10. Marketing Models for the Customer-Centric Firm

There are other ways of expressing this basic idea. For example, instead of talking about retention and development, Bolton et al. (2004) talk of the length, depth, and breadth of the relationship between a customer and a service provider, where “the depth of a relationship is reflected in the frequency of service usage over time [... and ...] in customers’ decisions to upgrade and purchase premium (higher margin) products instead of low-cost variants [, ... and ... ] the breadth of a relationship is reflected in cross-buying or ‘add-on’ buying; that is, the number of additional (different) products or services purchased” (p. 273).

See Deighton and Johnson (2013) for an examination of the complex network of firms that collect and use data about individuals for marketing purposes.

A notable exception is the work of Natter et al. (2015).

While a sale may be the “direct response” to the advertisement, it is frequently a referral (i.e., the individual revealing that they are a prospect very interested in the product or service being advertised), which may or may not result in a sale. Calli et al. (2012) and Tellis et al. (2000) are examples of work that model the response to direct response advertising on radio and/or TV, both focusing on referrals and making no distinction between new referrals (i.e., prospects) and repeat customers.

Note that the vast majority of the response/predictive/classification models presented in the direct marketing related literature are not acquisition focused. Rather, they consider the response to mailings to existing customers (as opposed to prospects). Any model that includes past purchasing behavior as a covariate obviously falls in this category. We review this work in Sect. 10.3.3.

It is also important to consider the impact of acquisition campaigns on the behavior of existing customers. For example, offering new customers better deals than existing customers can potentially result in customer dissatisfaction—“I’ve been a loyal customer for many years and I’m getting a worse deal than new customers!” See Lhoest-Snoeck et al. (2014) for a discussion and examination of these issues.

Other work looks at the long-term implications of customer behaviour at the time of acquisition. For example, Fader et al. (2007) compare the repeat-buying behaviour of two groups of customers that differ in terms of the size of their initial purchase, and find that those with a higher initial transaction value have higher repeat-buying rates and lower attrition rates. Padilla and Ascarza (2017) explore how differences in customer behavior at the time of acquisition explain differences in expected value and customers’ sensitivity to marketing actions.

This question is partially addressed by research on allocating marketing expenditures between acquisition and retention activities, which we consider in Sect. 10.4.

This section draws on material presented in Fader and Hardie (2009, 2015). Readers are referred to these references for a deeper review of this literature.

We may also wish to include the acquisition cost in the calculation of the second quantity.

A related stream of work uses homogeneous Markov chains to characterize customer behavior (e.g., Deming and Glasser 1968; Pfeifer and Carraway 2000; Soukup 1983). Such work does not account for heterogeneity in the underlying behavioral characteristics, which can lead to misleading interferences about the nature of buying behavior (e.g., Frank 1962). See Ching et al. (2004) for an example of how these simple Markov models of customer behavior can be embedded in broader marketing optimization models.

Of course, if it is possible to characterize these time-varying covariates by a separate stochastic process, we could take the expectation of the covariate-dependent process over the distribution of covariate paths. How the resulting estimates of E(CLV) and E(RLV) would differ from those based on models of customer behavior that do no consider time-varying covariates is an open question.

David Shepard Associates (1999) use the labels contractual and implied; “an implied relationship is one in which there is no obligation on either party’s part to do anything in the future” (p. 416).

This is not the same as Jackson’s (1985) lost-for-good versus alway-a-share classification. Following Fader and Hardie (2014a), we feel that the contractual versus noncontractual classification is a better way of thinking about the nature of a firm’s relationship with its customers, as the notion of latent attrition is missing from the basic always-a-share “model.”

What lies behind this death? It could be a change in customer tastes, financial circumstances, and/or geographical location, the outcome of bad customer service experiences, or even physical death, to name but a few possible causes. But given the modeling objectives, why this death occurs is of little interest to the analyst; the primary goal is to ensure that the phenomenon is captured by the model.

Churners are typically categorized as voluntary or involuntary. Voluntary churn occurs when the customer decides to terminate their relationship with the firm, whereas involuntary churn occurs when the firm terminates the relationship (e.g., as a result of nonpayment or fraud). Involuntary churners are typically excluded when developing a churn model or modeling survival (cf. Braun and Schweidel 2011).

Nitzan and Libai (2011) examine such effects in a duration time (i.e., survival) model.

Neslin et al. (2006) and Blattberg et al. (2008) use the abbreviation LTV, which we have replaced with CLV. Strictly speaking, this should be E(RLV), but the distinction raised in (10.1) and (10.2) is ignored in most of the literature, including the work of Blattberg et al. (2008) and Neslin et al. (2006).

Of course, a firm will make use of experiments to determine the best mailing package (e.g., Bult et al. 1997).

In addition to the frequency of response, a number of researchers have considered the impact of contact history (e.g., frequency of contact) on the customer’s likelihood of responding to the current mailing, including possible irritation effects; see Schröder and Hruschka (2016) for a review. This is especially an issue in today’s permission marketing world where, for example, too much contact could result in the customer opting-out of communications all together (e.g., Drèze and Bonfrer 2008).

Whereas the work discussed above implicitly assumes that customers can only place an order (i.e., respond) in a given period if they receive a mailing, Gönül et al. (2000) recognize that customers can place orders from old catalogs, even though they did not receive a catalog in the current period.

Note that most of this work has focused on which products to offer to the firm’s customers. Khan et al. (2009) develop a model for determining which promotions to offer, if any, over a finite planning horizon. The promotions they consider are transaction, not product, specific (i.e., free shipping offers, discount coupons, and loyalty program rewards).

Once at the firm’s website, a further form of customization matches the “look and feel” of a website to each customer (e.g., Hauser et al. 2009, 2014).

At first glance, it may appear that the work that embeds Blattberg and Deighton’s (1996) model in a brand-switching framework (e.g., Tsao et al. 2014; Williams and Williams 2015) would work in a noncontractual setting. However, this is not the case; the fact that someone purchases from a competitive firm between two purchases from the focal firm should not necessarily mean that they churned after the first purchase and were acquired (again) when they made their second purchase. The notions of acquisition and retention implicit in such models are quite different from those implict in most of the literature reviewed in this chapter. (See Fader and Hardie (2014a) for a further discussion of issues related to the treatment of competition in noncontractual settings.).