reach the status of an empirical generalization. We then discuss conceptual issues that are not empirical per se but influence our understanding of how to apply LTV. We close with a brief summary.

Empirical Generalizations

We have identified four findings that we believe qualify as empirical generalizations regarding LTV: customer satisfaction, marketing, cross-buying, and multichannel purchasing all have positive relationships with LTV.[1] Table 1 summarizes research addressing each generalization. Much of the evidence is based on using one of the earlier described components of LTV as the dependent variable. Because of equation (1), if there is a relationship with a component, there will be a relationship with LTV.[2]

  1. Customer satisfaction has a positive effect on LTV. There is a large literature relating customer satisfaction with loyalty and measures of firm performance. The consensus is that satisfaction has a positive relationship with loyalty, retention and profitability,[3] although Yeung et al. (2001) find that the strength of the relationship and the magnitude of the impact appear to vary with the choice of performance measures and industry sector. For example, the relationship is strong in the financial sector, but much weaker in the technology and communications sectors. This suggests that there are sector characteristics that moderate the relationship between satisfaction and profitability and points out the need for a richer theory. Yeung’s articles also find that the amount of variation in performance measures explained by satisfaction is generally small, suggesting other causal factors should be included. Most of the empirical work has used measures of firm profitability as dependent variables and more work is needed to investigate this relationship at the customer level.
  2. Marketing efforts are associated with higher LTV: The evidence summarized in Table 1 suggests a strong association between marketing efforts and customer duration. One study looked at customer “profitability” (Reinartz, Thomas, and Kumar 2005) and found a positive relationship. That marketing influence LTV is a requirement for LTV to be a useful marketing metric. However, the finding of a positive association is nontrivial. Firms might spend more money trying to rescue customers who have shorter duration; they might spend too much on marketing to customers who would have purchased anyway, reducing the profitability of these efforts.

The ramifications of this generalization are crucial because they suggest firms can formulate marketing activities to manage LTV over time. Note however that most of the evidence is based on associations. The problem is that firms might choose to expend higher marketing efforts on more valuable customers, and thusLTVcaused increased marketing rather than the reverse. A formal selectivity model (Woodridge 2002) would be needed to sort out the causality issue, where there would be two equations: one for customer profitability and the second for firm marketing efforts.

  1. Cross-buying is associated with higherLTV. Reinartz and Kumar (2003, pp. 81-82) give a literature review and develop a theoretical rational for cross-buying having a positive effect on duration. Table 1 shows that there is generally a positive effect, although some authors do not find a significant effect. There is some question as to whether the relationship between cross-buying and LTV is spurious. It could be that customers who highly prefer a company buy often from it and also buy from several departments. For example, if customers are loyal to a given electronics retailer because of its service, they are likely to make multiple purchases from that retailer such as TV’s, DVD’s, and computer equipment. Reinartz, Thomas, and Bascoul (2006) used Granger causality tests to assess the direction of causality, and found that profitability caused cross-buying rather than the reverse. So while we clearly have a positive association between cross-buying and LTV, we still need to determine causality.
  2. Multichannel purchasing is associated with higher LTV: This is a significant finding in the multichannel literature (see Neslin and Shankar (2008) in this issue of Journal of Interactive Marketing, as well as Neslin et al. (2006)). Multichannel purchasing could increase LTV because it creates switching costs or increases customer satisfaction. For example, customers who use a bank’s ATM, branch office, and online service, must extricate themselves from several contact points in order to switch to another bank. Hence switching costs are high. In a more positive vein, customers may be more satisfied because dealing with the bank is convenient andthey therefore giveit more business.

The evidence in favor of positive association is fairly strong, especially, although not exclusively, in the retail industry. One interesting exception can be found in Campbell and Frei (2006), who find that while adopters of online banking increase usage frequency, total revenue goes down, possibly due to customers managing their assets more effectively.

The question of causality rears its head again in the case of multichannel usage, similarly to the case of cross-buying. Conceptually, the phenomena are very similar – multichannel purchasing is “cross-buying” across the firm’s channels rather than the firm’s departments. Again, the multichannel shopper may become more satisfied with the company, and hence becomes more loyal and more valuable. The shopper may also receive more marketing simply by visiting various channels. On the other hand, high-value customers may self-select into using all the firm’s channels. This explanation is refuted by Ansari et al. (2007), but that is only one study. Further work is needed in this area. If indeed the relationship is multichannelincreases LTV, the implication is clear – firms should encourage multichannel usage.

