Table of Contents
2 Literature Review
2.2 Recommender Systems
2.3 Product categories and product familiarity
3.1 Research Model and Hypothesis Development
3.2 The effect between different types of recommender systems and cross-selling
3.3 The moderation effect of product categories and product familiarity
4 Research Method
4.1 Experimental Design
4.2 Experimental Task and Procedures
4.3 Measurement of Dependent Variable
The marketplace of the e-commerce is shaped by intense competition. The development of the internet makes it easier for sellers to display their products to the customer. However, it is getting more difficult for the sellers to get fully noticed by the consumers, as the online marketplace is becoming more and more crowded of competing sellers (Fong, 2017). Under this situation, online recommender systems are often used to recommend lots of products to the customers and guide them through the overwhelming range of offers on the online marketplace. Thus, they help them to decide which product to purchase.
One of the advantages of the online recommender systems is that it can reduce search costs (Pathak et al., 2014). These reduced costs are a result of the e-commerce websites’ possibility to use big data and personal analysis to provide content which is tailored to the unique customers’ needs (Ansari and Mela, 2003). Consumers are likely to value the reduced searching costs (Lynch and Ariely, 2000) as it is stated that liking role of advertisement is to reduce the cost of search by informing on where they can search and find products (Stiegler, 1961). Since customers also reacting positively to a greater flexibility in search (Ariely, 1999), many e-commerce websites, such as Amazon, eBay, Taobao and Jingdong have been absorbed in making full use of these capabilities to provide product information which can satisfy the needs of consumers.
In fact, this will result in more flow importing as well as the more potential sales for the sellers’ platform. Online recommender systems can also have an impact on the cross-selling of an online retailer since they can be designed to foster cross-selling (Pathak et al., 2014; Schafer, 1999). The practice of cross-selling describes the selling of an additional product or service to a customer and sometimes making single-product buyer to multi-product ones (Li, Sun and Montgomery, 2011; Kamakura, 2008). The cross-selling strategies mainly reflect in two aspects. One is by introducing the consumers from the single products into the whole store or from the online into the offline store, another is with the form of co-purchase links (Oestreicher-Singer and Sundararajan, 2012). Simply speaking, cross-selling means that customers consume products or services that they did not plan to purchase before. That is to say, cross-selling will bring extra sales.
In this study, we aim to investigate how two different types of online recommender systems affect the cross-selling of a retailer on a website using the online recommender systems. Furthermore, this will give us the chance to study the direct effect of targeting on cross-selling. Different from the previous studies, this study will mainly focus on two aspects.
On the one hand, the online recommender systems will be divided into two types, the targeting recommender systems and popular recommender systems. The first one is the personalized form through which we can give the targeted consumers the specific products recommenders based on the consumers’ purchase history. This type has been fully discussed in previous studies. The second type of a recommender system is the public form giving the recommendations based on the hot products, which is a common phenomenon in e-commerce platforms. In other words, recommended products or services come from common preferences of all consumers. Research on the relationship between cross-selling and this type of recommender system are still relatively lacking.
On the other hand, we will discuss the different moderating effects between two different types of the recommended products — search-type and experience-type products — as well as the influence caused by product familiarity on the relationship between the online recommender systems and cross-selling. These two types of products have been fully discussed in the area about the helpfulness of online reviews (Mudambi and Schuff, 2010; Pan and Zhang, 2011; Scholz and Dorner, 2013). In addition, familiar and unfamiliar products differ in terms of the knowledge regarding the products that a consumer has stored in memory. This will affect how consumers search and process the online recommender systems and co-purchase information. Therefore, we will consider these in the process of data collecting.
Through examining those previously described causal effects, we can make the following two contributions. Firstly, we can make further suggestions about how the choice for an online recommender system can influence cross-selling and thereby further contribute to the discussion about recommender systems in the e-commerce ecosystem. Secondly, we can classify the cross influence from the product types and product familiarity in the above stated relationship between online recommender systems and cross-selling.
2 Literature Review
In this section, we will review the literature about the recommender systems, cross-selling, product categories and product familiarity, we will further find something new to investigate.
The definition of cross-selling in our paper is the selling of one or more than one additional items to the customer, which he or she does not plan to purchase before, regardless of the form of purchase it is implementing (Li, 2011).
Cross-selling can have significant beneficial or negative outcomes for both sellers and customers. To the sellers, it can give more growth opportunities to get stronger and longer ties with the customers and thereby reduce churn (Kamakura, 2008; Kamakura et al., 2003). Additionally, cross-selling could also improve potential sales for the sellers as well as for the platforms (Pathak et al., 2014). Likewise, studies have shown that recommender systems can help to accelerate cross-selling, improve the cross-selling order size and also increase the generated profits from cross-selling without losing accuracy of the recommendation quality (Chen et al., 2008; Schafer et al., 2001).
Nevertheless, cross-selling can have negative effects if customers already have a large share of wallet to the firm as they have formed the fixed consumption structure (Kamakura, 2008). Moreover, promotions performed by the sellers to enhance cross-selling can be ineffective or even negative (Liu-Thompkins and Tam, 2013).
Therefore, the relationship between the recommender systems and cross-selling is not uniform. In our study, we tried to find the main factors and boundary conditions affecting the relationship between these two, in order to supplement the deficiencies in previous studies.
