Feature Overload Kaifu Zhang

V. Padmanabhan1

June 12, 2011

1 Kaifu

Zhang is a Ph.D. student in Marketing and V. Padmanabhan is The John H. Loudon Chaired Professor of International Management and Professor of Marketing at INSEAD, 1 Ayer Rajah Avenue, Singapore. [email protected], [email protected]. Preliminary draft. Comments welcome.

Feature Overload

Abstract

Feature overload refers to the phenomenon wherein consumers purchase feature rich products but subsequently don’t use all the features. We try to understand why this occurs as an equilibrium outcome. We focus on two aspects of consumer preference: the uncertainty about feature need and the complexity disutility from too many features. We show that consumer uncertainty creates an option value even if the feature is unused ex-post. In a monopoly setting, the firm offers the feature rich product both to enhance valuation and for pricing reasons. In the later case, feature rich product may be offered even if the overall complexity cost outweighs the option values of additional features. In the competitive case, feature rich product may be offered in two types of equilibria: firms may compete with each other on the number of features, leading to a prisoner’s dilemma situation where both firms offer the feature rich product. Firms may also differentiate on the number-of-feature dimension, therefore engaging in ‘uncertainty based segmentation’. In the later case, competing products use ‘functionality’ and ‘simplicity’ as their respective value propositions, and joint profit is maximized. Interestingly, higher complexity disutility lowers profit in the monopoly case but may raise equilibrium profits in the duopoly case. To provide support on our utility function assumptions, we develop a preference measurement methodology and empirically assess feature need uncertainty in a field study. The empirical results point to both the existence of and the heterogeneity in feature need uncertainty.

1

1

Introduction.

One of the authors had to purchase a camcorder at short notice because he realized the old one was broken and his daughter’s concert was just a few days away. The bewildering array of features within and across brands made the choice process quite difficult. How should one trade-off resolution, image stabilization, lux ratings, video capture formats, image sensors, interface options? Not being a techie and not knowing when he might need a particular feature, he proceeded to buy a Canon Legria with everything bundled in. However, it was not long before he found himself deeply lost in the plethora of options and an encyclopedia-like user mannual. Unfortunately, the author is not alone in his frustration with complex gadgets nowadays. A recent survey by a California-based research company reveals that people over 30 only use 12% of the features in their mobile phones. Moreover, one third of these respondents expressed ’deep frustration’ over their handsets. This ’feature overload’ phenomenon

1

has become an important

source of consumer dissatisfaction and complaints.2 Interestingly, anecdotal evidence suggest that many companies are aware of the feature overload problem as well as the possibility to profit from this problem. Sony, for example, offered to remove the pre-installed softwares in their laptops in 2008. Being aware of consumer complaints over unneccessary softwares, they planned to charge a price premium for the computers loaded with fewer softwares. Some other companies have hit success with simple and usable products. One of the most successful products in the camcoder category is the Flip with a market share of 13% that retails for $130. It just records videos and has no menus, no settings, no optical view-finder, no video light, no special effects, no high definition, no lens cap, no optical zoom and no memory card (New York Times, December 21, 2008). The feature overload phenomenon refers to bundling all the features within a single product. 1 Often

referred to as feature creep in the popular press (Financial Times - November 12, 2005, Business Week April 13, 2006 and New York Times - July 16, 2009). 2 In this study, we formally define feature overload as offering a product with several features even if it is common knowledge that none of the consumers will use all of these features post purchase.

2

Thompson, Hamilton and Rust (2005) in a recent paper provide a very insightful perspective on this phenomenon. They show through a series of experiments that consumer evaluations and choices are systematically influenced by the feature attributes of a product. In short, ex-ante consumers prefer feature rich product but ex-post prefer the less-feature product. In this paper, we attempt to build a consumer utility function that predicts this preference reversal phenomenon, and study firm decision making in an equilibrium framework. A careful examination of the academic and business literature on feature overload suggests that consumer uncertainty and complexity disutility are key drivers of the phenomenon. Uncertainty in this context refers to the ability to forecast likelihood of usage contexts but could as well be about credibility of the claims of value added benefits of features. Complexity disutility in this context refers to the cost on the consumer side as firms include more features in the product. We build a model that incorporates these two elements to answer questions that have not been studied earlier. Specifically, we seek to answer the following questions • What are the implications of consumer uncertainty and heterogeneity for firm’s decisions on feature provision in their product? How does this interact with the firm’s ability to market an assortment of products? • What are the equilibrium implications of the strategic interaction of firms in the context of consumer uncertainty and heterogeneity? Are the drivers behind the ‘feature overload’ phenomenon different in the monopoly case and under compeition? • What are the consequences of endowing firms with the ability to offer optional feature upgrades on equilibrium outcomes and welfare?

Our analysis reveals several key insights into the feature overload phenomenon. In a monopolistic setting, offering a feature rich product represents a basic trade-off between option value

3

and complexity cost. Interestingly, the firm may have the incentive to offer the feature-rich product even when the complexity cost outweighs the average option value provided by the additional feature. This points to the role of the feature-rich product as a device to reduce consumer heterogeneity, since it provides one-model-fits-all insurance against all possible consumption state. In the competitive setting, we find that offering the feature rich product can be either a prisoner’s dilemma outcome or an efficient outcome that maximizes joint profit. In the former case, firms try to outcompete each other by loading more features into their product. This leads to a situation where both firms offer feature rich product, and become worse off due to the heighten competition. When complexity cost is sufficiently large, however, firms can differentiate by offering the simple product and the feature rich product respectively. Simplicity and functionality become their respective value propositions, which lead to more focused positioning and less competition. These findings relate to anecdotal evidences. Interestingly, we observe that the firms’ profit can both raise when complexity cost becomes higher. The rest of the paper is organized as follows. We review the related literature in Section 3. The model setup is presented in Section 4. Section 4 and 5 present our key results and extensions. In Section 6, we propose a measurement technique of consumer uncertainty and apply it in a field study, which intend to illustrate the key factor underpins our analytical model. Section 7 concludes.

