ArticlePDF Available

Optimal Selection for Direct Mail

Authors:

Abstract

Direct marketing (mail) is a growing area of marketing practice, yet the academic journals contain very little research on this topic. The most important issue for direct marketers is how to sample targets from a population for a direct mail campaign. Although some selection methods are described in the literature, there seems to be not a single paper discussing the analytical and statistical aspects involved. The objective of this paper is to introduce a comprehensive methodology for the selection of targets from a mailing list for direct mail. At least theoretically, this methodology leads to more efficient selection procedures than the existing ones. The latter are not based on an optimal selection strategy, whereas we explicitly take the profit function into account. By equating marginal costs and marginal returns we determine which households should receive a mailing in order to maximize expected profit. In the empirical part we show that our methodology has great predictive accuracy and generates higher net returns than traditional approaches.
Copyright 1995, by INFORMS, all rights reserved. Copyright of Marketing Science is the property of
INFORMS: Institute for Operations Research and its content may not be copied or emailed to multiple sites or
posted to a listserv without the copyright holder's express written permission. However, users may print,
download, or email articles for individual use.
... Churn prediction models typically analyze historical data to identify patterns in customer behavior and early signs of attrition (Shirazi and Mohammadi, 2019). Recency, frequency and monetary value (RFM) analysis (Bult and Wansbeek, 1995;Kitchens et al., 2018;Martens et al., 2016) has become a popular technique for evaluating customer value. RFM assesses recency (time since the last activity or interaction, for instance, the last purchase), frequency (number of activities or interactions within a given period, for instance, the number of visits) and monetary value (amount of value or impact generated during that period). ...
Article
Purpose Although prior research has employed various variables to predict player churn, the dynamic evolution of the behavioral patterns of players has received limited attention. In this study, churn prediction models are developed by incorporating the progress level, in-game purchase, social interaction, behavioral pattern and behavioral variability (BV) of players in social casino games (SCGs). The study distinguishes churn prediction between two player groups: monetizers and non-monetizers. Design/methodology/approach This study employs three machine learning techniques—logistic regression, decision trees and random forests—using real-world player data from an SCG company to construct churn prediction models. Two experiments were conducted. In Experiment 1, BV was combined with four other variable categories to effectively predict churn behaviors across all players ( n = 52,246). In Experiment 2, churn prediction models were developed separately for monetizers ( n = 16,628) and non-monetizers ( n = 35,618). Findings The findings from Experiment 1 indicate that incorporating BV significantly improves the overall performance of churn prediction models. Experiment 2 demonstrates that churn prediction models achieve better performance and predictive accuracy for monetizers and non-monetizers when BV is calculated over the 3-day to 7-day and 7-day to 14-day windows, respectively. Originality/value This study introduces BV as a novel variable category for churn prediction, emphasizing within-person variability and demonstrating its effectiveness in enhancing model performance. Churn prediction models were independently constructed for monetizers and non-monetizers, utilizing different time windows for variable extraction. This approach improves predictive performance and highlights key differences in critical variables influencing churn across the two player groups. The findings provide valuable insights into churn management strategies tailored for monetizers and non-monetizers.
... To analyze game-related variables with respect to game items, the RFM model (recency, frequency, and monetary) was taken into account. The RFM model is frequently used to analyze customers' use patterns and loyalty [44]. R (recency) denotes the latest purchase date; the closer a customer makes a purchase, the more likely he or she is to make another purchase. ...
Article
Full-text available
The consumption of games has received increasing attention due to their high profits and addiction issues. However, previous studies have focused mainly on players’ in-game purchases, neglecting the purchase of game derivative products. This article provides the first exploration of the differences and similarities between in-game purchases and derivative product purchases with a mixed-method approach. A quantitative survey collected data from 9864 game players, and the results suggested that there were differences between in-game purchases and derivative product consumption in terms of consumption amount and number of participants, and that derivative product purchases had a stronger relationship with character attachment and game loyalty. Subsequent interviews were conducted with 22 players. The findings supported the quantitative results and revealed that players exhibited a distinct understanding of each type of purchase in terms of ownership. Overall, these findings contribute to the understanding of game derivative product purchases and explore the factors influencing in-game purchases and derivative product purchases. We strongly argue that the pattern of spending on game derivative products is very different from that of spending on in-game purchases and is, thus, worthy of dedicated research.
