3. Understanding Personalization

Using the framework presented in the previous section, we discuss the activities involved in each stage. We examine the research issues, link them to extant research, and then identify opportunities for researchers working in the management sciences.

3.1 Data Sources

To provide personalization effectively, firms need to capture data on customers, and then process or mine this data to derive relevant characteristics of customers. There are several sources of data that are valuable to a firm in learning about a customer’s tastes and preferences. Some of these data, like transactional data, are common across brick-and-mortar stores and web-based ones. When customers interact with a firm through its web site, all such interactions can be stored as well. These interactions can provide information that would typically be not available in conventional databases. We summarize the important data, their sources, and their uses in the personalization process. We refer the interested reader to (Mena, 2001) for additional details.

Transaction data/Point of sale data

This includes all information on items purchased, their prices, time of purchase, and all other information associated with a transaction. These data are typically captured directly in databases at the time the transaction occurs. A customer’s transaction history is a very important source of knowledge about the customer’s tastes. In traditional brick-and-mortar environments, the information on the customer, if at all collected, may be hard to deploy for personalization of services (particularly during the shopping process). With electronic stores, the customer information is usually mandatory (for payment and delivery of products), and the site can connect the customer information with prior purchase history. Identifying that a visitor is an existing customer is performed by requiring user registration or with the help of cookies, as discussed below.

Demographic data

These data are often used to profile customers based on characteristics such as age, gender, and income level. These data can be collected by asking users to register themselves for the firms products. In web-based environments, firms sometimes require visitors to first register themselves before they are allowed to browse the site. Physical and email addresses can also be used to help in profiling a user. For instance, advertising servers use ZIP codes to target advertisements for companies in that neighborhood. Demographic data can also be obtained from direct marketing companies, typically based on phone numbers and physical addresses.

Web logs

Log files record all visitor interactions at a site. While these files were originally designed to track server traffic, some of the data can be useful for personalization and customization related activities. Data captured include (i) the browser host IP (Internet Protocol) address; (ii) authentication information such as an ID or a password; (iii) date and time of the interaction; (iv) the Uniform Resource Locator (URL) for the page requested by the user; (v) the referrer field if any (e.g., search engine and keyword used to navigate to the site); and (vi) a cookie field that identifies if a visitor is new or a returning one. IDs and passwords are often used to customize a visitor’s site. The set of URL’s requested by a user is often referred to as clickstream data.

Cookies

Cookies are small text files that a web site server places on the hard disk of a browser host machine (client machine). A cookie helps the web site server identify a user both within a session, as well as across sessions. They typically include (i) the domain name of the server; (ii) how the cookie was created; (iii) the expiration date for the cookie; (iv) the name of the cookie; and (v) the cookie value that helps identify the browser host machine to the server. Cookies can be used in a variety of ways in the personalization process. Browsing behavior within a session can help the server understand the immediate needs of the customer. Across sessions, a cookie helps the server track repeat visits of users, which is very useful for sites that do not require authorization for access. Furthermore, the information in a cookie can help the server link a visitor to transactional and demographic data stored on that individual. This can enable the server to tailor content or recommendations to the user. In addition to tracking behavior within a site, cookies can be used to track a customer at multiple sites. For example, advertising servers (or Ad Networks, as they are referred to in the popular press) track a person’s visits to all those sites that are serviced by that server. These data enable the Ad Network to learn about a customer’s preferences that cannot be gleaned from navigation within a single site. The Ad Network can use that knowledge to better target advertisements and manage advertising campaigns.

In addition to the above, there exist other sources of data on customers that could help in the personalization process. These include customer service databases, warranty claims databases, and any other point of contact with the customer that is recorded by the firm. Firms are investing heavily in Customer Relationship Management (CRM) software to capture all information about customers’ interactions, which can enable the firms to arrive at a single unified view of each customer.

3.2. Learning about Consumers

As shown in Figure 1, firms would like to obtain information about consumers’ preferences, past purchases, decision processes such as search, evaluation of alternatives, the choice process, and the purchase process. We address each aspect in the ensuing discussion.

3.2.1 Preferences

The most obvious way of getting preference data from consumers is to ask them using surveys. Info Harvest Inc. is an example of a company that specializes in doing preference surveys using secure data gathering methods. On the web, consumers are given incentives to reveal information about themselves by filling out forms and surveys. Varian (2001b) observes that most consumers are willing to reveal some information about themselves (such as name, address, and some preferences) but not other types of information (such as income, or credit card information). Further, consumers wish to reveal their information to a very small set of firms in each need domain. For example, online consumers may register at one or two travel, news, or online financial service websites. The fear of misuse of their personal information and invasion of privacy make this mechanism limited in its scope. In addition, surveys may be erroneous due to people forgetting, overstating or understating responses, and instrument biases. There is also a self-selection bias in responses when customers who choose to fill surveys are different from the target population (Montgomery 2000). Therefore, there is a need to obtain information using other non-intrusive means.

The Internet provides an easy mechanism to track and archive consumers’ web-browsing behavior. This data can then be analyzed using Artificial Intelligence (AI) techniques such as decision tree induction (Quinlan, 1990), mining association rules (Agrawal et al, 1993), and neural networks to uncover patterns in consumer behavior. Clustering or other collaborative filtering techniques (Goldberg et al, 1992) can then be used to group consumers with similar buying patterns and make intelligent recommendations about products that a consumer would consider buying. Amazon and other online retailers have employed these techniques extensively. Companies such as Netperceptions, Broadvision, and Strategic Data Corporation, have developed personalization software using the above techniques.

One particular technique that is eminently suited for assessing the relative importance weights, and utilities of different features of a product or of a website is conjoint analysis (Green and Srinivasan 1990). It has been developed and used extensively in marketing in new product development. In this technique, respondents are asked to rank or choose among products that comprise of combinations of attributes that are carefully selected by the researcher according to an experimental design. Using the rank or choice data, researchers can determine which attributes of a product are important and what is the value of each attribute to a customer. On a website, a customer makes choices by clicking on different pages or features. Therefore, choice-based conjoint analysis can be employed with click stream data to quantify the utility of different types of content, tools, and features on a web page, and this can be done in real time for each customer. Conjoint analysis has been used in designing personalized websites (Dreze and Zufryden 1997), testing product concepts (Dahan and Srinivasan 1998) and in understanding purchase behavior (Brynjolfsson and Smith 2000). Montgomery (2000) describes nicely how conjoint analysis can be used to profile customers on the web.

When data strings for an individual consumer are sparse, modelers can combine data across different households to estimate individual level parameters using hierarchical Bayesian (HB) techniques (Allenby et al, 1995; Lenk et al, 1996). The use of HB techniques improves the stability of the estimates at the individual level and also allows a way of combining data from surveys with that from consumer’s choice data. Sawtooth Software Inc. provides the software to conduct conjoint analysis and has recently introduced a module that employs hierarchical Bayes estimation technique.

