Explore the vast range of titles available.

Firms using online advertising regularly run experiments with multiple versions of their ads since they are uncertain about which ones are most effective. During a campaign, firms try to adapt to intermediate results of their tests, optimizing what they earn while learning about their ads. Yet how should they decide what percentage of impressions to allocate to each ad? This paper answers that question, resolving the well-known “learn-and-earn” trade-off using multi-armed bandit (MAB) methods. The online advertiser’s MAB problem, however, contains particular challenges, such as a hierarchical structure (ads within a website), attributes of actions (creative elements of an ad), and batched decisions (millions of impressions at a time), that are not fully accommodated by existing MAB methods. Our approach captures how the impact of observable ad attributes on ad effectiveness differs by website in unobserved ways, and our policy generates allocations of impressions that can be used in practice. We implemented this policy in a live field experiment delivering over 750 million ad impressions in an online display campaign with a large retail bank. Over the course of two months, our policy achieved an 8% improvement in the customer acquisition rate, relative to a control policy, without any additional costs to the bank. Beyond the actual experiment, we performed counterfactual simulations to evaluate a range of alternative model specifications and allocation rules in MAB policies. Finally, we show that customer acquisition would decrease by about 10% if the firm were to optimize click-through rates instead of conversion directly, a finding that has implications for understanding the marketing funnel. Data is available at https://doi.org/10.1287/mksc.2016.1023 .

There is growing interest in \"customer-based corporate valuation\"—that is, explicitly tying the value of a firm's customer base to its overall financial valuation using publicly disclosed data. While much progress has been made in building a well-validated customer-based valuation model for contractual (or subscription-based) firms, there has been little progress for noncontractual firms. Noncontractual businesses have more complex transactional patterns because customer churn is not observed, and customer purchase timing and spend amounts are more irregular. Furthermore, publicly disclosed data are aggregated over time and across customers, are often censored, and may vary from firm to firm, making it harder to estimate models for customer acquisition, ordering, and spend per order. The authors develop a general customer-based valuation methodology for noncontractual firms that accounts for these issues. They apply this methodology to publicly disclosed data from e-commerce retailers Overstock.com and Wayfair, provide valuation point estimates and valuation intervals for the firms, and compare the unit economics of newly acquired customers.

We examine three sets of established behavioral hypotheses about consumers’ in‐store behavior using field data on grocery store shopping paths and purchases. Our results provide field evidence for the following empirical regularities. First, as consumers spend more time in the store, they become more purposeful—they are less likely to spend time on exploration and more likely to shop/buy. Second, consistent with “licensing” behavior, after purchasing virtue categories, consumers are more likely to shop at locations that carry vice categories. Third, the presence of other shoppers attracts consumers toward a store zone but reduces consumers’ tendency to shop there.

This paper develops a model of conversion behavior (i.e., converting store visits into purchases) that predicts each customer's probability of purchasing based on an observed history of visits and purchases. We offer an individual-level probability model that allows for different forms of customer heterogeneity in a very flexible manner. Specifically, we decompose an individual's conversion behavior into two components: one for accumulating visit effects and another for purchasing threshold effects. Each component is allowed to vary across households as well as over time. Visit effects capture the notion that store visits can play different roles in the purchasing process. For example, some visits are motivated by planned purchases, while others are associated with hedonic browsing (akin to window shopping); our model is able to accommodate these (and several other) types of visit-purchase relationships in a logical, parsimonious manner. The purchasing threshold captures the psychological resistance to online purchasing that may grow or shrink as a customer gains more experience with the purchasing process at a given website. We test different versions of the model that vary in the complexity of these two key components and also compare our general framework with popular alternatives such as logistic regression. We find that the proposed model offers excellent statistical properties, including its performance in a holdout validation sample, and also provides useful managerial diagnostics about the patterns underlying online buyer behavior.

The authors present a new model that links the well-known RFM (recency, frequency, and monetary value) paradigm with customer lifetime value (CLV). Although previous researchers have made a conceptual link, none has presented a formal model with a well-grounded behavioral \"story.\" Key to this analysis is the notion of \"iso-value\" curves, which enable the grouping of individual customers who have different purchasing histories but similar future valuations. Iso-value curves make it easy to visualize the interactions and trade-offs among the RFM measures and CLV. The stochastic model is based on the Pareto/NBD framework to capture the flow of transactions over time and a gamma-gamma submodel for spend per transaction. The authors conduct several holdout tests to demonstrate the validity of the model's underlying components and then use it to estimate the total CLV for a cohort of new customers of the online music site CDNOW. Finally, the authors discuss broader issues and opportunities in the application of this model in actual practice.

Many data sets, from different and seemingly unrelated marketing domains, all involve paths —records of consumers' movements in a spatial configuration. Path data contain valuable information for marketing researchers because they describe how consumers interact with their environment and make dynamic choices. As data collection technologies improve and researchers continue to ask deeper questions about consumers' motivations and behaviors, path data sets will become more common and will play a more central role in marketing research. To guide future research in this area, we review the previous literature, propose a formal definition of a path (in a marketing context), and derive a unifying framework that allows us to classify different kinds of paths. We identify and discuss two primary dimensions (characteristics of the spatial configuration and the agent) as well as six underlying subdimensions. Based on this framework, we cover a range of important operational issues that should be taken into account as researchers begin to build formal models of path-related phenomena. We close with a brief look into the future of path-based models, and a call for researchers to address some of these emerging issues.