2

Related Research

This paper is related to several streams of literature. The effect of product features on individual choices has been extensively studied in the consumer behavior literature. A robust finding was that more features are not always better. Simonson, Carmon, and O’Curry (1994) illustrated how additional features make consumers less likely to choose a product, even if these features clearly does not diminish the value of the brand. Mukherjee and Hoyer (2001) articulated the link between additional features and product complexity. They demonstrated that additional features may 4

decrease product evaluation for high complexity products. Moreover, the literature suggests that when choosing between feature rich product and simple product, consumers exhibit a preference reversal tendency. Thompson, Hamilton, and Rust (2005) provided a systematic experimental examination of how consumers’ product evaluations and choices are influenced by the number of product features. Their study revealed a preference reversal phenomenon: Before actual usage, consumers prefer the more feature products. Once having used the products, however, they tend to prefer the less-feature products. Furthermore, when the consumers are asked to rate the products on capability and usability, they give inconsistent ratings before and after usage. Meyer, Zhao, and Han (2008) find evidence of a valuation-usage disparity for product capabilities. Consumers have high willingness to pay for products with expanded set of features, but do not use these features ex-post. The above studies provided the behavior basis for our consumer utility function. In the next section, we will introduce the consumer utility function in our model. Our formulation of consumer utility function will predict the preference reversal phenomenon as in (Meyer et al. 2008, Thompson et al. 2005). Our study does not investigate the role of ’biases’ in explaining the feature overload problem. We show that feature overload emerges without the introduction of any behavior bias. Behavior ‘biases’ might further exacerbate the problem. For example, Shin and Ariely (2004) demonstrates that people are willing to make significant investment to ’keep options open’, even when the options themselves seem to be of little interest. Della Vigna and Malmendier (2006) demonstrates overconfidence about personal efficiency and self control may drive consumer purchase gym plans which they subsequently under utilize. Meyer et al. (2008) explain the disparity between valuation and usage with an intertemporal choice model, in which consumers are subject to hyperbolic discounting. Our study is also related to the economics and marketing literature on preference uncertainty. Preference uncertainty arises when consumers have state dependent utility, and are uncer-

5

tain about the states. It is a driver to various consumer behavior, such as brand loyalty (Villas-Boas 2004) and the choice of service contract and multi-part tariff. (Lambrecht, Seim, and Skiera 2007, Narayanan, Chintagunta, and Miravete 2007). The literature on preference uncertainty suggests a link between state dependent preference and the need to maintain consumption flexibility. For example, Hauser and Wernerfelt (1990) has shown that a consumer’s optimal consideration set usually contains more than one item when preference uncertainty is important. Guo (2006) find that consumers may purchase multiple products in face of state dependent utility. As a result, firms cease to compete with other, and larger product differentiation can lead to smaller profit. In similar veins, we argue that consumers may have state dependent utility for the features, and their preference for the feature rich product is driven by the need to maintain flexibility. In this respect, our model shares the logic with (Guo 2006), while the ’multiple buying’ behavior is on the product feature level. The main difference between our study and (Guo 2006) is our assumption of a ’base product’, which makes it too costly for the consumers to buy multiple base products with different features 3 . In the service literature, a few mechanisms have been proposed to leverage consumer preference uncertainty for greater profit, such as advance selling (Xie and Shugan 2001), service upgrade (Biyalogorsky, Gerstner, Weiss, and Xie 2005) and allowing service cancellation (Xie and Gerstner 2007). Advanced selling refers to selling the service to consumers before the resolution of their preference uncertainty. Service upgrade allows the consumers to upgrade their pre-purchased services when they observed the state of nature. Service cancellation gives the consumers the option to cancel their service, and the firm is able to profit by both charging cancellation fees and resell the freed capacity. All these instruments depend on the difference between ex-ante consumer valuation and ex-post consumer valuation, and in this light, share the same insight as our work. Finally, our study is related to the economic literatures of bundling and product line competition. Bundling (Bakos and Brynjolfsson 1999, Matutes and Regibeau 1992, McAfee, McMillan, 3 For

example, consider a GPS navigator producer who is offering a altimeter and a floating case as features. If the price of the GPS is high, the consumers will never buy a GPS with a altimeter and another GPS with a floating case.

6

and Whinston 1989, Nalebuff 2000, Stremersch and Tellis 2002) is defined as ’the sale of two or more separate products in one package’ (Stremersch and Tellis 2002). ’Separate’ products are products for which separate markets exist. In the bundling literature, the firm faces the choice between selling separate products in different markets, or selling the bundle. In our model, features are not sold as independent products.