3.2.2 Consumer decision processes

Traditional consumer decision process models include the following steps: problem recognition, information search, evaluation of alternatives, purchase decision, and post purchase behavior (Kotler, 2000). We examine how the consumer decision process is different for purchases made online, and how firms obtain an understanding of these processes for each customer to use in personalization.

Problem recognition

For most products, consumer realization of the need for a product does not differ significantly between online and offline purchases. On the Internet, there are two new aspects that affect problem recognition. First, personalized recommendations aid in accidental triggering of need for unanticipated products and thus create demand for products. For example, while purchasing a certain type of music at CDNow, if the website suggests other music labels, it could trigger further exploration and possible purchase. Second, community feedback might provoke some consumers to consider different purchases. The magnitude of the demand generation effect and the conditions that enhance product discovery through recommendations and community input need further study.

Information Search

The Internet allows easy access to abundant information with the click of a mouse. This glut of information impedes information search. However, portals and websites simplify search by organizing information and providing tools that allow customers to compare different product choices along a number of dimensions. For instance, at Edmunds.com, an online provider of information about automobiles, it is possible to search for many cars and conduct side-by side comparisons, and get an idea of invoice prices. There exist specialized software called price bots (e.g., at mysimon.com) that allow consumers to search for the lowest prices of any given product on the internet. Further personalized recommendations make search easier.

Haubl and Trifts (2000) study the effect of two interactive decision tools, namely, recommendation agents and comparison matrices on consumer decision making in an experimental online shopping situation. They find that these tools reduce consumer’s search costs by making the search process efficient and improve the quality of the decisions. These tools also reduce the size of the consumer’s consideration set (the set of items that a consumer considers before purchasing). They focus only on goal directed tasks, where subjects were asked to make a purchase. Further research should consider non-goal directed search, such as when consumers gather information without any immediate intention to purchase.

By making it easy for consumers to compare products, conventional wisdom suggests that consumers may become more price-sensitive. Using experiments, Lynch and Ariely (2000) test the conditions under which lowering search costs (by providing tools for comparison) would affect price sensitivity in the context of online wine purchase. They find that for commonly available wines, price sensitivity increased when cross-store comparisons were made easy while for unique wines there was no effect.

Search behavior can be tracked by analyzing clickstream data. In a recent study, Moe (2001) classifies people into buyers, browsers, and searchers based on the type of information that they were seeking. Further research is needed to understand the role of personalization in affecting search. Also novices and experts may well vary in their search patterns, duration of search and the choice of tools for searching. Experts tend to prefer keyword search, while novices favor browsing using menus (Pollock and Hockley, 1997). Research also needs to shed light on the differences in credibility of recommendations from manufacturer’s websites, from independent third party sites, and from partner websites. The issue of trust is important in this context.

Evaluation of alternatives

These are a number of theories about the processes and rules that consumers use to evaluate different alternatives. When faced with many choices consumers could follow a two- stage decision process. First they could screen many alternatives using various filters and narrow their selection to a few items and then conduct an in-depth evaluation of this smaller set. Haubl and Trifts (2000) find that recommendation agents are useful in the initial screening stage while side-by-side comparison matrices are employed in the in-depth evaluation phase. Consumers also employ a wide variety of rules such as lexicographic, conjunctive and disjunctive in combining information about multiple attributes. Such rules were employed by consumers in traditional purchase situations to minimize cognitive effort. When comparison tools and shopbots are available on the Internet, it remains to be seen whether consumers would prefer more complicated multi-attribute models. Firms can uncover such rules through experimentation or data mining techniques.

Purchase decision

Online purchase data reveals choices that have been made by consumers in an actual purchase environment. This data is a rich source of information about consumer’s price sensitivity, brand preferences, price-search behavior, and responsiveness to different types of promotions. The determinants of brand sales and market shares can be modeled at an aggregate level using regression analysis. Such models help managers understand the effectiveness of different promotions and price manipulations. In addition, one can understand the nature of competitive effects.

At an individual level, brand choice can be modeled using discrete choice models. In the last ten years, marketing researchers have developed a number of models that employ multinomial logit or probit models to understand brand choice behavior in grocery scanner data (Guadagni and Little, 1983; Rossi et al, 1996). These models can be used to explain the probability of choice of a given brand (or product) as a function of marketing variables such as price, promotions, and advertising communications.

The logit model can be estimated for each individual if a sufficient number of purchases have been made in a given category. Quite often, enough data points on an individual customer may not be available. In such cases, the choice models can be estimated at an aggregate level by combining data from many customers. When grouping customers, the estimates of the choice model will be biased if differences between individuals (such as differences in their preferences or in their response to marketing variables) are ignored (Guadagni and Little 1983). This issue is called heterogeneity and has been modeled extensively in marketing using mixture models (Kamakura and Russell, 1989), random intercept models (Chintagunta et al, 1991), random coefficient models (Gonul and Srinivasan, 1993) or multinomial probit models (Rossi and Allenby, 1993). Degeratu et al (1999) use choice models to model online grocery purchases.

Data on timing and duration of web visits or transactions can be used to predict traffic patterns at different times. Researchers have developed purchase incidence models to predict the frequency distribution of purchases in a given time period. In the popular Negative Binomial Distribution (NBD) model (Ehrenberg 1955), consumers are assumed to arrive according to a Poisson process with an average rate l, and the purchasing rates l are distributed over the population of customers according to a gamma distribution. The resulting distribution of number of purchases within a given time interval is described by a negative binomial distribution. Morrison and Schmittlein (1991) present a nice review of purchase incidence (NBD) models used in marketing to forecast the number of purchases/visits in any given time period. By comparing the actual frequency distribution with the predicted values, firms can determine whether promotions are attracting new customers or repeat customers or previous heavy buyers. Fader and Hardie (1999) and Moe and Fader (2000) use the purchase incidence model framework to model customer trial and repeat purchase over time at the Internet music retailer CDNow.

Alternately, instead of looking at number of purchases in a given interval, researchers can model the time between arrivals or the duration of visits on a website. Hazard rate models have been used to explain the factors that affect the inter-arrival (or inter-purchase) time for consumers (Jain and Vilcassim 1991). These models help managers understand the factors that affect the time between visits or purchases. Hazard models can also be employed to model the optimal number of catalogs (messages) to send to a customer, and the appropriate time interval between them.

The role of personalization in affecting each of the above mentioned consumer decision processes offer much scope for research in the coming years. For instance, the issue of whether personalization affects the choice itself, the price sensitivity of the consumer, the loyalty of the consumer, the visit/purchase frequency or the consideration sets are all important to managers and need to be quantified for different product/service categories.

A major challenge that one needs to address here is the issue of endogeneity of prices and promotions (Leeflang and Wittink, 2000). In traditional models, researchers have assumed that the prices and promotions are exogenous to the consumer as firms set these for most products and consumers could not affect these prices. However, on the Internet, firms engage in personalized pricing, and communications, which makes the assumption of exogeneity of prices untenable. Models of personalization need to treat prices and promotions as endogenous. This issue of endogeneity applies to every aspect of the transaction where the consumer has an input in creation of the product and/or the offer. Villas-Boas and Winer (1999) have developed a model which incorporates endogeneity in the context of brand choice models using scanner data. Other types of models need to incorporate endogeneity as well.