2.2 Recommender Systems
Online product recommendation systems usually use a utility model based on the attributes of the products and consumers’ preference to give appropriate suggestions to consumers. Through some recommender information, it can act as a decision support tool to help consumers make better decisions in online shopping environments (Xiao and Benbasat, 2007; Grenci and Todd 2002). In general, there are mainly two recommender systems entioned in the previous studies: collaborative (social) filtering and content-based/attribute-based systems (Adomavicius and Tuzhilin, 2005; Xiao and Benbasat, 2007).
Collaborative filtering recommender systems pay more attention to the social approach which is rooted in other people’s preferences, rather than the characteristics of other products (Adomavicius and Tuzhilin, 2005; Xiao and Benbasat, 2007). Through finding the relationship between new users and existing users and determining the similarities in preferences, the website can recommend something useful to newcomers (Adomavicius and Tuzhilin, 2005). In other words, the more similar a newcomer is to existing users, the more likely the system will recommend products liked by old users to newcomers (Degemmis et al., 2004). Content-based/attribute-based recommender systems recommend products to users through processing users’ profiles and product information (Adomavicius and Tuzhilin, 2005; Xiao and Benbasat, 2007). For example, when a Netflix user watches many comedies, the recommender system may recommend a movie labeled comedy to the user. Content-based/attribute-based recommender systems recommend products with similar characteristics as the users’ previous purchasing or watching history (Adomavicius and Tuzhilin, 2005).
The initiative purpose of the product recommendation was to help consumers to make better choices in the context of information overload, especially in online environments. Previous studies related to the impact of recommender systems indicate that consumers with the aid of recommendation systems can complete their shopping tasks in less time than consumers without the aid of recommendation systems (Hostler et al., 2005; Vijayasarathy and Jones, 2001). The recommendation systems can recommend the proper products to consumers given their preferences, which could reduce the search cost and improve the quality of decision made by consumers (Montgomery et al., 2004, Xiao and Benbasat, 2007; Ansari and Mela, 2003). However, consumers’ belief about the degree to which recommendation can satisfy their personalized needs is the essential factor in the recommendations’ success (Komiak and Benbasat, 2006). Furthermore, product recommendation can also contribute to strengthening the long-tail phenomenon of electronic commerce. In addition, it will have a positive effect on reducing the consumers’ sensitivity on price, which can reinforce the flexibility for retailers to adjust their price (Pathak et al., 2014).
Online recommender systems can also be designed to foster cross-selling for an online retailer (Pathak et al., 2014; Schafer et al., 2004). However, there have been concerns that online product recommendations may benefit consumers at the cost of the merchants’ interests. On the one hand, some scholars argue that product recommendation can bring negative effect on the sales diversity (Ahn, 2006; Fleder and Hosanagar, 2009; Mooney and Roy, 2000). On the other hand, some scholars think that consumers may find more diverse products because of the reduced cost (Toine and Antal, 2011; Brynjolfsson and Simester, 2011; Eriket al., 2006). In the previous studies, most of the proposed recommender systems were more or less related to the consumers themselves, such as similar search behaviors or purchase history. In this study, we divided the recommendation system into two types, one is the targeting recommendation based on the consumer’s purchase or browsing history, and the other is the popular recommendation for current popular products based on the common preferences of all consumers.
2.3 Product categories and product familiarity
Many research areas have focused on the important effect of the product categories, such as consumer information search behavior (Bei et al., 2004), product uncertainty perception (Weather et al., 2007), inconsistent online reviews (Zhang et al., 2014), comments usefulness (Huang et al., 2013) and so on. Specifically, based on the information search cost from Information Economics, Nelson (1970) divided the product categories into search-type, experience-type and credence-type products. As credence-type products are less common and unusual in the online background, we will focus on search-type and experience-type products like other previous studies. Consumers’ judgments on search-type products are cognitive-driven, instrumental and goal-oriented (Strahilevitz and Meyers, 1998), while on the contrary, consumers’ judgments on experience-type products depends on aesthetic emotional experience and sensory pleasure (Holbrook, 1978). Some studies point out that product category is the most important factor affecting consumer search costs. Consumers can use multiple ways of information search and functional judgment to easily obtain the quality perception of search-type products, while experience-type products require more personal experience (Mudambi, 2010). In addition, when the experience costs of a product are high, information search is relatively more important. On the contrary, when information search costs are high or the information value obtained from search is very limited, experience is more important (Nelson, 1974). Therefore, this study believes that in the context of product recommendation, different types of recommended products will have a significant impact on consumers' decisions related to cross-selling. There is still a relatively lack in investigating the above stated impact in the existing literature.
Product familiarity reflects the extent of a consumers’ direct or indirect experience with a product. (Alba and Hutchinson, 1987; Kent and Allen, 1994). Although many marketing products are familiar to consumers, many others may be unfamiliar, either, because they are new in the market or because consumers have not yet been exposed to them (Stewart, 1992). To the familiar products, consumers may soon know how the products are positioned, packaged as well as how the product quality is, while to the unfamiliar products, consumers lack many associations for them because they have not had any of these types of experiences with them (Campbell and Keller, 2003). Thus, when faced with product recommendations of different familiarities, consumers may have different processing results, especially in the cross-selling environment. This effect is worth further study.