4

The absence of multiple markets, and the presence

of a base product make the feature overload problem conceptually different from a bundling story. However, as we will detail in the result section, some of our results share the same intuitions behind the bundling phenomenon. In considering multiple product competition, our model is related to the product line rivalry literature (Brander and Eaton 1984, Gilbert and Matutes 1993, Klemperer 1992, Verboven 1999). These authors found that firms may either offer interlaced product lines or identical product lines, depending on the degree of brand-level differentiation. The rest of the paper is organized as follows. The next section develops the basic model. Section 4 analyzes firms’ product decisions in both the monopoly scenario and under competition. Section 5 presents an extension of the model in which we allow the firms to sell optional feature upgrades after selling the base product. In section 6, we introduce the measurement technique and access consumer uncertainty in a field study.

3

The Model

3.1

Consumers

We consider consumers who derive consumption values from both the base product (e.g., a GPS navigator) and the product features (e.g., a built-in altimeter or a floating case). The consumers have state independent preference for the base product and state dependent preference for the features. When they purchase the feature rich product, they incur a state independent complexity 4 Consider

our GPS example again. While this looks similar to bundling the GPS navigator with the altimeter, we focus on a case where the GPS producer doesn’t become sellers of both GPS and standalone altimeter. Each consumer need only one unit of the base product, and additional features only influence their valuations of the base products.

7

cost. Let’s assume that the firm offers a base product, which could include two additional features, a and b. A consumer’s valuation of each feature is state dependent and cannot be foreseen at the time of purchase. There are two states of nature, Ω = {A, B}. Feature a will be useful only in state A, and feature b only in state B. In our GPS example, the base product is the GPS navigator, and the features a and b correspond to the altimeter and the floating case. The consumer need for the features depend on the type of the activity: the altimeter is useful only if the consumer uses the GPS for hiking, while the floating case is useful if she uses the GPS for surfing. Consider a consumer who is still deciding where she will live for the next year: a seaside city in southern France or the mountainous region of Switzerland? Depending on the location, she will engage in one activity or the other (hiking or surfing), and will therefore need one feature or the other. The above example illustrates a case where feature need uncertainty arise from a consumer’s uncertainty about the consumption context. The uncertainty can also arise from the lack of technical knowledge. For example, a consumer may consider a built-in camera in a laptop computer useful. However, will the built-in camera deliver acceptable video quality? The consumer may be uncertain about this if she is not familiar with computer cameras in general. These sources of uncertainty are particularly important in fast developing markets of technical products, such as the consumer electronics market. We consider three possible product offerings: the product with only feature a, the product with only feature b, and the product with both features. For simplicity, we denote a product by the features it has: product a (b) refers to the product with only the a (b) feature,and product ab refers to the product with both features. We denote a consumer’s state dependent valuation of product i under state ω as v(i, ω). For example:

 v(b, ω) = 8

v ω =B v ω =A

For product a, the consumer valuation is v only if the state of nature is A. This formulation of state-dependent preference is widely used in the theoretical works on preference uncertainty(Guo 2006, Villas-Boas 2004, Xie and Shugan 2001). For product ab, the consumer has to bear a state independent complexity cost. However, he has the flexibility to pick which feature to use after the state of the world is observed. Formally, v(ab, ω) = max{v(a, ω), v(b, ω)} − γ Given the above notations, v is the state independent utility from the base product. The parameter γ stands for the complexity disutility associated with the feature rich product. Complexity decreases consumer valuation of the product, either for engineering reasons (the basic functionality of the product degrades as more features are included) or usability reasons (product becomes more difficult to learn and operate). Finally, such cost can be a result of consumer perception (Chernev 2007): where a ‘generalist’ product is perceived as inferior on each dimension to specialist products even if the features offer the exact same functionality. We assume that the consumers are aware of the cost associated with feature-rich product before actual usage, following the empirical findings by Thompson et al. (2005). We capture consumer uncertain by a belief parameter, θ . θ is the consumer’s belief that the state of nature is B. If θ is close to 0 or 1, the consumer is relatively certain about his feature need. If θ is close to 0.5, the consumer is highly uncertain 5 . We assume that the consumers are heterogeneous with respect to their uncertainty. A dedicated surfer is certain that he is not going to use the altimeter in the GPS, while a person who is interested in both surfing and hiking is uncertain about his feature need if he is still deciding which city to stay and which activity to learn. Without loss of generality, we assume θ is uniformly distributed on [θmin , θmax ] with density 1. Since we do not restrict θmin + θmax = 1, this setup allows 5 Alternatively,

θ could be interpreted as the expected usage frequency of this feature. Under this interpretation, the consumer indeed need both features in the long run, although he will not use both features simultaneously in any time period. See Guo (2006) for further explaination on this interpretation.

9

us to capture the asymmetries in feature need. We call θ the type of consumers. Will consumers prefer the feature rich product or the simple product? Net of price, the ex-ante and ex-post preference can be illustrated by the following figure: Figure 1: Ex-ante and Ex-post Product Valuations

Figure 1 illustrates the valuations for the possible products. For a sufficiently uncertain consumer, the ex-ante valuation of the ab product exceeds that of either the b or the a product. However, after the resolution of uncertainty, the consumer strictly prefers one of the simple product to the ab product. This pattern of individual preference is consistent with the preference reversal phenomenon documented in previous behavior literatureThompson et al. (2005). In the case of firm competition, we further assume that consumers have heterogeneous brand preference. This captures the product differentiation due to the factors other than product features. Firm 1 is located at x = 0 and firm 2 is at x = 1. Consumers are uniformly distribution on [0, 1] with ‘transportation cost’ t. Overall, the distribution of consumers is uniform with f (x, θ ) = 1, (x, θ ) ∈ [0, 1] × [θmin , θmax ]. For ease of exposition, we choose to present the results when θ is distributed from [θmin , 1]. This special case is able to provide all the main insights. A few important caveats apply to our model of consumer utility function. First, we do not assume heterogeneity in complexity cost. One important feature of many consumer electronic markets is that consumers have largely heterogeneous technical expertise. The cost of complexity (i.e., more product features) is higher for novice consumers and lower for experts. Our initial analysis of this case indicates that heterogeneity in complexity cost is a moderator of our results. Second, we assume a Bernoulli distribution for feature need and a uniform distribution for consumer uncertainty. In reality, consumers may be certain about their need for one feature yet uncertain about another. Moreover, they may need multiple features simultaneously or none at all. In addition, the distribution of consumer belief may be non-uniform. These are important possibilities we do not 10