Another issue with the use of clickstream data is that most data about customers is restricted to their browsing behavior within a given site. For instance, the analysis of web usage data by Moe and Fader (2001) focus on site-specific data. Padmanabhan et al (2000) point out that this could lead to significant bias in results relative to interpretations drawn from data on browsing behavior collected across multiple web sites. They show that user-centric data outperforms site-centric data by a significant margin in predicting purchases in future sessions. Future research needs to develop models to account for incomplete data. Since most websites will have incomplete data, this is an important problem. Potentially erroneous solutions can be reached when using incomplete data.

Firms generate additional revenue by cross selling goods on the Internet. There are few choice models that look at this phenomenon. Models of consumer shopping with basket data (many product categories) as opposed to a single product category are now becoming popular in marketing (Manchanda, et al, 1999; David et al, 1998). New methods such as hierarchical Bayesian estimation techniques are being employed. These techniques are able to capture covariances in preferences (and responsiveness to promotions) across multiple product categories with limited number of time series observations for each individual.

There are a number of recurring issues across all the management science models discussed above. The first issue deals with the tradeoff between speed and sophistication. Individual level choice models which control for unobserved heterogeneity and endogeneity may be sophisticated but are computationally burdensome. Other data mining techniques may offer the benefits of less accurate but faster solutions. Ideally research needs to develop techniques that deliver both benefits. Alternately, research is needed to determine optimal tradeoffs between speed and accuracy. A second issue deals with integration of data across multiple sources. For example, how should clickstream data be combined with offline data, or transaction data be merged with survey data? Researchers need to consider the effect of aggregation on statistical distributions. Further, managers need method to handle discrepancies across multiple data sources.

Postpurchase behavior

Consumers continue to evaluate their purchase after the item has been bought. It is important for firms to retain customers and so they need to understand how to convert a customer into a repeat customer. In this connection, there is some research on retention of customers and churn rates. Battberg and Deighton (1996) use a decision calculus approach to link marketing expenditures to customer acquisition and retention rates. Bolton (1998) explores the link between customer satisfaction and retention. Since it is feasible to track customers on the Internet, research could study how online transactions translate into post purchase behaviors (such as word of mouth and posting of messages on community boards). Further, the relative effects of positive and negative word of mouth publicity on other customers is relevant for managers.

3.2.3 Effect of environment and community

There are other elements of the environment that can affect a consumer’s decision process. Family members, friends, salespersons, and other people often influence decisions in many ways. In addition, marketing variables, competitive factors, and situational factors affect the purchase process. An interesting fact is that a large number of products (about 70%) that are put into shopping carts by online customers are abandoned before reaching checkout. This is a large opportunity loss and firms need to understand the factors that affect such behavior.

This raises interesting issues for personalization that have not been currently addressed. In group-decision making situations (such as when a family makes a decision regarding purchase of a car or insurance), if firms had data on both the wife and husband, how could they combine the different pieces of information to provide a personalized recommendation? What models are relevant for aggregating preferences of the members of a household or a group?

Another interesting aspect of the Internet is the development of communities, where consumers can go online to community spaces to gather or share information about vendors, prices, products, recommendations, and experiences of other consumers. Chat rooms, instant messaging, and bulletin boards are all online tools that are made available by firms to facilitate discussion among its customers. For instance, AOL has 33 million customers, and over 120 million registered users of ICQ, the instant messaging software. AOL members generate 1.2 billion messages everyday and spend 10 million hours per week in chat rooms. Given the widespread popularity of online communities and active participation by many members, firms find it attractive to build and maintain online communities (Hagel and Armstrong, 1997).

There is very little research on how communities shape beliefs and perceptions about products and whether firms can manage these communities to their advantage. Can providing access to communities enhance differentiation, and can this differentiation be used to enhance personalization to individual consumers? Another issue is to quantify the nature of network externality that may accrue due to membership in a community.

3.2.4 Shopping Experience

Internet is a fascinating experience, especially for novice consumers, who tend to become totally absorbed and are surprised at the amount of time they have spent browsing. This idea has been called flow (Csikszentmihalyi, 1991) and has been researched by Hoffman and Novak (1996). They argue that flow is very important for firms selling entertainment through the Internet. However, achieving flow is hampered by the need for firms to get the consumers to engage in profit generating activities either by providing information, making purchases, or even looking at banner advertisements. Research needs to study the balance between providing a stimulating environment and practical considerations of commerce. In addition, research needs to understand whether personalization enhances or mitigates flow. Consumers may get irritated if personalization were to interfere with their enjoyment.

3.3 Personalization Mechanisms

There are a number of mechanisms to deliver personalized services to customers. CDNow and Amazon popularized the use of collaborative filtering techniques to provide recommendations for music and books. The recommendations are based upon purchase information from other customers who match the profile of a given customer. Research is being conducted to develop better matching and recommendation algorithms. Collaborative filtering is appropriate when consumer tastes vary considerably, and the product is not expensive or risky. Companies like Net Perceptions and Art Technology Group specialize in collaborative filtering to develop targeted content and product recommendations. An alternative set of recommendation technique uses rule-based systems. These systems rely on modeling expertise and knowledge of customers to develop rules to deliver recommendations that are beneficial not only to the customer but also serve the needs of the firm. Firms that employ rule-based engines include Blaze Software and Broadvision. Research needs to clearly define the conditions which favor each of these techniques.

Another domain of personalization deals with delivering targeted communications, which can be either advertisements or e-mails. For example, based on visitor profiles, advertising server software places advertisements for appropriate product categories. Ad Networks schedule banner advertisements for their clients, keeping in mind the site requirements and customer preferences. Adler et al (2001) and Kumar et al (2000) have developed scheduling algorithms to maximize advertising revenues for a site. This is a growing area of research and newer models need to address optimization issues not only over multiple advertisements and web sites, but also over time. Permission based communications, in which consumers opt-in to receive promotional offers in certain categories, is becoming popular and provides better response rates relative to direct mail communications. There is a need for research on techniques that address optimal scheduling of messages while dynamically generating appropriate messages.

Dynamic and customized pricing mechanisms are being developed to maximize revenues for online firms. Firms need a good understanding of consumer’s willingness to pay to be able to extract the right price. Customers may also learn and adapt to the pricing mechanisms. Research also needs to explore methods to extract rents from customers without affecting their satisfaction. There was a public outcry when it was reported that Amazon tried to implement dynamic pricing strategies.

Yet another factor that has become important is how to control congestion at the back end servers that deliver personalized content and handle transactions. Priority queueing approaches based on customer lifetime value, coupled with caching mechanisms, can be used to optimize the use of such resources. VanderMeer et al (2000) have examined caching approaches, although they do not consider prioritizing services based on customer classes. As these technologies mature and become better understood, it will become increasingly important for firms to optimize personalization related operations to derive full benefit from them.