... The positive effect of number of transactions on CLV has been proven by the following studies: Foster et al. (1996), Gurǎu and Ranchhod (2002), Jašek et al. (2018), Sifa et al. (2018), while Bult and Wansbeek (1995), Zhang et al. (2010) are more careful in their conclusions and recommend using experiment to test causal relationships. The association between CLV and customer loyalty (measured by number of repeat purchases) can thus be confirmed i.e. we confirm a very strong association of customer loyalty on CLV, including in the context of SOEFIOR e-shops. ...
Article
Full-text available
This paper investigates customer lifetime value (CLV) in e-shops, particularly those operated by small on-platform evolving financially independent online resellers (SOEFIOR) e-shops. The aim is to identify factors predicting CLV and assess their associations with CLV. Given the nested structure of the data, where transactions by customers are clustered within e-shops, a multilevel model is employed as the analytical framework. While classical linear regression assumes independence of observations within a sample, our dataset operates across three hierarchical levels: transaction level (I), customer level (II), and e-shop level (III). This hierarchical structure challenges the validity of inferences drawn from linear regression models, as transactions by one customer are not independent, and customers within a single e-shop may exhibit interdependencies. Therefore, a multilevel model is utilised to appropriately address the dependence among transactions within this nested data structure. The analysis reveals that the “number of transactions” exhibits the strongest positive association with CLV, followed by “days to transaction” and “session duration”. Furthermore, we discovered that “direct access” exhibits a positive association with CLV compared to access through Google campaigns, whereas access through Facebook campaigns demonstrates a negative association with CLV when compared to Google campaigns. Additionally, using the e-shop on mobile and landing on the product details page both show negative associations with CLV compared to desktop usage and landing on the e-shop’s home page, respectively. Our research identifies several variables that are associated with CLV in e-shops. This enables e-shop managers to effectively target and engage customers through marketing activities, thereby maximising revenues, financial performance, and customer CLV.
... Scale, the Community sample gave signi cantly more money to charity than the Student Sample, and it gave more in Time 1 than Time 2. This may re ect the fact that community members (older, employed individuals) have more disposable income than students (young people with little money), and so they give more to charity [47]. Over time in a crisis, however, individuals are encouraged to rein in expenses to help get them through tough times. ...
Preprint
Full-text available
In the midst of the COVID-19 pandemic’s peak, many individuals faced unprecedented levels of perceived threat. That being said, it has been reported that many individuals also reported positive psychological outcomes and engagement in altruistic or prosocial behaviour. This quasi-experimental study delves into the interplay of sample types (students versus community members) and sample timing (pre- and post-vaccine availability) on psychological disruption variables tied to positive psychological outcomes. Through a comprehensive questionnaire comprising seven measures, administered online to participants from distinct populations at two crucial time points, our findings unravel a compelling narrative. Notably, the community and Time 1 sample consistently exhibit heightened levels of psychological disruption, posttraumatic growth, and positive psychological outcomes. As anticipated, mediation analysis underscores the pivotal role of a disruption in core beliefs in mediating the relationship between threatening events and posttraumatic growth. This study not only sheds light on these intricate dynamics but also holds implications for researchers navigating the multifaceted landscape of psychological outcomes in the face of unprecedented global challenges.
Preprint
Full-text available
Customer Lifetime Value (CLV) is an important metric that measures the total value a customer will bring to a business over their lifetime. The Beta Geometric Negative Binomial Distribution (BGNBD) and Gamma Gamma Distribution are two models that can be used to calculate CLV, taking into account both the frequency and value of customer transactions. This article explains the BGNBD and Gamma Gamma Distribution models, and how they can be used to calculate CLV for NFT (Non-Fungible Token) transaction data in a blockchain setting. By estimating the parameters of these models using historical transaction data, businesses can gain insights into the lifetime value of their customers and make data-driven decisions about marketing and customer retention strategies.