consider in our simplified model. Our analysis leads us to believe that our main results will remain valid under two conditions: the consumers’ need for feature are state-dependent, and consumers are heterogeneous with respect to their uncertainty about state. Intuitively, these condition means that the feature rich product provides ‘all-wheather’ insurance against various consumption needs, but the value of such insurance is different for each consumer. It implies that the ex-ante valuation of the feature rich product should be close to its ex-post valuation, while the ex-ante and ex-post valuation of the simple product are more likely to be different. Empirical findings from previous literature do seem to support this implication (See Table 1 of (Thompson et al. 2005)).

3.2

Firms

The firms have constant marginal cost c for product a and b, and c for product ab. We assume that c is sufficiently high compared to γ, and the difference between c and c is small. This is true when the base product is relatively costly. Consider the example of the GPS navigator, this assumption says that the production cost of the GPS device is much higher than the cost of adding the altimeter or the floating case. In the monopoly case, we consider a two stage game. In the first stage, the firm chooses its product offering and set prices. In the second stage, the consumers purchase their most preferred product. The expected surplus from product i for consumer type θ is: Uθ (i) = (1 − θ ) ∗ v(i, A) + θ ∗ v(i, B) − pi Here, pi is the price of product i. We assume consumers are risk neutral expected utility maximizers. When firms offer both ab and b products, for example, a consumer will prefer ab to b iff Uθ (ab) = v − γ − pab > Uθ (b) = (1 − θ ) ∗ v + θ ∗ v − pb From the locations of the marginal consumers, we can derive the demand for each product, 11

and obtain the firm’s profit function. In the competitive case, we consider a three stage game. In the first stage, the firms decide on their product offerings. In the second stage, the firms set their prices. In the third stage, consumers choose which product to purchase. In this case, a consumer preference depends on both the product features and his horizontal location. For example, a consumer located at (x, θ ) derives a utility from consumption of product b from firm 1:

Ux,θ (b) = (1 − θ ) ∗ v + θ ∗ v − t ∗ x − p1b Similarly, he derives the following utility from consuming product ab by firm 2: Ux,θ (ab) = v − γ − t ∗ (1 − x) − p2ab The marginal consumers who are just indifferent between the product offerings can be derived accordingly. The marginal consumer is defined by an implicit function of x and θ . For each product offering, we obtain a large number of possible demand schedules depending on the parameters. The derivations are detailed in the appendix. Our model can be considered a product differentiation model with two dimensions of consumers heterogeneity. Different from a simple model (e.g., Hotelling) of 2-D product differentiation, the differentiation between the feature rich product and the simple product can be either horizontal or vertical. The nature (i.e., vertical vs horizontal) and degree of differentiation is driven by both consumer uncertainty and complexity cost. Conceptually, we believe these two factors captures the specificities of the number-of-feature decision. Technically, these factors create complexities beyond the simple model of 2-D product differentiation. We provide detailed explaination in the analysis section.

12

4

Analysis

Table 1 represents our roadmap of analysis. We vary our model assumptions along two dimensions: first, firms can offer either a single product or multiple products. Second, firms may be either monopoly or in competition. Economies of Scale? Single-Product

Multi-Product

Monopoly

Section 4.1

Section 4.3

Duopoly

Section 4.2

Section 4.4

Competition?

Table 1: Modeling Roadmap

The single-product versus multi-product distinction captures an important aspect of firms’ cost structure. When firms face strong economies of scale in production, producing several different models will dramatically increase the average production cost. When economies of scale is weak, firms may produce multiple products to satisfy diverse consumer tastes. Although we do not model economies of scale explicitly (i.e., model the average cost as a function of quantity produced), considering these polar cases sheds light on the cost side of the feature overload decisions, therefore complementing our emphasis on the demand side explanation for this phenomenon. The monopoly versus duopoly distinction captures the level of competition in the marketplace. Our results indicate that competition dramatically changes the firms’ incentives of offering feature rich products.

13

4.1

Single-Product Monopoly

4.1.1

Equilibrium Results

A single-product monopoly faces a choice between offering the simple product a, b or the feature rich product ab. The firm always prices the product to extract as much consumer surplus as possible. When consumers have heterogeneous beliefs θ about the state of nature, their valuation of the product features are also heterogeneous. Ex-ante, both features may have option values even if a consumer knows he will need only one feature post purchase. These intuitions translate into two drivers behind the firm’s incentive to offer the feature rich product: • Option Value The firm offers the feature rich product because all product features provide consumers option values. Higher valuation leads to higher price and profit. • Reduced Consumer Heterogeneity While consumer valuations for the simple product is state dependent, their valuations for the feature rich product is state independent. Put differently, the feature rich product provides ‘consumption insurance’ in every state of nature. When consumers are heterogeneous with respect to their beliefs but not on the features’ ex-post usage value, uniform pricing is a more effective tool to extract consumer surplus.