Other personalization mechanisms include using consumer profiles to enhance full text searches (e.g., Verity, Infoseek, and Fulcrum). In addition, Vignette and ATG match profiled users to relevant content. A new idea is to develop user driven agents that track web accessible sites to provide personalized information. Some firms provide customized products or web pages based on an explicit statement or an inference about consumer’s preferences. Mobasher et al (2000) have used association rule mining to dynamically include interesting links to visitor’s web pages based on their browsing behavior.

3.4 E-metrics

A number of firms feel that personalization has not delivered on its promise. Many firms have not seen tangible results such as increased sales or profits due to their personalization efforts. Recent industry efforts have been directed towards development of e-metrics to quantify the benefits of e-commerce in general. NetGenesis has a white paper on e-metrics in which they define a personalization index, in addition to discussing traditional metrics such as reach, acquisition, conversion and retention. We discuss the e-metrics that are relevant for measuring the effectiveness of personalization.

Personalization is meant to improve customer satisfaction and increase the loyalty of customers. Loyalty (also referred to as retention) can be measured either as an action tendency (typically repeat purchase) or as strong positive attitudes (e.g., “I will only buy at Amazon”). There are currently no known studies on how personalization affects loyalty. Similarly, personalization could affect conversion of browsers into customers and these benefits need to be quantified. Customer satisfaction has traditionally been measured as the gap between expectations and actual performance and many metrics are developed in the literature (Zeithaml et al, 1988). On the Internet, expectations of consumers are significantly different from that in traditional shopping environments. For instance, consumers expect faster service, want past purchases to be remembered, want access to past transactions, demand customization, and want the experience to be fun and interactive (Hanson 1999). The manner in which the changed expectations affect the way customer satisfaction is managed and measured is an important research issue. Recent research has developed the dimensions of e-service quality (Parasuraman et al, 2000).

In addition, more research needs to be done on exploring the links between personalization and cross selling or direct marketing efforts, churn rate (the rate at which existing customers leave), life-time value (the discounted stream of profits from a customer over the lifetime of her relationship with the firm), stickiness of a website, and word-of-mouth publicity. In this connection, it may be beneficial to consider the effects of personalization on visitors and customers separately.

4. Customization, Personalization, and Firm Strategy

The ability to personalize products and services can provide considerable strategic advantage to a firm. The strategic impact can manifest itself in several different ways. For example, personalization can help firms differentiate their services from their competitors, leading to competitive advantage. For instance, customization and personalization strategies can help a firm perform price discrimination (Dewan et al, 1999, Ulph and Vulcan, 2000), and provide, in some industries, first mover advantage (Resnick and Varian, 1997).

In examining the impact that personalization may have on a firm, it is important to understand how value is created using these kinds of technologies, and consequently to recognize the key players in the firm’s value chain. We use a modified version of the Value Net [Brandenburger and Nalebuff, 1995] to provide a schematic map of the relevant players, and the interdependencies across these players. This schematic serves as a useful framework for identifying the various ways in which personalization technologies can become important to a firm’s strategic behavior.

Figure 2: The Enhanced Value Net

In the Value Net approach, a firm interacts with customers and suppliers in the vertical dimension, and with competitors and complementors in the horizontal dimension. Typically, transactions occur in the vertical dimension, with products and services flowing from suppliers to customers, and money flowing in the reverse direction (i.e., top-down). Customer information, the critical ingredient for personalization, also flows top-down. Competitors and complementors impact a firm’s ability to transact with its customers and suppliers. Since customer information flows to Competitors and complementors as well, the ability of a firm to effectively differentiate its products and services is also affected by the actions of these players. We enhance the model of Brandenburger and Nalebuff by including the entity channel between the firm and its customers. The channel could, for example, be a retailer for a manufacturing firm, or a portal for a content provider on the worldwide web. Since the flow of customer information to a firm would typically go through such a channel (if such an intermediary exists), the channel can become an important player in a firm’s personalization strategies. In the following discussion, we look at the interactions between a firm and the other players, and examine the strategic role of personalization for each of these interactions.

4.1. Firm and Customer

The important strategic consideration between a firm and its customers is the bargaining power of the customer. Effective personalization strategies can help shift the power in favor of the firm. We examine broadly the following issues: product differentiation, price discrimination, bundling, privacy, and information asymmetry (strategic behavior of customers). We elaborate on how personalization impacts each of these issues, and pose questions in the context of these issues that need further research.

Personalization techniques enable firms to better differentiate their products or services. Most goods are differentiated to some degree or other and the economic explanation for differentiation rests on two premises. One is that there are differences in consumer tastes between individuals (or even for the same individual over time). The second premise is that individuals prefer, and sometimes are willing to pay more, for products that are more suited to their own tastes. Firms, therefore, have an incentive to develop multiple variants of a product to satisfy this need for variety.

By developing products that are tailored to customer’s tastes, firms can charge a premium price for their product. For example, a custom-made pair of Levi jeans is priced at a premium of $10 over the standard product’s prices (the premium price typically offsets the additional costs incurred, thereby, providing higher margins). This is an example of price discrimination as firms can charge different prices to different customers who have different valuations for products. Personalization techniques allow firms to precisely estimate their customers’ valuations at low costs, and hence enable then to engage in finer price discrimination.

A taxonomy commonly used for price discrimination considers three types (Varian, 2001). When a firm is able to charge different prices to different customers, it is termed first-degree price discrimination. A firm engages in second-degree price discrimination when it makes available a set of related offerings with fixed prices associated with each, and customers choose the product that best fit their tastes. This phenomena is also referred to as product line pricing or versioning (Varian, 2001b). Examples include the many versions of Quicken accounting software, different versions of DVDs of movies (basic and collector’s edition), and even stock quotes (real time versus 20 minute delayed). In third-degree price discrimination, firms charge different prices to different groups (as distinct from individuals, which is of course first-degree)[3].

In the past, first-degree price discrimination was not a practical approach in most markets because it was quite expensive or sometimes impossible for a firm to gauge the consumer’s willingness to pay. With access to enormous amounts of customer data in electronic form, and the tools to analyze these data in close to real time, firms have over time become better able to estimate customer valuations. Further, technology now permits the gathering of information about consumer tastes at low costs. By analyzing consumers’ click-stream data and purchase history on the Internet, a firm is better able to price its products based upon the willingness to pay of the customer. Thus, personalization enables better differentiation of products offered, which in turn can lead to better extraction of consumer surplus. It is becoming practical for companies to develop a larger number of variants of products and, in some cases, even serve individual customers profitably. Formal analysis of how personalization enables first-degree price discrimination under different conditions is needed. For example, it may be possible for a firm to estimate a customer’s valuation for a product, and use customized coupons to match the effective price of the product to an individual customers’ valuation. Under what conditions should we expect to see the proliferation of products and services? Should we expect this to be more or less pronounced for information goods (that usually have close to zero marginal costs)? A related issue for potential research is to understand how personalization technologies can be used to deliver dynamic pricing strategies in real time. This enhances the ability of a firm to perform price discrimination (by providing an additional dimension to consider in it’s pricing scheme), and may lead to substantial gains in traditional as well as spot market environments.