Article
Full-text available
Introduction: In the modern competitive landscape of football, clubs are increasingly leveraging data-driven decision-making to strengthen their commercial positions, particularly against rival clubs. The strategic allocation of resources to attract and retain profitable fans who exhibit long-term loyalty is crucial for advancing a club's marketing efforts. While the Recency, Frequency, and Monetary (RFM) customer segmentation technique has seen widespread application in various industries for predicting customer behavior, its adoption within the football industry remains under explored. This study aims to address this gap by introducing an adjusted RFM approach, enhanced with the Analytic Hierarchy Process (AHP) and unsupervised machine learning, to effectively segment football fans based on Customer Lifetime Value (CLV). Methods: This research employs a novel weighted RFM method where the significance of each RFM component is quantified using the AHP method. The study utilizes a dataset comprising 500,591 anonymized merchandising transactions from Amsterdamsche Football Club Ajax (AFC Ajax). The derived weights for the RFM variables are 0.409 for Monetary, 0.343 for Frequency, and 0.248 for Recency. These weights are then integrated into a clustering framework using unsupervised machine learning algorithms to segment fans based on their weighted RFM values. The simple weighted sum approach is subsequently applied to estimate the CLV ranking for each fan, enabling the identification of distinct fan segments. Results: The analysis reveals eight distinct fan clusters, each characterized by unique behaviors and value contributions: The Golden Fans (clusters 1 and 2) exhibit the most favourable scores across the recency, frequency, and monetary metrics, making them relatively the most valuable. They are critical to the club's profitability and should be rewarded through loyalty programs and exclusive services. The Promising segment (cluster 3) shows potential to ascend to Golden Fan status with increased spending. Targeted marketing campaigns and incentives can stimulate this transition. The Needs Attention segment (cluster 4) are formerly loyal fans whose engagement has diminished. Re-engagement strategies are vital to prevent further churn. The New Fans segment (clusters 5 and 6) are fans who have recently transacted and show potential for growth with proper engagement and personalized offerings. Lastly, the Churned/Low Value segment (clusters 7 and 8) are fans who relatively contribute the least and may require price incentives to potentially re-engage, though they hold relatively lower priority compared to other segments. Discussion: The findings validate the proposed method's utility through its application to AFC Ajax's Customer Relationship Management (CRM) data and provides a robust framework for fan segmentation in the football industry. The approach offers actionable insights that can significantly enhance marketing strategies by identifying and prioritizing high-value segments based on the club's preferences and requirements. By maintaining the loyalty of Golden Fans and nurturing the Promising segment, football clubs can achieve substantial gains in profitability and fan engagement. Additionally, the study underscores the necessity of re-engaging formerly loyal fans and fostering new fans' growth to enable long-term commercial success. This methodology not only aims to bridge a research gap, but also equips marketing practitioners with data-driven tools for effective and efficient customer segmentation in the football industry.
Article
In recent years, various novel techniques have emerged in the realm of deep learning for enhanced pattern recognition. Multimodal learning is a widely used approach that enables simultaneous data input across multiple modalities, including video, audio, and text.For customer relationship management in the field of marketing, a comprehensive analysis using combinations of multiple data, such as the behavior log and survey responses, is actively pursued to assess customer loyalty and predict future behavior. However, these modalities of datasets are typically aggregated into a single dataset comprising variables that are hand-crafted by the analysts. This may not fully exploit the high predictive performance achieved by feature extraction in deep learning techniques.In this study, we employ a source--target multihead attention transformer encoder in conjunction with serial feature fusion, which enables the creation of a multimodal and context-aware deep learning model. This model can simultaneously process two high-dimensional datasets―time-series panel data that aggregates daily smartphone app usage and cross-sectional data containing demographic variables and survey responses. Finally, the both data are effectively fused at the upper layers of the model for the output. Results of the exhaustive analysis demonstrate that the proposed model outperforms other major deep learning architectures.
Article
Full-text available
The authors examine the effects of a manufacturer coupon on brand choice behavior. The level of coupon redemption and changes in brand choice behavior after redemption are examined as a function of the household's prior probability of purchasing the promoted brand, likelihood of buying a favorite competitive brand, and coupon face value. A model of the coupon redemption decision is developed to predict response to the coupon promotion by different consumer segments. Predictions from the model are tested by using scanner panel data from a field experiment on coupon face values. Coupon redemption rates are found to be much higher among households that have purchased the brand on a regular basis in the past. The results also suggest that most consumers revert to their precoupon choice behavior immediately after their redemption purchase. These and other findings have important implications for the profitability of coupon promotions.
Article
Full-text available
Several authors have noted that the profitability of a coupon promotion depends on the incremental sales generated by the coupon. However, most prior research on coupon promotions has focused on redemption rates and little is known about the characteristics of households that make incremental purchases. The authors develop and test several hypotheses about the characteristics of households that make incremental purchases in response to a direct mail coupon promotion. For the product tested, coupons produced greater incremental sales among households that were larger, more educated, and were homeowners. The findings suggest that directing coupons to the most responsive market segments can increase profits significantly.