The following proposition characterizes firms’ product decision. Proposition 1. A single-product monopoly offers the simple product b if (1 − θmin )(v − v) > γ + c − c, and offers the feature rich product ab otherwise. Figure 1 illustrates the above proposition. We plot the optimal decision as a function of θmin (consumer heterogeneity) and

γ (1+θmin )(v−v)/2

value).

14

(the ratio of complexity cost to average feature

As figure 1 illustrates, the monopolist is more likely to offer the feature rich product when the cost/value ratio is low and θmin is small. Smaller cost/value ratio and smaller θmin both lead to greater option value, and the heterogeneity reduction effect is strong when consumers are more heterogeneous (as in the case of small θmin ). Observation 1. A single-product monopoly may offer the feature rich product even when γ (1+θmin )(v−v)/2

> 1.

One interesting observation is that the firm may offer the feature rich product even if the cost/value ratio is greater than 1. This represents a case where the large complexity cost overweights the option value associated with more product features. This result is driven by the heterogeneity reduction effect of feature-rich product, which provides the same expected consumption value regardless of consumer uncertainty. Although total consumer surplus becomes smaller due to the complexity cost, the firm can extract surplus more effectively. The intuition behind this finding is similar to that in the bundling literature (Bakos and Brynjolfsson 1999, McAfee et al. 1989), where the bundling of multiple product reduces the heterogeneity in consumer willingness to pay. The intuition carries through even when seperate markets for the product features do not exist. One caveat applies to the above results. While feature rich product provides option values as long as consumer are uncertainty about feature need, the heterogeneity reduction effect is driven by our assumption that the consumers are heterogneous with respect to their uncertain beliefs, but they agree on the usefulness (i.e., value) of a feature once they know they need it. As a result of this assumption, consumers have state-independent and homogeneous valuations of the feature rich product. While this assumption might not hold exactly in reality, we believe that the heterogneity reduction effect of feature rich product will remain true (albeit weaker) in most situations, especially when the number of feature is large 6 . The important take-away is that the feature rich product is not only offered to enhance consumer valuation, but also for pricing reasons. 6 see

(Bakos and Brynjolfsson 1999) for an argument in a similar context, based on the Central Limit Theorem

15

4.1.2

Anecdotal Observations

The single-product monopoly is perhaps an oversimplified description of real world scenarios. The single-product monopoly model, nevertheless, sheds light on the basic trade-off driving the firm’s product features decision. Recent press stories provide an abundance of cases where companies make major effort in reducing their product features, after keeping adding features for many years. An examination of these stories reveal on interesting pattern. The process of including more and more features is often an effort to accomodate diverse or conflicting consumer needs, while the decision to reduce features is often a result of accumulating consumer complains about feature need. A dominant market leader is often observed in the software markets, which exhibit strong network effects. Consider the case of the voice-call software, Skype. Skype 1.0 was a tremendous success praised for its extreme simplicity and great functionality. In the subsequent versions, the development team started to incorporate more features and third-party ‘Skype Extra’ applications in the software. As the complexity grows, consumer complaint has been heard. In a dialogue7 between a few Skype users and an engineer from the development, the users complained about too many unused features, which consumed system resource and lowered the reliability of simple voice calls. The development engineer, on the other hand, defended these features by claiming that the features are incorporated because of stated needs from users. Moreover, potential consumers seem to be in support of more product features when they are surveyed. Nevertheless, in response to the users’ demand for simplicity, Skype has started reducing the number of features in the most recent versions. It has terminated the ‘Skype Extra Developer’ programs and taken out some of the in-house developed applications. 7 http://forum.skype.com/index.php?showtopic=92264

16

4.2 4.2.1

Single-Product Duopoly Equilibrium Results

To model firm competition, we consider heterogeneous consumers with respect to both feature need and brand preference. We first provide intuitions on the consumers’ choice problem, which drives the firms’ product decisions. When competing firms offer different products, consumer choices are driven by both brand preference and feature need. The relative magnitudes of t and v − v determine which factor will drive consumer choice in equilibrium. When t is relatively large, brand preference is the major factor driving consumer choices. The firms divide the market along the horizontal dimension regardless of their product choice. In equilibrium, the consumers who prefer one brand strongly (x=1) will never buy the other brand regardless of the product feature. Similarly, when v − v is relatively important, feature need will drive consumer choices in equilibrium. If both the simple product and the feature rich product is offered, an uncertain consumer will never buy the simple product regardless of its brand. The above observation is formalized in Lemma 2 in the appendix. When firms are in competition, they seek to maximize the degree of product differentiation. When brand preference is not sufficiently important, firms can engage in uncertainty-based segmentation: one firm sells ab to the uncertain consumers, while the other firm sells b to the certain consumers. As such, we identify two major incentives for a firm to offer the feature rich product: • Option Value: When brand preference is driving consumer choices, offering the feature rich product does not increase product differentiation, but may lead to higher consumer valuation and therefore larger market share. • Uncertainty-Based Segmentation: When feature need is driving consumer choice, firms can increase differentiation by offering different feature configurations. The feature rich product is able to target the uncertain consumers when the competitor offers the simple 17

product and targets the certain consumers.

