From the very beginning of the development of business-to-consumer (B2C) e-commerce, comparison-shopping has been a popular activity among consumers due to lower search costs in the online environment. However, the potentially unlimited availability of choice alternatives in online shopping has also made comparing these alternatives and selecting a most preferred product a daunting task for consumers. As the volume of B2C e-commerce increases further, it is very likely that the efficiency and effectiveness of comparison-shopping will be a major bottleneck in the effort to achieve customer satisfaction [2].
In the last decade, a large number and variety of product comparison agents have emerged to assist consumers making choice decisions in online environments. The earliest of this breed that gained wide popularity was the bargainfinder introduced by Andersen Consulting in 1995. It allowed a shopper to enter the name of a desired music CD and the software would then search for the selection at nine online stores and return the price list to the shopper. Subsequently, many other types of comparison agents have emerged, such as pricescan.com, mysimon.com, epinion.com, and bizrate.com, operating in very different ways to assist consumers in the same task.
Many researchers have contributed valuable insights about the impact of product comparison agents. However, we do not yet have a formal classification system for identifying, naming, and grouping these agents that often operate in different ways. For example, there is a lot of confusion about even the names of these agents: essentially the same product comparison agent is often referred to as shopbot, comparison-shopping agent, recommendation agent, buyer’s agent as well as aggregator [24]. The issue of providing a comprehensive and consistent taxonomy is important to both industry experts and academic researchers dealing with product comparison agents. For example, by identifying the operational mechanism of various comparison-shopping agents and matching it to the cognitive processes underlying consumer decision making in online settings, we may be able to find new opportunities for providing improved comparison-shopping services.
In this article, we use a basic classification concept in biology—the ecological food chain—as an analogy to classify these agents. If we regard the Internet as an ecosystem that includes online shoppers, vendors, and various service agents, then comparison-shopping agents fit very well in this ecosystem as a class of herbivores that feed on plants (original data producers like online vendors) and in turn, become the target of carnivores (upstream agents or meta-agents). Sitting on the top of the information food chain are the shopping demands of online shoppers, and their click-stream will in turn be the information “food” for online vendors. Together, these participants—the shoppers, vendors, comparison-shopping agents and upstream agents constitute the complete ecosystem of e-commerce world, as depicted in Figure 1.
We propose a conceptual scheme for classifying comparison-shopping agents into different groups. Essentially, comparison-shopping agents emerged and evolved as a response to the demand for assistance in decision making by online shoppers. Consumers may desire varying types of assistance depending upon the specific cognitive processes that they employ during the task of comparison. Hence, our classification scheme is based on the nature of specialized assistance that agents provide for the various cognitive processes involved in comparison-based decision making.
For simplification of discussion, we use PCA (acronym for Product Comparison Agent) to refer to the entire class of comparison agents. Further, for clarification, we formally define PCA as Web-based online services that can retrieve, aggregate, and process product and/or service information from heterogeneous online data sources as well as online consumers and present it in an appropriate format to online shoppers for choice-related decision making.
Anatomy of Comparison and Classification of PCAs
Just as biologists group organisms according to their shared characteristics, we must first identify the shared characteristics of PCAs in order to classify them. To do so, we focus on the characteristics or “the anatomy” of comparison. Researchers have suggested that comparison is a cognitive process that involves two modes of thinking: associative reasoning and rule-based reasoning [5]. Associative reasoning operates by grouping objects with other objects that are most similar. It is related to subjective experience and operates at a low level of consciousness. Rule-based reasoning operates using abstract symbols based on logical rules that have a certain well-specified symbolic logical structure. Past research suggests that these two modes of thinking dominate at different stages of thinking or may work simultaneously.
The goal of a rational comparison process is to pick the choice that maximizes consumer utility, based on preferences. For example, if a consumer wants to choose between two car models, he will first find out the differences between them on various aspects such as price, reputation, after-sale service, and other factors. Then by selecting those differences that matter most to him, he will evaluate these differences and combine them into an aggregated utility for comparison. These differentiation and evaluation processes may be repeated for other alternatives in the choice set, until he feels that he can make a decision based on the current evaluation result. Often, he may not find a dominant choice if he has to trade off between attributes. Then, he may have to identify a subjective preference to proceed, say, that comfort is more important than price. And based on that preference, he can make a decision.
