Optimal Mailing of Catalogs: A New Methodology Using Estimable Structural Dynamic Programming Models

Abstract

We investigate the key determinants of the optimal direct mail policy in a dynamic environment where customers maximize utility and the direct mailer maximizes profits. We measure the sensitivity of the customers to receiving a catalog in the mail, while controlling for customer characteristics such as elapsed time in responses and number of purchases. We apply our model to a database from a national cataloger that markets nonseasonal products. We summarize the results of our model that are valid for these types of products. We find that the dynamic model significantly outperforms its single-period counterpart. We find that it is not optimal to mail to individuals at low recency levels because they are likely to buy anyway. It is better to save the mailing dollars for customers at higher recency levels. We find that it is optimal to mail to customers who have purchased only a small or a medium number of times to induce them to continue to buy from this catalog and not switch to others. It is not necessary to mail often to customers who have purchased many times before from the company unless they have high recency values. We find that under the optimal mailing policy the cataloguer enjoys higher profits than under the current mailing policy.Structural Models, Principle of Optimality, Catalogs, Mailing Policy

An empirical investigation of the impact of communication timing on customer equity

“…Bitran and Mondschein (1996) study list selection decisions in an environment where budget constraints force the company to divide resources between sending costs and inventory costs. Further, Gönül and Shi (1998) make consumer response endogenous by linking the purchase decision to past actions. When maximizing expected revenues, these papers do not explicitly study the duration of the relationship.…”

“…While much of the recent research provides marketers with powerful analytical tools for selecting customers to receive a (direct) marketing communication (e.g., Bult and Wansbeek 1995, Gönül and Shi 1998, Gönül and Ter Hofstede 2003, there has been little attempt made to treat the timing of such marketing activities endogenously when calculating or maximizing the value of customer names.…”

Customer Targeting: A Neural Network Approach Guided by Genetic Algorithms

“…Routine mailings to existing customers are typically based on a statistical analysis of the household purchase history (DeSarbo and Ramaswamy 1994, Schmittlein and Petersen 1994, Bult and Wansbeek 1995, Rao and Steckel 1995, Berger and Nasr 1998, Gönül and Shi 1998, Reinartz and Kumar 2000. Marketing consultants often implement the so-called RFM (recency, frequency, monetary) approach, which targets households using summary measures computed from the customer's purchase history (Schmid andWeber 1998, David Sheppard Associates, Inc. 1999).…”

Efficient Allocation of Marketing Resources using Dynamic Programming

“…Although the use of Markov Decision Processes and dynamic programming is not new in the 5 For confidentiality reasons we cannot show the historical policy and the optimal policy in terms of marketing actions undertaken by the company. Figure 4: Comparison of the long-term value generated in 12 months when using the optimal and the historical policy.…”

“…To the best of our knowledge, with the exception of [18] where customer states are built using a supervised clustering algorithm, all the models found in the literature (e.g. [1], [5], [14], [3]) assume a given state representation ad hoc, without providing any theoretical justification. Most of the models are based on the recency, the frequency, and the monetary value (RFM) segmentation.…”