In many environments there will be a limit to the number of variants that can be produced because of increasing returns to scale, especially in traditional products. To recover the costs of developing and supplying different variants, firms need a sizeable market. In this context, Dewan et al (1999) have examined the range of standardized and customized products that form the optimal product spectrum for a firm in a monopolistic setting, when the firm engages in second-degree price discrimination. In deriving their results, they assume specific functional forms for the marginal costs of production. Results obtained using more generalized functional forms should be of great interest to both academicians and practitioners.

The Internet allows for the reproduction and distribution of information goods at very low marginal costs. This has interesting implications on the bundling of information goods (Bakos et al., 1999). Chuang and Sirbu (1999) show that allowing customers to self-select a bundle consisting of a subset of goods (rather that pre-designating the goods in a bundle) can often improve a firm’s outcome. Hitt and Chen (2001) extend this stream of work to show that for a monopolistic setting, such a mechanism outperforms individual selling and pure bundling when marginal costs of providing the goods are greater than zero, and customers have heterogeneous preferences.

Protecting the privacy of individuals has become a very important issue because of the low costs associated with collecting and disseminating information on the Internet and otherwise. Currently, the market on personal information is based on the notion that the institution that has gathered the information also owns the information [Laudon, 1996]. While privacy laws are being enacted to guard against unauthorized use of personal information, there is likely to remain a significant market in personal information. For a customer to be willing to share personal information with a firm, she must have a clear idea about benefits she can expect to receive, how the information will be used by the firm, and how it may be shared with other organizations. Laudon suggests the possibility of creating a National Information Market in which information about individuals is bought and sold at a market-clearing price. In this kind of a market, an individual would have the ability to grant to institutions the right to use their personal information for a predetermined period of time and specified nature of use. There are several interesting research issues that warrant examination in this context. For instance, a firm would like to obtain as much information on a customer as possible before engaging in a transaction with her. The customer, on the other hand, would like to obtain perfectly personalized service by providing as little information as possible. It would generally be beneficial for the customer to share information that would enable the firm to provide the right product to her. At the same time, the customer would not like to provide information that would reveal her reservation price for the product. The firm should provide incentives to the customer in order to convince her to share some of this information. Incentives could be, for instance, of a monetary nature or a mandatory requirement for receiving recommendations (Resnick and Varian, 1997). A related phenomenon is that of users deliberately providing incorrect personal data in an effort to obtain the desired recommendations without divulging those details that they consider too personal. Therefore, incentives must be such that users do not falsify their data. Implications for one-time purchase products and repeat purchase situations need to be examined. Yet another topic for study is the impact of such an information market on transaction costs associated with personalized products, and thereby its impact on the social welfare.

Many sites such as IMDb, CDNow, and Amazon base their recommendations on ratings of products obtained from users. The ratings provided are useful in identifying customers with similar tastes. Since users are typically able to provide their ratings anonymously, it is possible for interested parties (e.g., producers of these products) to manipulate the ratings. An interesting question here is what kinds of incentive mechanisms are required to elicit unbiased ratings.

4.2 Firm and Competitors

Competitors pose the threat of substitutes to a firm. This is clearly a very important aspect of the personalization and customization strategies that a firm has to consider, and consequently has many important research implications. The issues we examine here are: differentiation, price discrimination and price competition, switching costs and lock-in, first-mover advantage, and network effects. We briefly survey the existing literature, and then identify some questions that warrant additional research.

A firm’s personalization and customization efforts have the strategic effect of increasing differentiation, which in turn helps reduce price competition (Shaked and Sutton, 1976), generate greater loyalty among consumers, and in some cases, generate price premiums (by extracting greater consumer surplus). Shaked and Sutton have shown that increased product differentiation leads to reduced price competition in equilibrium.

However, as more firms start personalizing their services, there is also an enhanced competition effect, which reduces the benefits of surplus extraction [Ulph and Vulkan, 2000]. This intensified competition comes about because firms are competing for smaller and smaller segments of consumers. Ulph and Vulkan show that when consumers are homogeneous in taste, the competition effect dominates the surplus effect making firms worse off with personalized pricing. Using a duopoly setting, they characterize when it is profitable for both firms to engage in first-degree price discrimination, and when the firms are both worse off. In related work [Ulph and Vulkan, 2001], they examine situations where firms are able to customize a range of products at the same marginal costs. They show that a firm is always better off using price discrimination if it also mass-customizes.

Their results are along the same lines as those of Dewan et al [2000a] who show that when firms in a duopoly simultaneously adopt customization there is reduced differentiation, which should lead to greater price competition. However, firms charge higher prices on customized products and this compensates for the lower prices due to price competition. In their model, the authors assume that firms incur an additional cost in order to customize their products. They further assume that the firms price discriminate to the second-degree. They show that when one firm adopts a customization strategy, it is able to improve its market share and profits at the expense of other firms. However, it then becomes optimal for other firm’s to also adopt customization, which in turn leads to excessive investments in customization leading to lower profits for all the firms.

In another work, Dewan et al [2000b] examine if there exist any first-mover advantages for a firm to adopt customization, and find that when firms adopt customization sequentially, there is an advantage for the early adopter. Further, they show that by investing heavily in customization, a firm can deter entry of potential rivals.

The above studies provide a good starting point for further research on the competitive effects of personalization and customization. There are a number of interesting research issues that deserve further attention. Existing studies assume that all firms have full information and do not allow some firms to possess greater knowledge about customers than other firms. Even though switching costs are lowered on the Internet, customers may find it costly to provide information about their preferences to firms and therefore be unwilling to engage in such exercises with many firms. When does a firm lock-in its customers using personalization? How will this affect the market equilibrium? In a related vein, for repeat purchase environments, a firm can over time acquire customer information that enables the firm to be able to better customize their product offering, as well as, improve their ability to discriminate on prices. How should a firm invest in personalization and customization technologies to ensure that they can sustain their advantage over their competitors?

Many personalization techniques (most notably collaborative filtering) are more effective when implemented with a large customer base. Consequently, for products with high search costs where personalization adds significantly to the value of the product, there are indirect network effects to customers for shopping at sites that are well established. Therefore, there may exist important first mover advantages. What implications does this have on the market structure in equilibrium? Resnick and Varian (1997) speculate about the competition across recommendation systems themselves, positing that one or two systems would emerge as survivors in each product category. More formal analysis needs to be done in this regard.