Specifically, when consumers compare two or more items, they aim to differentiate them and to evaluate the differences based on certain preferences. These tasks in the process of comparison use aspects of both associative reasoning and rule-based reasoning. We refer to the PCAs that assist consumers in each of these specific tasks as product differentiation PCA, product evaluation PCA, and consumer preference identification PCA respectively.
Differentiation is essentially a cognitive process that is dominated by rule-based reasoning. It abstracts the products in the choice set into symbols such as price, weight, and quality. It then operates on these symbols according to specific rules, such as “if all other attributes are the same, the lower-priced object should be selected.” Many PCAs specialize in collecting product attribute information from different vendors, aggregating the data and then presenting it in a tabular format to the consumer. We refer to such PCAs that assist consumers in attribute-based differentiation of choice alternatives as product differentiation PCA. A good example of this category would be pricescan.com. In general, such PCAs are useful when consumer choice in the product category is determined by concrete, quantifiable attributes such as price, physical dimensions, warranty period, or other performance-related characteristics.
Evaluation is a mixed process involving both objective and subjective information, so that different decision makers may have different evaluations of the same product. Consider restaurants as a category; consumers make choices based on objective dimensions such as price range and location as well as subjective dimensions such as service level, food quality, and presentation. An evaluation outcome can be drawn either from our past experience directly (by associative reasoning) or from quantified information about the product (by rule-based reasoning) or from a mix of these two strategies. Product evaluation PCAs collect objective data (such as price of product) and quantified subjective evaluation data (such as quality of service) from consumers, aggregate it, and then provide it to other consumers. BizRate.com is a prominent agent of this type.
Consumer preference identification is mainly achieved through associative reasoning. Often, consumers may need usage context or scenario information (such as “a PC for my home office” or “a wine to serve at Thanksgiving dinner”) rather than product information per se in order to identify a preferred choice. In some instances, consumers can make an association between current context and their own past experiences, and make a decision based on associated preferences. However, for new products or services, or unfamiliar contexts, consumers may need to rely on peers’ experiences and preferences, and they may identify preferences based on similarity of context with others. Consumer preference identification PCAs (such as Epinion.com) can help in this situation. They collect product usage experience from peers. The information collected is not numeric but text-based descriptive information. This enables them to provide different situational contexts for the comparison and evaluation of products, instead of comparing products per se. Then consumers can use peer experiences as surrogates to identify their preferences and make choice decisions. This is particularly relevant for services (such as hotels, restaurants, concerts, cruises) where preferences can vary significantly across situational contexts (including occasions, seasons, and goals). The table here summarizes the key aspects of the three categories of PCA.
An Empirical Investigation
In order to relate this classification scheme to the PCAs that are operational at the current time, we collected PCA information from directory and agent portals such as Yahoo directory, SearchEngineWatch.com, and pricingcentral.com from late September to early October in 2003. This exercise was repeated in October 2005 to validate previous entries and identify new entries. We identified a total of 90 PCAs by October 2005 that conform to our qualification criteria: Web-based; collect data from heterogeneous online sources including consumers; and provide data to consumers for decision making. Of these, 77 are Differentiation PCAs, two are Evaluation PCAs and three are Preference Identification PCAs (see Figure 2a) based on our classification scheme. We found eight PCAs that do not belong to any single category described previously. They are “composite” PCAs in the sense that their functions can be mapped to multiple categories and belong to what we called derivative PCAs, which we will discuss in more detail later.
Differentiation PCA Dominates
The majority of the PCAs in existence today are differentiation PCAs. This is not surprising because the advancement of searching technology makes it extremely easy to retrieve quantitative product attribute information from heterogeneous data sources on the Internet [2]. As a result, differentiation PCAs proliferate in every consumer-focused industry, including books, video, audio, electronics, computers, automotives, insurance, or personal finance.
We further categorized the 77 differentiation PCAs based on their industry focus. We found that approximately 35% of them offer a comparison service across two or more industries, and we refer to these as “consolidated PCAs.” We noted that even though academic researchers seem to have focused all their attention on PCAs in the books and entertainment industry, they make up only 13% of the differentiation PCA population. Perhaps future research and attention should also be invested in other differentiation PCAs, such as those in insurance (9% share of PCAs) and the travel and leisure industries (14% of PCAs).
Online drug prescription was once a controversial topic in e-commerce due to its privacy and feasibility concerns. If we can use the existence of comparison-shopping tools such as differentiation PCAs as indicative of the maturity of an online industry, then we have to say this industry shows signs of vitality. We found two differentiation PCAs focusing on this industry that had emerged in 1999 (Price-rx.com) and 2000 (Destinationrx.com), and continue to operate today.
