Network-based marketing

In their paper Network-based marketing: Identifying likely adopters via consumer networks, Shawndra Hill, Foster Provost and Chris Volinsky argue that there are three, possibly complementary, modes of network-based marketing.

Explicit advocacy: Individuals become vocal advocates for the product or service, recommending it to their friends or acquaintances. Particular individuals such as Oprah, with her monthly book club reading list, may represent “hubs” of advocacy in the consumer relationship network. The success of “The Da Vinci Code,” by Dan Brown, may be due to its initial marketing: ten thousand books were initially delivered free to readers thought to be influential enough (e.g., readers, booksellers) to stimulate the traffic in paid-for editions (Paumgarten 2003). When firms give explicit incentives to consumers to spread information about a product via word of mouth, it has been called viral marketing, although that term could be used to describe any network-based marketing where the pattern of awareness or adoption spreads from consumer to consumer.

Implicit advocacy: Even if individuals do not speak about a product, they may advocate implicitly through their actions—especially through their own adoption of the product. Designer labeling has a long tradition of using consumers as implicit advocates. Firms commonly capitalize on influential individuals (such as athletes) to advocate products simply by conspicuous adoption. More recently, firms have tried to induce the same effect by convincing particularly “cool” members of smaller social groups to adopt products (Gladwell 1997; Baker 2005).

Network targeting: The third mode of network-based marketing is for the firm to market to prior purchasers’ social-network neighbors, possibly without any advocacy at all by customers. For network targeting, the firm must have some means of identifying these social neighbors.

 

Network-based marketing assumes interdependency among consumer preferences. When interdependencies exist, their effects should be accounted for in the models. Studies in network-based marketing attempt to measure these interdependencies through implicit links, such as matching on geographic or demographic variables, or through explicit links, such as direct observation of communications between actors.

Work in network-based marketing, spans the fields of statistics, economics, computer science, sociology, psychology and marketing. In this section, we organize the prominent work in network based marketing by five types of statistical research: 1) econometric modeling; 2) network classification modeling; 3) surveys; 4) designed experiments with convenience samples; 5) diffusion theory; and 6) collaborative filtering and recommender systems.

Econometrics Models

Econometrics is the application of statistical methods to the empirical estimation of economic relationships. In marketing this often means the estimation of two simultaneous equations: one for the marketing organization or firm and one for the market. Regression and time-series analysis are found at the core of econometric modeling, so econometric models are often used to assess the impact of a target marketing campaign over time.

Network classification models

Network classification models use knowledge of the links between entities in a network to estimate a quantity of interest for those entities. Typically, in such a model an entity is influenced most by those directly connected to it, but is also affected to a lesser extent by those further away. Some network classification models use an entire network to make predictions about a particular entity on the network (Macskassy and Provost 2004). However, most methods have been applied to small datasets, and have not been applied to consumer data.

One example that models a consumer network for maximizing profit is (Domingos 2005), where a social network of customers is modeled as a Markov random field. They model the probability that a given customer will buy a given product as a function of the states of her neighbors, attributes of the product, and whether or not the customer was marketed to. In this framework they are able to assign a “network value” to every customer, by estimating the overall benefit of marketing to that customer, including the impact that the marketing action will have on the rest of the network (for example, through word of mouth). The authors test their model on a database of movie reviews from an Internet site, and find their proposed methodology outperforms non-network methods for estimating customer value. Their network formulation uses implicit links (customers are linked when a customer reads review by another customer and subsequently reviews the item herself) and implicit purchase information (they assume a review of an item implies a purchase, and vice versa).

Surveys

Most research in this area does not have information on whether or not consumers actually talked to each other. To address this shortcoming, some studies use survey sampling to collect comprehensive data on consumers’ word-of-mouth behavior. By sampling individuals and contacting them, researchers can collect data that is difficult (or impossible) to obtain directly by observing network-based marketing phenomena (Bowman and Narayandas 2001). The strength of these studies lies in the data themselves and the richness and flexibility of the answers that can be collected from the responders. For instance, researchers can acquire data on how customers found out about a product, and how many others they told about the product. An advantage is that researchers can design their sampling scheme to control for any known confounding factors, and devise fully balanced experimental designs that test their hypotheses. Since the purpose of models built from survey data is description, simple statistical methods like logistic regression or ANOVA typically are used. (Bowman and Narayandas 2001) surveyed more than 1700 purchasers of 60 different products who previously had contacted the manufacturer of that product. The purchasers were asked specific questions about their interaction with the manufacturer and its impact on subsequent word-of-mouth behavior. The authors were able to capture whether the customers told others of their experience and if so, how many people they told. The authors find that self-reportedly “loyal” customers are more likely to participate in word-of-mouth when they are dissatisfied, but interestingly not more likely when they are satisfied. Although studies like this one collect consumers’ word-of-mouth behavior, the researchers do not know which of the consumers’ contacts later purchased the product. Therefore, they do not address whether word-of-mouth actually affects individual sales.