Firms can differentiate from competition by using other strategies such as developing a strong brand, and partnering with strong and highly visible companies. Research is needed to quantify the magnitude of differentiation that can be obtained from personalization relative to other sources of differentiation. We need to understand the conditions under which personalization is a significant source of differentiation relative to other alternatives. In other words, which products and services would most benefit from personalization? What environmental conditions (consumer characteristics, market structure, etc.) enhance the effect of personalization in a competitive setting? What kinds of interaction effects exist between the multiple sources of differentiation? For instance, does personalization enhance or diminish the effect of branding?

4.3 Firm and Suppliers

The bargaining power of suppliers is the important strategic consideration here. Of interest here is how customization and personalization strategies of a firm impact its suppliers. While the impact is indirect in nature, the following issues appear to have interesting implications: product proliferation, information sharing, and forward and backward integration.

The ability of a firm to provide customized or personalized service may be dependent on the firm’s ability to harness its supply chain in an effective manner. An important assumption often made in the literature on customization is that it can be performed at uniform marginal costs, and these costs are low. For a firm to be able to achieve this efficiently, it is important that the product proliferation that typically results from customization should not require very high fixed costs (Varian, 2001b). There are several unanswered questions in this context. How does product proliferation for a firm impact the firm’s ability to transact efficiently with its suppliers? Does it require the firm to use a larger number of suppliers with higher costs of engaging in such supplier relationships? What kind of revenue sharing would be optimal for a firm to align its supplier’s incentives to its own? Should the firm share the customer information with its suppliers? While these questions are interesting in general, they are even more so for digital products. Consider a portal that has contracted out the delivery of content. Should the portal make the customer data available to the content providers? This could have important implications on the competitive landscape, as this may allow the content provider to compete for the customer’s directly. Over time, this may even enable the content provider to gather knowledge about its competitors that are also supported by the portal. Finally, if the transaction costs become too high due to product proliferation, it may be worthwhile for either the firm or the supplier to consider vertical integration.

4.4 Firm and Complementors

Complementors could play an important role in a firm’s customization and personalization strategies. The bargaining power of the complementor is the primary strategic consideration for the firm. We identify bundling and information sharing as the important issues in this interaction.

By engaging in strategic partnerships with its complementors (e.g., hardware from Dell bundled with MS Windows software from Microsoft), a firm can greatly increase the ability to customize its products for a large customer base. The coverage of the product space can increase substantially, provided the complementary firms ensure that different versions of their respective products are compatible. This can lead to a very fine-grained level of customization at relatively low costs, leading to a higher ability to price-discriminate than traditional bundling of complementary products.

There are several questions of interest here. How should the additional consumer surplus extracted by the complements be shared? If versioning is more costly for one firm than the other, how should it impact revenue sharing? How do customization capabilities drive the choice of a complementor, when several possibilities exist? What incentives should a firm provide to its complementors to achieve compatibility of offerings? When would firms want exclusive rights over its complementors’ products?

Another related, but distinct, set of issues pertain to how information should be shared across complementors. At first sight it would appear that the firms would benefit from sharing customer information as they would be better able to customize their offerings, and also engage in cross-selling their products. However, sharing of this information could lead to a shift in the balance of power between these firms. Should any customer information be even shared with the complementors? Or would there exist some intermediate level of sharing that would be optimal for the firms? It is possible that one firm can sell its customer information to the other. For example, Microsoft may find Dell’s customer list quite valuable, as this would enable Microsoft to target these customers for software upgrades.

4.5 Firm and Channel

Two strategic considerations are important here, the bargaining power of the channel, and also the threat of substitution. Important issues in this context are information sharing and coopetition.

Many firms find it necessary to use some intermediary in order to reach their eventual customers. These intermediaries, or distribution channels, traditionally include wholesalers, retailers, and agents. While wholesalers and retailers typically resell the firm’s merchandise, agents usually negotiate with customers on behalf of the firm. Since these intermediaries have first hand information on the customers, they play an important role in personalization of products and services.

The first question that arises is how much of the customer information should a channel share with the firm? Information regarding customers’ tastes would help the firm in identifying important market segments and thereby in better targeting its products. At the same time, information on a customers’ valuation of products may enable the firm to negotiate better terms with the channel. With the relatively low costs associated with setting up storefronts on the Internet, a firm may find it profitable to directly target its customers and compete with its existing channels. Therefore, the channel may well find it disadvantageous to share all of the customer information.

In some situations, channels could serve as intermediaries that enable competitors to share data that is mutually beneficial. For example, an electronic mall can track visitors’ movements across all storefronts, and make that information available to participating stores, perhaps at a nominal price. Going one step further, the mall could also collect transactional data from the stores, and provide information at some level of aggregation back to all of the stores (resulting in some amount of coopetition across these stores). This would enable the stores to better assess the customer’s preferences, and help them determine what products to recommend to them. Several research issues emerge in this kind of a marketplace. When would it be worthwhile for the marketplace to engage in provisioning this type of personalization services? How should the personalization provider charge for their services? What are the implications to firms who do not participate (or, stated differently, what kinds of firms would prefer to not participate)? How does it impact the customer, and what are the privacy implications in this context?

Table 1 provides a summary of the key research issues that we have identified for understanding the impact of customization and personalizing on firm strategy.

Table 1: Summary of impact of customization and personalizing on firm strategy

Interaction	Research Issues
Customer	Differentiation, Price discrimination, Bundling, Privacy
Competitors	Price discrimination, Lock-in, First mover advantage, Network effects
Suppliers	Transaction cost economics, Incentives for information sharing, Vertical integration
Complementors	Bundling, Information sharing
Channel	Information sharing, Coopetition

5. Conclusion

The advent of e-commerce/e-business has generated new opportunities for personalization and customization. The widespread availability of Internet technologies, along with the steeply falling prices of computers, has changed the economics of personalization. With improved technologies in flexible manufacturing and in developing digital products, constraints in providing customized products have been mitigated in several areas. While neither concepts of personalization and customization are new, the shift towards e-tailing has made these phenomena of critical importance to firms in a large number of industries.

In this article, we highlight aspects of personalization and customization that we consider offer great opportunities to researchers in the management sciences. Taking an interdisciplinary approach that spans the areas of marketing, economics, information technology, and operations, we establish links between the multitude of research issues and the different disciplines. We have approached these issues at two different levels. First, we identify research issues at the level of the firm, focusing on the activities a firm must perform to effectively provide personalization. In this context, we have developed a framework that delineates the various stages of the personalization process. This framework enables us to not only discuss the research issues in the context of the different activities involved in personalization, but also helps us recognize which aspects are largely unexplored in extant research. At the second level, we step back and look at the role of customization and personalization in a firms value system. The framework we use for our analysis here is a modification of the well-known Value Net framework. We examine the role of customization and personalization in the interactions between a firm and other key players in the firms value system, survey extant research, and suggest avenues that we consider are promising to management science researchers.