Multiservice providers, such as telecommunication and financial service companies, can benefit from understanding how customers' service portfolios evolve over the course of their relationships. This can provide guidance for managerial issues such as customer valuation and predicting customers' future behavior, whether it is acquiring additional services, selectively dropping current services, or ending the relationship entirely. In this research, we develop a dynamic hidden Markov model to identify latent states that govern customers' affinity for the available services through which customers evolve. In addition, we incorporate and demonstrate the importance of separating two other sources of dynamics: portfolio inertia and service stickiness. We then examine the relationship between state membership and managerially relevant metrics, including customers' propensities for acquiring additional services or terminating the relationship, and customer lifetime value. Through a series of illustrative vignettes, we show that customers who have discarded a particular service may have an increased risk of canceling all services in the near future (as intuition would suggest) but also may be more prone to acquire more services, a provocative finding of interest to service providers. Our findings also emphasize the need to look beyond the previous period, as in much current research, and consider how customers have evolved over their entire relationship in order to predict their future actions. This paper was accepted by Pradeep Chintagunta, marketing.

This paper examines search across competing e-commerce sites. By analyzing panel data from over 10,000 Internet households and three commodity-like products (books, compact discs (CDs), and air travel services), we show that the amount of online search is actually quite limited. On average, households visit only 1.2 book sites, 1.3 CD sites, and 1.8 travel sites during a typical active month in each category. Using probabilistic models, we characterize search behavior at the individual level in terms of (1) depth of search, (2) dynamics of search, and (3) activity of search. We model an individual's tendency to search as a logarithmic process, finding that shoppers search across very few sites in a given shopping month. We extend the logarithmic model of search to allow for time-varying dynamics that may cause the consumer to evolve and, perhaps, learn to search over time. We find that for two of the three product categories studied, search propensity does not change from month to month. However, in the third product category we find mild evidence of time-varying dynamics, where search decreases over time from already low levels. Finally, we model the level of a household's shopping activity and integrate it into our model of search. The results suggest that more-active online shoppers tend also to search across more sites. This consumer characteristic largely drives the dynamics of search that can easily be mistaken as increases from experience at the individual level.

Today’s managers are very interested in predicting the future purchasing patterns of their customers, which can then serve as an input into \"lifetime value\" calculations. Among the models that provide such capabilities, the Pareto/NBD \"counting your customers\" framework proposed by Schmittlein et al. (1987) is highly regarded. However, despite the respect it has earned, it has proven to be a difficult model to implement, particularly because of computational challenges associated with parameter estimation. We develop a new model, the beta-geometric/NBD (BG/NBD), which represents a slight variation in the behavioral \"story\" associated with the Pareto/NBD but is vastly easier to implement. We show, for instance, how its parameters can be obtained quite easily in Microsoft Excel. The two models yield very similar results in a wide variety of purchasing environments, leading us to suggest that the BG/NBD could be viewed as an attractive alternative to the Pareto/NBD in most applications.

Several researchers have proposed models of buyer behavior in noncontractual settings that assume that customers are \"alive\" for some period of time and then become permanently inactive. The best-known such model is the Pareto/NBD, which assumes that customer attrition (dropout or \"death\") can occur at any point in calendar time. A recent alternative model, the BG/NBD, assumes that customer attrition follows a Bernoulli \"coin-flipping\" process that occurs in \"transaction time\" (i.e., after every purchase occasion). Although the modification results in a model that is much easier to implement, it means that heavy buyers have more opportunities to \"die.\" In this paper, we develop a model with a discrete-time dropout process tied to calendar time. Specifically, we assume that every customer periodically \"flips a coin\" to determine whether she \"drops out\" or continues as a customer. For the component of purchasing while alive, we maintain the assumptions of the Pareto/NBD and BG/NBD models. This periodic death opportunity (PDO) model allows us to take a closer look at how assumptions about customer death influence model fit and various metrics typically used by managers to characterize a cohort of customers. When the time period after which each customer makes her dropout decision (which we call period length ) is very small, we show analytically that the PDO model reduces to the Pareto/NBD. When the period length is longer than the calibration period, the dropout process is \"shut off,\" and the PDO model collapses to the negative binomial distribution (NBD) model. By systematically varying the period length between these limits, we can explore the full spectrum of models between the \"continuous-time-death\" Pareto/NBD and the naïve \"no-death\" NBD. In covering this spectrum, the PDO model performs at least as well as either of these models; our empirical analysis demonstrates the superior performance of the PDO model on two data sets. We also show that the different models provide significantly different estimates of both purchasing-related and death-related metrics for both data sets, and these differences can be quite dramatic for the death-related metrics. As more researchers and managers make managerial judgments that directly relate to the death process, we assert that the model employed to generate these metrics should be chosen carefully.