PCA Evolution
By observing the launch year of the PCAs in these three categories, we found that product differentiation PCA is the first category that emerged on the Web. Though it is widely believed the first comparison-shopping agent is the bargainfinder developed by Andersen Consulting in 1995, our investigation found two other independent PCAs that were already operational in 1995—Pricewatch.com and Netquote.com; both remain active today. Considering that other differentiation PCAs may have ceased to exist after the crash of the Internet bubble in 2000, the earliest comparison-shopping agent may well have began before 1995.
The first product evaluation PCA (Bizrate.com) emerged in 1996 and the first consumer preference identification PCA (Epinions.com) emerged in 1999. Here we can see a clear trend of the need for assistance at increasingly complex levels of cognitive processing from online shoppers: consumers are not satisfied with merely differentiating between products now; they need evaluation information in some cases and a usage context to anchor their preferences in other cases. This trend is also confirmed by the evolution of PCA functions. Before 2000, few product differentiation PCAs provided vendor evaluation information in their product comparison tables. However, since 2000, almost all surviving PCAs include vendor-rating information in their product comparison lists. The PCA evolutionary trend is clearly shifting from a product focus to a consumer-centric focus.
Will we experience the emergence of another wave of PCAs? With the growth of B2C e-commerce in the past several years, we can expect the online availability of products and services will increase dramatically. In this context, online shoppers will need even more assistance from PCAs in their decision making. The first wave of emergence of PCA reached its peak in 1999 as indicated in Figure 2a. With the saturation of comparison-shopping service in commodity market, a second wave seems to be already well on its way with more ubiquitous services. For example, in 2005, several highly specialized comparison-shopping services were operating: a public-funded PCA for hospital services sponsored by the State of Massachusetts and a comparison-shopping site specializing in dental products (www.net32.com); another is a PCA for virtual goods used in online virtual games (www.eyeonmogs.com).
The distinctive feature of a derivative PCA is its meta-searching nature of aggregating comparison-shopping information from other PCAs instead of online vendors or shoppers.
More importantly, a new type of PCA, what we call the derivative PCA, has emerged in the last few years. We expect this type to become a dominant trend in future, and is therefore, particularly noteworthy.
Emergence of Derivative PCA
We define derivative PCAs as comparison-shopping agents that collect information from other PCAs and aggregate them for use. A derivative PCA could be a composite PCA by aggregating multiple categories of comparison information or could be a differentiation PCA that only provides differentiation information (for example, Kayak.com only collects airfare from other PCAs.) The distinctive feature of a derivative PCA is its meta-searching nature of aggregating comparison-shopping information from other PCAs instead of online vendors or shoppers (see Figure 2b).
A derivative PCA could be a loose-coupled agent system that searches other PCAs in an ad hoc manner or a close-coupled agent system that integrates different PCAs into its architecture.
Interestingly enough, as early as 2003, we found two derivative PCAs and each fits into the two design principles described previously. A PCA called Bestbuyfinder.com perfectly matches our definition depicted in Figure 2b. It is a piece of downloadable software that allows the user to enter a product name. Then it searches almost all major PCAs, compares the results returned by them and then presents the final result to the user. The derivative PCA that conforms to our close-coupled design is Shopping.com, a PCA re-branded from Dealtime.com. A critical factor leading to its emergence as a derivative PCA is its acquisition and integration of Resellerratings.com (an Evaluation PCA) and Epinion.com (Preference Identification PCA) in 2003.
In 2004, we witnessed the emergence of more derivative PCAs like Bizrate.com and Froogle.com that fit these two design principles. Bizrate.com extended its original customer rating and evaluation service to include product price comparison service and renamed itself as Shopzilla.com in 2004 to re-brand the new integrated services. Froogle.com, operated by Google, searches online vendors, product evaluation PCAs, and consumer preference identification PCAs for product and context information in a loose-coupled way. As a beta version, it is still in a nascent stage in terms of efficient ways of interacting with online shoppers and providing effective assistance in making online shopping decisions.
In 2005, more sophisticated, hybrid-style derivative PCAs like Pronto.com emerged that retrieved the three category of comparison-shopping information from different channels. For example, it not only allows vendors to feed the price and other product related information into its search engine but also scours the Web to find additional vendors and its product information. Instead of building its own product review database, it retrieves the product reviews from the product’s representative online retailers in an ad hoc manner. Figure 3 shows the distribution of PCAs across business categories.