Overall, the equilibrium strategy choice is described by Proposition 2. Proposition 2. In a duopoly case, the equilibrium outcome is determined by three parameters: θmin , β = v − v, γ. The second parameter is the relative importance of feature need (compared with brand differentiation). The equilibrium is characterized by the following: When feature need is not important, firms will offer identical products,

(a) Both firms will offer the feature rich product (ab) if the complexity disutility is small; (b) both firm will offer the simple product b if the complexity disutility is large;

When feature need is important, firms will differentiate by including different features in their products.

(c) When θmin is small or complexity disutility is large, firms offer simple products with different features; (d) When θmin is large and complexity disutility is intermediate, firms offer the simple product and the complex product respectively. Figure 2: Equilibrium Product Offering of a Single Product Duopoly

The exact ranges where each outcome occurs as equilibrium are given in the appendix. Figure 2 illustrates the above proposition. Our analysis reveals that there are two situations where the ab product is offered in equilibrium. Interestingly, we find that offering feature rich product may be either a prisoner’s dilemma outcome or a socially efficient equilibrium.

18

Observation 2. The ab − ab equilibrium is a prisoners’ dilemma situation: both firms are better off if they offer the simple products, but ’both overload’ emerges as an equilibrium outcome. The ab − b equilibrium maximizes joint profit. Firms will offer the feature rich product in two types of equilibrium. In the ’both ab’ equilibrium, feature need is less important compared to brand preference, and complexity disutility is low. As a result, brand preference will be driving consumer choice. The firms cannot increase the degree of differentiation by offering products with different feature configurations. Compared to the simple product, the feature rich product receive lower valuation from the certain consumers but higher valuation from the uncertain consumers. When complexity disutility is small, the benefit of offering the feature rich product will exceeds the cost. As a result, both firms offer the feature rich product in equilibrium. The benefit from the feature rich product will be competed away, and the equilibrium profit is no more than the profit if both firms offer the simple product. In the uncertainty based differentiation equilibrium, firms offer ab and b respectively. The joint profit is maximized in equilibrium. Said differently, when feature need is important, uncertainty based segmentation is more effective than the horizontal brand differentiation. This result leads to the following observation. Observation 3. For some parameter combination θmin , β , there exists γ1 > γ2 , where the equilibrium profits for both the competitors are higher with γ1 . Profits are increasing in complexity disutility. The above result indicates that complexity cost can actually increase firm profits in the competitive case. This counterintuitive result takes place when the feature need is important compared to brand preference. In this scenario, the joint profit is maximized when firms engage in uncertainty based segmentation by offering the simple product(b) and the feature rich product(ab) respectively. However, without complexity disutility, b − ab differentiation corresponds to a case 19

of vertical differentiation. All the consumers strictly prefer the feature rich product. As such, the profit is not divided evenly between the competitors. The simple product is perceived as inferior to the feature rich product. Consequently, ab − ab emerges as an equilibrium, leading to a prisoner’s dilemma situation. Complexity disutility qualitatively change the nature of feature based differentiation. It turns the ab −b vertical differentiation into a type of horizontal differentiation. While the uncertain consumers still prefer the feature rich product, the most certain consumers will find the option value from the additional feature too small to justify the complexity cost associated with the feature rich product. In equilibrium, the competitors can segment the market based on consumer uncertainty, wherein the simple product is perceived as different from the feature rich product. This leads to higher profits for both firms. 4.2.2

Anecdotal Observations

The above analyses resonate with the anecdotal evidences from many product markets. In many markets of high tech products, adding features become more and more feasible with the progress in design and manufacturer technologies. Firms start to compete by keeping up with each others’ product features. In the MP3 player category, for examples, leading competitors offer largely similar product lines in terms of the feature configuration. The pairwise correlations between the product lines of four leading brands range from 0.86 to 0.98 8 . Our analysis reveals that this may be considered a ‘feature arm-racing’ pattern of product competition, and may be detrimental for all competitors. As firms put more and more features into their products, complexity and usability become 8 We

acquire product line data from a consumer review website, epinions.com. We consider the top four features which most number of products include. Thus, there are sixteen possible products in terms of feature configuration. The product line of each firm is defined by a distribution function of their products over the feature configurations. The correlation between the product lines of two brands equals one if for every possible feature configuration, the two firms have identical percentage of their products. The correlation is zero if two firms offer completely non-overlapping product lines.

20

a more important consideration. In many markets, consumers have clearly divided preferences for simple and feature rich products. Consider a discussion thread hosted on Google community 9

about Google talk vs Windows Messenger. The participants gave reasons why they favored

different IM messengers. Most people favor GTalk because its ‘simple interface’, and some favored Windows Messenger because it has ’nice features’, ’emoticons’ and ’video call’ function. Our analysis suggests a case where firms’ profits raise as a result of higher complexity disutility. This result partly depends on the assumption of a duopoly without entry. In practice, we often observe that as products become more sophisticated and complexity cost becomes a greater concern, startups or incumbents spot novel market oppurnities by introducing extremely simple products. In 2004, the Swiss mobile operator, Orange, launched a three-button mobile phone called Mobi-Click or ’The mobile for grandparents’. The model was a huge success among the elder people: consumers who are certain about their feature need and who find complexity extremely costly. Released in 2006, the Flip camcorder by Pure Digital is an example of simple product that has hit a mass market success. The company’s best selling model, Flip Extra, has been praised as the “World’s simplest video camera” and won a market share of 13 percent in 2007. Revenue at Pure Digital grew more than 44000 percent by the end of 2008, the highest rate among Silicon Valley firms. In more and more product categories, both entrants and market incumbents start to use simplicity as a key point of differentiation.