Diffusion Models

Diffusion theory provides tools, both quantitative and qualitative, for assessing the likely rate of diffusion of a technology or product, and additionally identifies numerous factors that facilitate or hinder technology adoption (Fichman 2004) The social factors that influence product adoption are discussed in detail in prior research (Rogers 2003) In general, the quantitative diffusion research involves the empirical testing of predictions from a given diffusion model, which is often informed by economic theory.

The most notable and most influential diffusion model was proposed by (Bass 1969; Dodds 1973; Ueda 1990; Bass, Krishnan et al. 1994; Evans 1995; Satoh 2001; Niu 2002). The Bass model of product diffusion predicts the number of users that will adopt an innovation at a given time t. It hypothesizes that the rate of adoption is a function solely of the current proportion of the population having adopted. Specifically, let F(t) be the cumulative proportion of adopters in the population. The diffusion equation, in its simplest form models F(t) as a function of p, the intrinsic adoption rate, and q, a measure of social contagion. When q > p, this equation describes an S-shaped curve, where adoption is slow at first, takes off exponentially, and tails off at the end.

This model can effectively model word-of-mouth product diffusion, but only at the aggregate, societal level. In the first Bass study, the model is tested empirically against data for eleven consumer durables. Bass finds that the model yields good predictions of the sales peak and the timing of the peak when applied to historical data. Bass utilizes linear regression to estimate the parameters for future sales predictions. He measures the forecasting accuracy (R2 value) of the model for eleven consumer durable products. The success of the forecasts suggests that the model may be useful in providing a long-range forecasting for product sales or adoption. There has been considerable follow-up work on diffusion since this groundbreaking work. Recent work on product diffusion explores the extent to which the Internet (Fildes 2003) as well as globalization (Kumar and Krishnan 2002) play a role in product diffusion. (Mahajan, Muller et al. 1984) review this work. In general, the empirical studies that test and extend accepted theories of product diffusion rely on aggregate-level data for both the customer attributes and overall adoption of the product (Ueda 1990; Evans 1995). They typically do not incorporate individual adoption. Models of product diffusion assume that network-based marketing is effective. Since the understanding of when it occurs and to what extent it is effective is most important for marketers, these methods may benefit from using individual level data. Data on explicit networks would enable the extension of existing diffusion models, as well as the comparison individual vs. aggregate level data.

Collaborative filtering and recommender systems

The purpose of collaborative filtering systems is to automate the process of "word-of-mouth" by which people recommend products or services to one another. Collaborative filtering systems make personal recommendations to individual consumers based on the purchase preferences of the consumers most similar to them. Collaborative filtering marketing techniques involve associating customers primarily based on descriptive data or transaction data.

Collaborative filtering models enable the exploitation of information in the links between customers. Huang, Chung et al. (Huang, Chung et al. 2004) establish the connection between the recommendation problem and Statistical Relational Learning (Getoor 2005), through the application of a recently developed statistical relational learning method called Probabilistic Relational Models (PRMs) (Getoor, Friedman et al. 2001). Huang et al. use purchase data from a large Internet company, which has individual level response data but no explicit interaction between customers. Likewise, (Newton and Greiner 2004 ) applied PRMs to the collaborative filtering task. They use recommendations as their links between customers and whether or not the customer sees the recommended movie as the target. Huang, Terry et al. (Huang, Chung et al. 2004) use physical proximity between mobile devices to help users filter incoming information and determine its relevance. Although collaborative filtering is related to explicit consumer network-based marketing, the goal is to automate recommendations as opposed to studying the effect of explicit consumer behavior for the purpose of better targeting for a single product. We suspect firms that use recommendation systems could benefit from the additional link of explicit consumer interaction. This would allow the collaborative filtering to include one additional, perhaps quite important, aspect of similarity.

© 2017-2018 Intelligencia Limited. All Rights Reserved.

Contact

MACAU:

Rua da Esrela, No. 8, Macau

Macau: +853 6616 1033

 

HONG KONG:

505 Hennessy Road, #613, Causeway Bay

HK: +852 5196 1277

andrew.pearson@intelligencia.co