|
Article Excerpt
We introduce Dynamic Multilevel Modeling (DMLM) to a multicatalog-brand environment to determine the optimal frequency, size, and customer segmentation of direct marketing activities. This optimization method leverages multicatalog-brand effects including the utilization of prior customer ordering behavior, maximization of customer value and customer share, and economies of scale and scope in printing and mailing. This enhancement of the original DMLM-approach is called Dynamic Multidimensional Marketing (DMDM). With DMLM alone, Rhenania, a German direct mail order company, turned its catalog mailing practices around and consequently rose from the number 5 to the number 2 market position. The DMLM approach was so effective that two major competitors could be bought out. Improvements provided by DMDM were threefold: more efficient resource allocation across all catalog brands, more accurate customer microsegmentation, and more effective reactivation. Presently, the company's target is to transform single-brand customer relationships into two- or three-brand relationships with higher revenue per customer. As a consequence, the Rhenania group's performance was decoupled from the overall market trend.
Key, words: customer value analysis; database marketing; direct marketing; dynamic investment analysis; mail-order business; managerial decision making; Markov processes
History: This paper was received June 5, 2003, and was with the authors 3 months for 2 revisions; processed by Gary L. Lilien, ISMS Practice Prize Editor.
1. Introduction
Rhenania was founded in 1946 and is a medium-sized mail-order company that sells books, CDs, and related products through catalogs. In the mid-1990s, the company mailed up to 20 catalogs per year to a large number of current and prospective customers. Rhenania's customer data is stored in a proprietary database containing customer addresses, contact and buying history, and sociodemographic data. It augments this list of customers with names that it either rents from commercial address brokers (rental lists) or gains directly through campaigns in print or similar media. In the mid-1990s, Rhenania's database contained about 600,000 names, and each year, it mailed about 2,400,000 catalogs.
After the German reunification in 1990, the German economy was booming. However, this boom did not last. Unemployment has increased steadily since the early 1990s. Consequently, the propensity to consume has decreased in almost every consumer market. Thus, Rhenania has to deal with lower response rates, fewer active buyers, and lower annual sales and profits, especially after 1994.
Until 1997, Rhenania had followed the standard marketing approach to managing customer contacts and selecting profitable customers for each mailing. This traditional approach is also recommended in academic publications on direct marketing, for example, by Bult and Wansbeek (1995) or Colombo and Jiang (1999). In other words, Rhenania only sent catalogs to clients if expected revenue exceeded the total cost of the following: merchandising, order fulfillment, and the mailing itself. This strategy led to an increased profitability of single direct mail campaigns, but the longer-term consequence was a dramatic overall company downturn, because of a shrinking active customer base, declining sales, market share, and profits.
Given that promotion and mailings in particular constitute the most effective marketing mix element in the catalog mail-order industry, Rhenania decided it was necessary to rethink its overall mailing strategy. More specifically, three major issues had to be addressed (Bitran and Mondschein 1996, Elsner et al. 2003):
* How often and when should customers receive mailings over a given period (frequency and timing of direct marketing campaigns)?
* How many customers or customer segments in the company's database should be contacted (size of direct marketing campaigns)?
* Which customers should be included in which campaign (customer segmentation)?
As will be shown in the following sections, Rhenania was able to address these issues successfully by applying dynamic multilevel modeling (DMLM), a multilevel modeling approach for determining the optimal number, frequency, and size of direct marketing activities, as well as customer segmentation over a rolling horizon of one year. This general notion of coordinating and optimizing direct marketing contact over time has been discussed in a number of publications by direct marketing researchers and practitioners, such as Wunderman (1997), Kestnbaum et al. (1998), or Malthouse (2003). However, none offers any evidence that such an approach works, nor does anyone spell out any details. Our innovative concept not only helped turn Rhenania's mailing business around in 2001, but also DMLM enabled Rhenania to acquire Akzente and Mail Order Kaiser, which were two of Rhenania's major competitors. This led to a new challenge, namely, optimizing direct marketing activities dynamically across several catalog brands. Although this problem was new to Rhenania, it is quite common among large catalog retailers: Many mail-order companies that deal with several catalog brands face multidimensional analysis problems. For example, consider the OTTO Combined Group, the world's largest mail-order company. OTTO Combined Group serves the German full-line mail-order market with its major brands OTTO, Schwab, Witt Weiden, and Baur (see http://www.otto.com). In an interview, a top executive of the OTTO Combined Group confirmed that a substantial number of OTTO's customers order products from two, or even more, of the group's catalog brands.
In a multicatalog-brand environment like the one described above, an optimization method is needed to leverage multicatalog-brand effects. This method includes utilizing prior-customer ordering behavior across different catalog brands and exploiting the potential for maximizing customer value and customer share across those brands, as well as economies of scale and scope in printing and mailing. However, the DMLM approach was developed to optimize mailing strategies for a single catalog brand only. The modeling concept does not take into account any interdependencies across a group of catalog brands. In order to leverage these multicatalog-brand effects, a more enhanced approach had to be developed and applied simultaneously to Rhenania's catalog brands. This new approach is called dynamic multidimensional marketing (DMDM) and will be explained in more detail later.
This article is organized as follows. In [section] 2, we review the literature on current approaches and optimization models in direct marketing. In our view, no single concept can provide answers to the three important marketing questions described earlier. For this reason, we developed the DMLM model. Section 3 provides a nontechnical description of the basic philosophy behind the DMLM approach. Section 4 conveys mathematical details of the model that helps optimize mailing strategies over a rolling horizon. The successful application of DMLM to Rhenania's customer base has been remarkable and is described in [section]5. However, after the acquisition of Akzente and Mail Order Kaiser in 2001, applying DMLM to Rhenania's three catalog brands separately revealed that multicatalog-brand effects were not adequately leveraged by the model. Therefore, we developed a new multidimensional approach called dynamic multidimensional marketing (DMDM). This model extension is described conceptually in [section]6 and mathematically in [section]7. The practical implications of DMDM are discussed in [section]8. In [section]9, the paper concludes with a summary, in which further issues such as transferability to other industries and the limitations of the DMLM and DMDM approaches are discussed.
2. Literature Review
Prior to developing our own model, we reviewed the relevant direct marketing literature. Several concepts aim at facilitating an appropriate decision about which customers in a company's database should be targeted. A common characteristic of these methods is that they claim to indicate the future value of catalog company customers. A classic and very common approach is to use an RFM model, in which an estimation of the likelihood of a future response is based on recency R, frequency F, and the monetary value M of past responses. Regression methods are a key technique for modeling the relationship of RFM to response (Colombo and Jiang 1999). Since response is usually measured by a buy or no-buy decision, logistic regression is used frequently (Hosmer and Lemeshow 2000) and, particularly where many other variables are available and interactions are suspected, tree-based regression methods (e.g., AID, CHAID, CART) have proven their efficiency (Haughton and Oulabi 1993). Neural networks have been used to uncover relationships between response and behavioral, demographic, and other predictor variables (Zahavi and Levin 1995).
Another common approach to predicting the future behavior of customers is the use of direct marketing scoring models. A scoring model is defined as a data-mining model that predicts the likelihood of some behavior, based on other information available about a prospect or customer. A scoring model assigns every observation in a database a score indicating how likely someone is to engage in a particular behavior (Malthouse 2003). The functional form and estimation of...
|