We can also expect that PCAs will include more sophisticated user interaction modes to accommodate the more experienced online shoppers.
With the pressure of existing derivative PCAs, we expect the entry barrier for future PCAs to be increased considerably. Comparison-shopping information that used to be obtained from multiple PCAs is now expected to be obtained from one PCA only. The challenge for new PCAs will not only be about technical design but also about information brokerage and collaborations with other agents and vendors.
Although popular, derivative PCAs are not the only solution for new comparison shopping. In 2005, Microsoft relaunched its MSN shopping site to enter into the comparison-shopping market. It provides similar comparison-shopping information like Shopping.com but it collects price information from vendors and rating and review information from consumers directly. Certainly, this strategy is feasible due to the established popularity and financial strength of MSN and Microsoft.
Internet Structure and Web-based Agents
Many researchers have suggested that the hyperlink structure of the Internet is not randomly connected but connected in a way that follows power law distribution [1]. That is to say, a majority of the Web sites have only a limited number of hyperlinks pointing to them while a few popular Web sites have a huge number of hyperlinks pointed to them. With a Web structure following that power law distribution, small online vendors have the risk of not being “discovered” by consumers even though they may provide high-quality services. For example, search engines like Google utilize this power law distribution structure to provide search services. When we use Google to search for certain products, we may invariably get directed to major branded vendors like Amazon.com or buy.com, rather than to smaller vendors. Those small but quality vendors lose visitors because of their limited linkages and resulting weak Web visibility. This Web feature provided a fertile ground for the emergence of PCAs because they not only reduced the search cost of online shoppers but also made the smaller vendors visible to consumers.
However, the emergence of PCAs only partially solved the problem of weak visibility of small online vendors: while they do collect and aggregate information from these small vendors, they at most manage to cover only a small part of the vast Internet due to constraints on response time from consumers. Given the need to better serve this segment of small online vendors and the specialized consumer groups they target, we may reasonably expect that more types of derivative PCAs will inevitably emerge to collect and aggregate relevant information from existing PCAs and naturally replace some of them to assist online shoppers in decision making. For example, the online travel booking industry was initially dominated by three main PCAs: Expedia, Orbitz, and Travelocity. Since 2004, however, several new derivative PCAs have become popular like Bookingbuddy, Kayak, and Sidestep. Instead of utilizing an existing network like SABRE to locate flight information from airline companies, they search existing PCAs, especially the big three as well as other travel sites, and then aggregate the result to consumers. When the consumer selects the preferred flight, these derivative PCAs will route the traffic to the originating PCA or travel site. For the time being, the attitude of PCAs toward these derivative PCAs is ambivalent: Orbitz works with Sidestep and its airline fares often pop up as top deals; while Expedia regards them as competitors.
We can also expect that PCAs will include more sophisticated user interaction modes to accommodate the more experienced online shoppers. For example, several new PCAs, like smarter.com and become.com, developed a refine-as-you-type (RAFT) functionality that could automatically display the subcategories of the product keyword entered and narrow the scope of category automatically when one enters more keywords until the moment that only one category remains. This greatly increases the search efficiency of the PCA with more relevant results for the consumer.
Other Web-based agents are also emerging to accommodate other specialized demands of consumers. One niche category is the “deal” agents like techbargains.com. These agents aggregate information on shopping deals that are offered by various online vendors. Consumers can search through available deals at deal agent sites and then click to the deal page of online vendors directly via the link provided by deal agents. By increasing visibility of deals while also encouraging comparisons between competitive deals, these agents are both an opportunity and threat to online vendors in their promotional efforts. This is also a domain of opportunity for derivative PCAs to integrate them in their services in the future.
We have offered a cognitive information processing-based schema to classify information intermediaries like PCAs in this article. PCAs as well as other Web-based intermediaries are virtual entities on the Web, whose input and output are both digital in nature. The coordination and transaction cost that were used to explain the behavior of traditional intermediaries are minimal for these entities, making it easy for them to operate, collaborate, and morph into innovate forms. We hope future research will fully explore the impact of this phenomenon. We also hope this explorative research will trigger inspiration for other useful frameworks to study Web-based information intermediaries in general.
Join the Discussion (0)
Become a Member or Sign In to Post a Comment