4.3

Multi-Product Monopoly

Next, we analyze the product line decision of a multi-product monopoly. The major insights from the single-product case remain valid. In addition, our analysis reveal an additional driver to a multi-product firm’s decision to offer the feature rich product, which we call Uncertainty Based Price Discrimination. Recall from the previous sections, when consumers have heterogeneous 9 http://www.googlecommunity.com/about9101-0-asc-0.html

21

belief about the state of nature, they also have heterogeneous preference about the simple product versus the feature rich product. When the feature rich product is offered together with the simple product, the consumers will self select to purchase their preferred product. To summarize, the firm has three major incentives to offer the feature rich product: • Option Value The firm offers the feature rich product because all product features provide consumers option values. Higher valuation leads to higher price and profit. • Consumer Heterogeneity Reduction While consumer valuations for the simple product is state dependent, their valuations for the feature rich product is state independent, since the feature rich product provides the needed feature in every state of nature. State independence leads to more homogeneous preference when the consumers are heterogeneous with respect to their preference uncertainty. • Uncertainty Based Price Discrimination The feature rich product is offered together with the simple product, allowing better market segmentation. We describe the firm’s product line decision in the following proposition. When launching products are costless, the firm always offers a full product line. Thus, we introduce Fab , the fixed cost of offering the feature rich product, into the analysis. To illustrate the Uncertainty Based Price Discrimination effect, we compare the case where the firm can observe the belief of each consumer to the case where types are unobservable. In the later case, the firm has stronger incentives to offer multiple products. Proposition 3. Denoting Fab as the fixed cost of introducing the feature rich product, the multiproduct monopolist’s product line decision can be described by the following table: Fab < C1 C1 < Fab < C2 C2 < Fab Product Decision:Introduce {ab} or Not? Types Observable Yes No No Types Unobservable Yes Yes No Fixed Cost of ab

22

C2 is the incremental benefit of offering ab when consumer type is unobservable, and C1 is the incremental benefit of offering ab when price discrimination is possible. C2 > C1 and the exact expressions are provided in the appendix. When the fixed introduction cost is intermediate, the option value alone is insufficient to justify the introduction of product ab. However, better price discrimination makes the feature rich product valuable. Thus, the firm has greater incentive to offer the feature rich product when price discrimination is not possible, since offering a menu of products helps the firm to screen consumers based on their uncertainty about feature need, which facilitates price discrimination. In such cases, the firms may charge lower price for the feature rich product. The above finding is not surprising considered in the context of the economics literature of asymmetric information and screening. We argue that offering the feature rich product can be a mean to screen the consumers based on their feature need uncertainty. Observation 4. The firm may charge either a higher or a lower price for the feature rich product, compared to the simple product. When the firm offers multiple product with different feature configurations, the feature rich products may not necessarily be considered superior and sold at higher prices. Instead, it is considered as a ‘generalist’ product and comes cheaper than the ‘specialist’ product which is perceived to excel on a certain dimension. This resonates with the empirical findings by Chernev (2007). The result also points to one of our recurring theme: the feature rich product may not lead to higher willingness to pay on the consumer side. Insead, it may be offered for pricing reasons.

4.4

Multi-Product Duopoly

In this section, we model the product line rivalry (Brander and Eaton 1984, Gilbert and Matutes 1993, Verboven 1999) of multi-product firms. In the first stage of the game, the firms choose their 23

product portfolio. There are 23 = 8 possible product portfolios, and each product portfolio is a subset of {a, b, ab}. In the second stage, firms simultaneously set prices for all their products and consumers decide which product to buy. In the third stage, consumers learn their feature need and get consumption benefits. When firms offer several products, the substitution pattern is much more complicated. When a firm lowers the price of a certain product, it not only steals customers from the competitor, but also attracts the buyers of other products in its own product line. This intra-firm profit interaction (Cabral and Villas-Boas 2005) will creates a more complicated scenario, especially in our asymmetric setting. The complete analysis of the multi-product equilibrium is technically prohibiting. Instead, in previous studies (Brander and Eaton 1984, Gilbert and Matutes 1993) on multi-product competition, the authors focused on the following questions: whether the competing firms will offer interlaced product lines or identical product lines? In the context of feature configuration, we provide answer to this question in the following proposition. Proposition 4. When the ratio

v−v t

is sufficiently large, the firms will offer non-overlapping product

lines. The equilibrium product line configuration is similar to the single product line case. In particular, firms offer ab and b products respectively when γ is not too large and θmin is close to 1. Otherwise, the firms offer a and b product respectively. When the ratio

v−v t

is sufficiently small, the firms will offer identical product lines. In

particular, both firms will offer a, b and ab if the complexity cost is small and θmin if small. As expected, the above results confirm the main insights from the single product section. In particular, we find that competing firms may either offer identical product lines and differentiate on the ‘brand’ attribute, or offer products with different features. In the later case, simplicity may imply higher quality for some consumers, and may be used as a point of differentiation.

24

5

Measuring Consumer Uncertainty: Method and Empirical Evidences

In this section, we intend to provide an empirical examination of our modeling assumptions. While the existence of complexity disutility is well supported by the previous behavior literature, feature need uncertainty and the heterogeneity in such uncertainty have not been illustrated. Since these assumptions drive our equilibrium results, we develop a measurement technique that attempts to quantify feature need uncertainty and provide empirical supports to the heterogeneity of feature need uncertainty in a field study. We provide a brief description of the method and the results from a field study. Details about the methodology as well as simulation validation results are delegated to the appendix.