While we have attempted to provide a reasonably comprehensive survey of the issues involved, we recognize that a single article cannot do justice to all of the issues that are worthy of study or to present all of the research conducted in these areas. Our hope is to draw attention to the importance of personalization and customization in management science research, and to illustrate some of the more important problems and opportunities for researchers. There do exist several challenges to execute high quality research in these areas. Given the interdisciplinary nature of these issues, researchers must be able to view problems from the different perspectives and be able to bring to bear tools and techniques from the different disciplines in order to make significant contributions. The difficulty involved in doing this well is further compounded by the pace at which technology changes are coming about, that lead to newer and more innovative ways in which firms can personalize.

References

Adomavicius, G. and A. Tuzhilin, (2001), “Using Data Mining Methods to Build Customer Profiles,” IEEE Computer, February 2001, pp. 74-82.

Adler, M., P. Gibbons, and Y. Matias (2000), “Scheduling Space Sharing for Internet Advertising,” Journal of Scheduling, to appear.

Agrawal, Rakesh, Tomasz Imielinski, and Arun Swami (1993), “ Mining Association Rules between Sets of Items in Large Databases” Proceedings of the 1993 ACM SIGMOD International Conference on Management of Data, pp. 207-216.

Ansari, Asim, Scander Essegaier, and Rajeev Kohli (2000), “Internet Recommendation Systems,” Journal of Marketing Research, August 2000, pp. 363-375.

Bakos, Yannis and Brynjolfsson, Erik (1999),”Bundling Information Goods: Pricing, Profits, and Efficiency,” Management Science, Vol. 45, No. 12, pp 1613-1630.

Belkin, Nicholas J., (2000), “Helping People Find What They Don’t Know,” Communications of the ACM, Vol. 43, No. 8, August 2000, pp. 58-61.

Brandenburger, A. M. and B. J. Nalebuff (1995), “The Right Game: Use Game Theory to Shape Strategy,” Harvard Business Review, July-August, 1995.

Brandenburger, Adam and Barry Nalebuff (1996), Co-opetition, Currency Doubleday, New York.

Breese, John S., David Heckerman, and Carl Kadie (1998), “Empirical Analysis of Predictive Algorithms for Collaborative Filtering,” Technical Report MSR-TR-98-12, Microsoft Research.

Brynjolfsson, Erik and Michael D. Smith (2000), “The Great Equalizer? The Role of Shopbots in Electronic Markets,” Working Paper, Sloan School, MIT.

Chan, Yee Man, Jonathan Womer, Jeffrey K. Mackie Mason, and Sugih Jamin, (2000), “One size doesn’t Fit All: Improving Network QoS Through Preference-Driven Web Caching,” Working Paper, University of Michigan.

Chintagunta, Jain and Vilcassim (1991), “Investigating Heterogeneity in Brand Preferences in Logit Models for Panel Data,” Journal of Marketing Research, 28 (November), 417-28.

Chuang, C. I. and Sirbu, M. A. “Optimal Bundling Strategy for Digital Information Goods: Network Delivery of Articles and Subscriptions”. In Information Economics and Policy, 1999.

Csikszentmihalyi, Mihaly (1991), Flow: The Psychology of Optimal Experience, Harper Collins.

Dahan, Ely and V. Srinivasan (1998), “The Predictive Power of Internet-Based Concept Testing Using Visual Depiction and Animation,” MIT, Sloan School of Management, Working Paper, http://web.mit.edu/edahan/.

David, B., R. James, and M. Lattin, “Shopping Behavior and Consumer Preference for Store Format: Why 'Large Basket' Shoppers Prefer EDLP", Marketing Science, Vol. 17, No. 1, 1998.

Degeratu, Alexandru, Arvind Rangaswamy, and Jianan Wu (1999), “Consumer Choice Behavior in Online and Traditional Supermarkets: The Effects of Brand Name, Price, and Other Search Attributes,” Ebusiness Research Center Working Paper, 03-1999, Pennsylvania State University, http://www.ebrc.psu.edu/papers/pdf/03-1999.pdf.

Dewan, Rajiv, Bing Jing, and Abraham Seidmann (1999), “One-to-One Marketing on the Internet,” Proceedings of the 20^th International Conference on Information Systems, pp. 93-102.

Dewan, Rajiv, Bing Jing, and Abraham Seidmann (2000a), “Adoption of Internet-Based Product Customization and Pricing Strategies,” Journal of Management Information Systems, Fall 2000, Vol. 17, no. 2, pp. 2-28.

Dewan, Rajiv, Bing Jing, and Abraham Seidmann (2000b), “Achieving First Mover Advantage Through Product Customization on the Internet,” Proceedings of the 21^th International Conference on Information Systems, pp. 272-285.

Dreze, Xavier and Fred Zufryden (1997), “Testing Web Site Design and Promotional Content,” Journal of Advertising Research, Vol. 37, No. 2, 77-91.

Ehrenberg, A.S.C. (1959), “The Pattern of Consumer Purchases,” Applied Statistics, 26-41.

Fader, Peter and Bruce Hardie (1999), “Forecasting Repeat sales at CDNOW: A Case Study,” Working Paper, June 12, 2000, University of Pennsylvania, http://fourps.wharton.upenn.edu/ideas/pdf/fader-cdnow.pdf.

Geoffrion, A.M. and R. Krishnan (2001), “Prospects for Operations Research in the E-Business Era,” Interfaces, Vol. 31, No. 2, March-April.

Goldberg, D., D. Nickols, B.M. Oki, and D. Terry (1992), “Using Collaborative Filtering to Weave an Information Tapestry,” Communications of the ACM, Vol 35, No 12, December 1992, pp. 61-70.

Gonul, Fusun and Kannan Srinivasan (1993), “Modeling Multiple Sources of Heterogeneity in Multinomial Logit Models: Methodological and Managerial Issues,” Marketing Science, Vol. 12, No. 3, (Summer), 213.

Green, Paul and V. Srinivasan (1990), “Conjoint Analysis in Marketing: New Developments with Applications for Research and Practice,” Journal of Marketing, Vol 54, No. 4, pp 3-19.

Guadagni, Peter and John Little (1983), “A Logit Model of Brand Choice calibrated on Scanner Data,” Marketing Science, Vol. 2 (Summer), pp 203-38.

Hagel, John and Arthur Armstrong (1997), Net Gain: Expanding Markets Through Virtual Communities, Harvard Business School Press, Boston.

Hanson, W. (2000), Principles of Internet Marketing, South-Western College Publishing, 2000.

Haunl, Gerald and Valerie Trifts (2000), “Consumer Decision Making in Online Shopping Environments: The effects of Interactive Decision Aids,” Marketing Science, Vol. 19, No. 1, Winter, 4-21.

Hitt, L.M., and P. Chen (2000), “Customized Bundling: Effective Pricing Strategies for Goods with Low Marginal Costs,” Paper presented at the Workshop on Information Systems Economics, December 2000.

Hitt, L.M., and P. Chen (2001), “Bundling with Customized Self-Selection: A Simple approach to Bundling Low Marginal Cost Goods,” Working Paper, August 2001 (at http://grace.wharton.upenn.edu/~lhitt/bundling.pdf).