Article
Full-text available
Simulated annealing is a stochastic strategy for searching the ground state. A fast simulated annealing (FSA) is a semi-local search and consists of occasional long jumps. The cooling schedule of the FSA algorithm is inversely linear in time which is fast compared with the classical simulated annealing (CSA) which is strictly a local search and requires the cooling schedule to be inversely proportional to the logarithmic function of time. A general D-dimensional Cauchy probability for generating the state is given. Proofs for both FSA and CSA are sketched. A double potential well is used to numerically illustrate both schemes.
Article
Full-text available
In many regression applications, users are often faced with difficulties due to nonlinear relationships, heterogeneous subjects, or time series which are best represented by splines. In such applications, two or more regression functions are often necessary to best summarize the underlying structure of the data. Unfortunately, in most cases, it is not known a priori which subset of observations should be approximated with which specific regression function. This paper presents a methodology which simultaneously clusters observations into a preset number of groups and estimates the corresponding regression functions' coefficients, all to optimize a common objective function. We describe the problem and discuss related procedures. A new simulated annealing-based methodology is described as well as program options to accommodate overlapping or nonoverlapping clustering, replications per subject, univariate or multivariate dependent variables, and constraints imposed on cluster membership. Extensive Monte Carlo analyses are reported which investigate the overall performance of the methodology. A consumer psychology application is provided concerning a conjoint analysis investigation of consumer satisfaction determinants. Finally, other applications and extensions of the methodology are discussed.
Article
Full-text available
When the binary choice probability model is derived from a random utility maximization model, the choice probability for one alternative has the form F[V(z, θ)]. Here V(z, θ) is a given function of the exogenous variables z and unknown parameters θ, representing the systematic component of the utility difference, and F is the distribution function of the random component of the utility difference. This paper describes a method of estimating the parameters θ without assuming any functional form for the distribution function F, and proves that this estimator is consistent. F is also consistently estimated. The method uses maximum likelihood estimation in which the likelihood is maximized not only over the parameter θ but also over a space which contains all distribution functions.
Article
In this paper we are concerned with estimation of a classification model using semiparametric and parametric methods. Benefits and limitations of semiparametric models in general, and of Manski's maximum score method in particular, are discussed. The maximum score method yields consistent estimates under very weak distributional assumptions. The maximum score method can very easily be used in situations where it is more serious to make one kind of classification error than another. In this paper, we use a so-called threshold-crossing model to discriminate between credit card holders and nonholders. The estimated parameters of the logit model differ significantly from the estimates of maximum score. Given an asymmetric loss function, maximum score performs better than the logit model.
Article
In this paper an approach is developed that accommodates heterogeneity in Poisson regression models for count data. The model developed assumes that heterogeneity arises from a distribution of both the intercept and the coefficients of the explanatory variables. We assume that the mixing distribution is discrete, resulting in a finite mixture model formulation. An EM algorithm for estimation is described, and the algorithm is applied to data on customer purchases of books offered through direct mail. Our model is compared empirically to a number of other approaches that deal with heterogeneity in Poisson regression models.
Article
The multivariable techniques most prevalent in direct marketing research today for modeling response to a mailing (responders versus non-responders) are multiple regression, discriminant analysis, and AID (abbreviations are explained in the Introduction). This article shows why these traditional approaches may provide erroneous and misleading results for direct marketing use, which tend to be corrected by newer, more statistically appropriate methods, known as CHAID, logit, and log-linear models. We conclude with an application of the CHAID technique that resulted in a substantial lift in response for an Amoco Oil Company promotion to its credit card file.
Article
A general customer purchase model is presented for segmenting direct marketing customer files for direct mailings. The model specifies the functional relationships among the traditional recency, frequency, and monetary value (RFMV) variables. Because of the generic nature of the model, it can be used by different direct marketing businesses for a variety of different customer mailings. Strategic implications of the model are discussed for: a) defining active versus inactive customers, b) selecting customers for special mailings, c) comparing multiple customer lists for a single company, d) an overall company purchase score for a company with multiple programs/customer files, and e) product purchase scores for special product offers.
Article
This paper reports on the operational characteristics of maximum score estimation of a linear model from binary response data. A series of previous articles have shown that in theory the maximum score method makes possible binary response analysis under very weak distributional assumptions. Here, we present evidence on the properties of maximum score estimation in practice.After reviewing the known asymptotic theory of maximum score estimation, the paper describes an algorithm for maximum score estimation and characterizes its performance. Then findings from a Monte Carlo study comparing maximum score and logit maximum likelihood estimation are reported. Finally, the accuracy of bootstrap estimation of maximum score root mean square errors is evaluated.