An Issue with Conflicting Empirical Evaluations

  1. How does RFM affect LTV? Previous purchase behavior is often summarized by the time since the most recent purchase (R=recency), the number of previous purchases (F=frequency), and the total amount spent (M=monetary value)– RFM. Since these variables are widely known for existing customers, they are often included in LTV modelsand used to make customer-level estimates. Frequency and monetary are often highly correlated and it could be argued that, at least in some situations, they are different measures of the same underlying construct, previous buying intensity. In our discussions with practitioners, most believe that recency has a negative relationship with LTV, and frequency and monetary to have a positive relationship with customer value,[4] but this has not been found consistently in the academic literature. Nagar and Rajan (2005) find a positive relationship loan volume and firm profitability. Reinartz and Kumar (2003) find positive relationships between previous spending levels and lifetime duration. The models in Malthouse and Blattberg (2005) were optimized for predictive accuracy and some had substantial multicollinearity, inhibiting the interpretation of effect signs, but across over 100 catalog companies, a software company, a nonprofit organization, and an educational service provider they consistently found frequency and monetary variables had positive relationships with individual-level long-term value and recency had a negative effect (the longer a customer has been inactive the lower LTV is). Li (1995), however, finds a negative effect on duration at the customer level and Niraj et al. (2001) find a negative relationship with customer profit, but it is unclear how Li defines his “Usage” variable and Niraj et al. define their “Number of Orders” variable. Clearly this area needs more research, but our view is that frequency and monetary are likely to have a positive effect and recency a negative effect on LTV.

Issues with Few Empirical Evaluations

  1. How does pricing affect LTV? There has been very little research on this question. Thomas et al. (2004) study the effects of reacquisition pricing strategies for newspaper subscriptions on the likelihood of reacquiring a customer and on the duration after reacquisition. In a longitudinal study, they examine the price offered at the time of reacquisition as well as prices charged after reacquisition. They conclude that an optimal pricing strategy would be to offer a low reacquisition price and higher prices after a customer has been reacquired. Reinartz and Kumar (2000) study the converse question of how a customer’s length of tenure with a firm affects LTV. They find that the shortest-life segment of customers have a significantly higher average price paid for a single product item for a catalog company but warn that this finding could be confounded with other variables. Further empirical work is necessary to understanding pricing effects. The direction of causality is also not clear.
  2. How do promotions affect LTV? Anderson and Simester (2004) studied the long-term impact of promotion depth for a catalog company using three randomized controlled experiments. All customers in the study received a catalog including merchandise indicated as being “on sale,” but the prices for the sale items were lower for the “promotion” group. In all tests promotions increased short-term sales, but those in the promotion group purchased less in the 12 months after the promotion than during the subsequent year, suggesting acceleration. They also found that deeper price discounts increased future purchases by first-time customers, but reduced future purchases by established customers. This conclusion should be tested for other firms and in other industries. Li (1995) also finds that the hazard of cancelling long-distance telephone service decreases when a discount plan is offered. Overall, there is some evidence, mostly through Anderson and Simester’s study, that database marketing communications can act like promotions and induce the same long-term effects observed in the promotions literature (e.g., see Neslin 2002 or Blattberg et al. 1995), but more empirical work is needed to measure these effects in other settings. Optimal contact models should consider these issues. For example, a communication may accelerate a purchase, so it would not make sense to communicate again until sufficient time has elapsed for the customer to become ready to buy again.
  3. What is the relationship between the number of contacts and response rates? Can too many contacts actually decrease response rates? These questions are important because they provide a non-cost reason to limit the number of contacts – fewer contacts could potentially produce more sales. The answer to the second question seems to be yes, but further work is necessary on the first to know the points at which customers “wear out” and understand the theoretical reasons they do so. Ansari et al. (2007) find a negative interaction between purchase frequency and like communications as well as different communications. For example, they find that two catalogs delivered at nearly the same time have a smaller impact on purchase frequency than if the catalogs are delivered at more highly separated times. They find the same negative interactions between successive emails and between successive emails and catalogs. In evaluating a new way of deciding which customer should receive a catalog, Gönül, Kim and Shi (2000) conclude that “it is better to send (not too many) a few catalogs with (not too long but) a moderate amount of time between them to encourage purchase (pp. 2-3),” also supporting wear out. Using self-reported data, Eastlick, Feinberg and Trappey (1993) conclude that the relationship between the number of catalogs and purchase history (e.g., frequency and monetary) is fit better by a quadratic inverted-U function than a linear one. Ganzach and Ben-Or (1996) critique their conclusions and rationale, and Feinberg, Eastlick, and Trappey (1996) defend them. Campbell et al. (2001) suggest the relationship is concave with diminishing returns, but do not report that the returns become negative at some point.

In summary, it seems clear that wear-out is a real phenomenon, in that there are decreasing returns to marketing within a given time period, and that the spacing between contacts can counteract wear-out. While this makes it plausible that a company could produce higher LTV with fewer communications, even if the communications are costless, this specific finding needs to be demonstrated. Further if the shape of the curve is an inverted-U, it is critical to determine the wear-out point. Also, theories need to be developed which explain why the shape is an inverted-U.