Hoffman, Donna, and Thomas Novak (1996), “Marketing in Hypermedia Computer-Mediated Environments: Conceptual Foundations.” Journal of Marketing, 60, July, 50-68.

Jiang, Z., and L. Kleinrock (1998), “Web Prefetching in a Mobile Environment,” IEEE Personal Communications, October 1998, pp 25-34.

Kamakura, Wagner and Gary Russell (1989), “A Probabilistic Choice Model for Market Segmentation and Elasticity Structure,” Journal of Marketing Research, November, Vol. 26, No. 4, p379-90.

Kautz, Henry, Bart Selman, and Mehul Shah (1997), “Referral Web: Combining Social Networks and Collaborative Filtering,” Communications of the ACM, Vol. 40, No 3, March 1997, pp. 63-65.

Kotler, Philip (2000), Marketing Management: The Millenium Edition, Prentice Hall, New Jersey.

Kumar, S., V. S. Jacob, and C. Sriskandarajah (2000), “Scheduling Advertisements at a Web Site,” Tenth Annual Workshop on Information Technologies and Systems, December 2000, pp. 37-42.

Laudon, Kenneth C., (1996), “Markets and Privacy,” Communications of the ACM, Vol. 39, No. 9, September 1996, pp. 92-104.

Leeflang, Peter S.H. and Dick Wittink (2000), “Building models for Marketing Decisions: Past, Present and Future,” International Journal of Research in Marketing, Vol. 17, No. 2-3, 105-126.

Levitt, Theodore (1980), “Marketing Success Through Differentiation-Of Anything,” Harvard Business Review, January-February.

Lynch, John and Dan Ariely (2000), “Wine Online: Search Costs Affect Competition on Price, Quality, and Distribution,” Marketing Science, Vol. 19, No. 1, Winter 83-103.

Manchanda, P, A. Ansari, and S. Gupta (1999), “The Shopping Basket: A Model for Multi-Category Purchase Incidence Decisions,” Journal of Marketing Research, Vol. 18, No. 2, 1999.

Mahajan, Vijay and R. Venkatesh (2000), “Marketing Modeling for E-business,” International Journal of Research in Marketing, Vol. 17, No. 2-3, 215-227.

McCarthy, John, (2000), “Phenomenal Data Mining,” Communications of the ACM, Vol. 43, No. 8, August 2000, pp. 75-79.

Mobasher, B., R. Cooley, and Jaideep Srivastava., (2000), “Automatic Personalization Based on Web Usage Mining,” Communications of the ACM, Vol. 43, No. 8, August 2000, pp. 142-151.

Moe, Wendy (2001), “Buying, Searching, or Browsing: Differentiating Between Online Shoppers Using In-Store Navigational Clickstream,” Working Paper, University of Texas at Austin.

Moe, Wendy and Peter Fader (2001), “Capturing Evolving Visit Behavior in Clickstream Data,” Working Paper, The Wharton School, University of Pennsylvania, http://www-marketing.wharton.upenn.edu/people/faculty/fader.html#working.

Montgomery, Alan (2000), “Applying Quantitative Marketing Techniques to the Internet,” Interfaces, Vol. 31, No. 2, March-April.

Morrison, Donald and David Schmittlein (1988), “Generalizing the NBD Model for Customer Purchases: What are the Implications and Is it worth the Effort?” Journal of Business and Economic Statistics, Vol. 6, No. 2, (April), 145-159.

Mena, Jesus (2001), “WebMining for Profit,” Digital Press, Woburn, Massachusetts.

Nunes, Paul F., and Ajit Kambil (2001), “Personalization? No Thanks,” Harvard Business Review, April 2001, pp. 2-3.

Peppers, Don and Martha Rogers (1997) Enterprise One to One: Tools for Competing in the Interactive Age, Currency Doubleday, New York.

Pollock, Annabel and Andrew Hockley (1997), “What’s Wrong with Internet Searching,” D-lib Magazine, March, http://www.dlib.org/march97/bt/03pollock.html.

Resnick, Paul and Hal R. Varian, (1997), “Recommender Systems,” Communications of the ACM, Vol. 40, No. 3, March 1997, pp. 56-58.

Rossi, Peter and Greg Allenby (1993), “A Bayesian Approach to Estimating Household Parameters,” Journal of Marketing Research, Vol. 30, No. 2, (May), 171-182.

Rossi, Peter, Robert McCullogh, and Greg Allenby (1996), “The value of Purchase History Data in Target Marketing,” Marketing Science, Vol. 15, No. 4, 321-340.

Stone, Bob, and Ron Jacobs (2001), “Successful Direct Marketing Methods,” McGraw Hill Publishers, New York.

Terveen, L., Will Hill, and Brian Amento (1999), “Constructing, Organizing, and Visualizing Collections of Topically Related Web Resources,” ACM Transactions of Computer-Human Interactions, Vol. 6, No. 1, March 1999, pp. 67-94.

Ulph, David, and Nir Vulcan, “Electronic Comerce and Competitive First-Degree Price Discrimination,” Technical Report, University College, London, 2000, http://www.ecn.bris.ac.uk/www.ecnv/welcome.htm.

Ulph, David, and Nir Vulcan, “Electronic Comerce, Mass Customization, and Price Discrimination,” Technical Report, University College, London, 2001, http://www.ecn.bris.ac.uk/www.ecnv/welcome.htm.

Vandermeer, D., K. Dutta, A. Datta, K. Ramamritham, and S. Navathe, (2000) “Enabling Scalable Online Personalization on the Web,” ACM Conference on Electronic Commerce, pp. 185-196.

Varian, Hal R., (2001a), “Differential Pricing and Efficiency,” FirstMonday, Vol.1, No.2, August 5^th, 1996 (www.firstmonday.dk/issues/issue2/index.html).

Varian, Hal R., (2001b), “Economics of Information Technology,” Working Paper, University of California at Berkeley, (http://www.sims.berkeley.edu/~hal/people/hal/papers.html), October 2001.

Villas-Boas, J.M. and Russell Winer (1999), “Endogeneity in Brand Choice Models,” Management Science, 45, 1324-1338.

Vilcassim, Naufel, and Dipak Jain (1991), “Modeling Purchase-Timing and Brand Switching Behavior Incorporating Unobserved Heterogeneity,” Journal of Marketing Research, Vol. 28, (February), 29-42.

Zeithaml, Berry, and Parasuraman (1988), "Communication and Control Processes in the Delivery of Service Quality," Journal of Marketing, April 1988, pp. 35-48.

[1] For expositional simplicity, we use the term products to refer to both products and services in the ensuing discussion.

[2] As one may recall, Amazon experimented with such a pricing mechanism, which they later withdrew due to pressure from its customers.

[3] The strategic role of personalization in the context of third-degree price discrimination has not been examined in extant literature. While one may expect this to be quite analogous to first-degree price discrimination for many kinds of products, this could deserve further reflection.

Sumit Sarkar*

1. Introduction

Figure 1: The Personalization Process Framework