5.1

Description of the Method

In our theoretical model, we assumed that the consumers will need one and only one of two features. This was a deliberately made choice to illustrate that firms may include two features into one product even if it’s common knowledge that no consumer will need both features simultaneously. In this section, we allow that consumers have independent need for each feature, so that they may need both features at the same time or neither feature at all. We propose a preference measurement method that explicitly estimate the consumers’ uncertainty about feature needs. The method is based upon the standard conjoint analysis methodology. We focus on estimating a linear utility function, such as:

k=K

ui j =

∑ θik ∗ wˆik ∗ x jk − wip ∗ p j + εi j

k=1

where x j1 to x jK are binary variables for the levels of features, and p j is the price of the jth option. wˆin is the ex-post usage value of feature k. θin is the probability that the consumer will need 25

the feature or the estimated long term usage frequency 10 . When consumer is uncertain about his feature need, win = θin ∗ wˆin corresponds to the expected value of the feature. The term εi j represents a random error term. Based on revealed preference data, such as stated choice between different product profiles, traditional conjoint analysis methods are able to estimate win for each feature. However, disentangling θin and wˆin becomes challenging. Put differently, it is difficult to infer ex-post feature valuation (e.g., usage value) and feature need uncertainty from the estimated option value. In order to disentangle the option value of a product feature and the ex-post usage value, we introduce two types of product profiles into the choice questions. In the first type of product profiles, the product is described by its built-in features and a price; in the second type of product profiles, the features are explicitly offered as options, which is not included in the base product, but available as an upgrade. By observing a consumer’s stated choices between product profiles, we infer his ex-ante willingness to pay for the built-in feature, and his ex-post willingness to pay for the upgrade. When consumers are uncertain about their feature need, the ex-ante WTP is lower than the ex-post WTP. We can infer the feature need uncertainty from this discrepancy in willingness to pay. In the appendix, we provide a detailed description of the preference models and the estimation method which underly the above intuition. The methodology has three key features: • When a product feature is offered as an option, the consumer may or may not exercise (e.g., upgrade) it even if he needs it. We predict the (post purchase) upgrade decision based on a given set of utility parameters. • Based on the predicted usage decisions, we infer the ex-ante utility and predict the choice behaviors, based on a given set of utility parameters. 10 The

long term usage frequency results from either uncertainty about usage in each period (As in Guo (2006)) or infrequent but certain needs. These represent cases that are mathematically equivalent in our framework

26

• By minimizing the deviation of the predicted choices and the observed choices, we estimate the utility parameters, including consumer uncertainty and the ex-post WTP. As a methodological note, our method is closely related to the Discrete/Continuous choice models in economics (Hanemann 1984) and marketing (Iyengar, Jedidi, and Kohli 2008). In the Discrete/Continuous framework, the researcher observes the consumers choice behavior but not the post-choice consumption decision (which is usually represented by a Continuous variable). By assuming that consumers make consumption decisions to maximize utility conditional upon choice, the researcher is able to estimate utility parameters and predict both choice (Discrete) and consumption (Continuous) decisions of the consumers. In our framework, we treat the post-choice feature upgrade decision as a binary random variable and model its impact on choice. When the consumer uncertainty parameter θ is treated as a continuous consumption variable, there exists a mathematical equivalence between our method and that developed by Iyengar et al. (2008)11 .

5.2

Field Study

In this section, we briefly present some initial results from a field study. We measure consumer uncertainty about several value-added services in mobile phone plans based on a standard CBC study. We include a total of four non-price product features: ’unlimited video and picture messages’(MMS), ’the MobiTV service’, the ’Mobile E-mail PLUS’ feature, and the ’Happy Roamer’ feature. For detailed description of these features, please refer to the appendix. We measured consumer uncertainty about the ’Mobile E-mail PLUS’ feature and the ’Happy Roamer’ feature. A priori, we conjecture that consumers are uncertain about both features. All the product features have two levels, 1 when the feature is included in the plan, 0 otherwise. Price has two levels for the basic service plan. When any of the feature is offered as an upgrade option, there is an upgrade price to be paid. The upgrade price has three levels. 11 A

formal illustration is out of the scope of this paper. We refer interested readers to the appendix and the methodology section of Iyengar et al. (2008) for detailed information

27

The service plan requires a one year contract. Thus, the consumers cannot switch between different service plan during the course of usage. They can only obtain consumption flexibility by including the feature upgrade option. For the briefing text for the choice questionnaire, please refer to the appendix. We estimate consumer part-worth based on eight questions without the upgrade option. In the second stage, we use four questions each to measure consumer uncertainty about the abovementioned features. The two features are never offered both as upgrade options simultaneously. We present the estimation results in the following tables: Attributes: Estimated Partworth

MMS MobiTV 0.1317 0.2473

Mobile E-mail Plus 0.6577

Happy Roamer 0.5896

Price -0.3759

Table 5. Estimated Partworths

Table 5 presents the population average of the estimated partworth. Overall, the subjects find Mobile E-mail PLUS and the Happy Roamer as the most important features. The negative price parthworth means a lower utility when the price level is 1 (high price). Based on these parthworth, we estimate consumer uncertainty in the second stage estimation task. The data exhibit two stylized patterns: First, for otherwise identical options, many subjects prefer the option where the feature is offered as an upgrade, even if the total price (e.g. the price paid if the upgrade is purchased every month.) is higher for this option. Second, for options offered at the same total prices, many subject prefer the option where the feature is offered as an upgrade, even if this option has less other features (for example, MobiTV or MMS). The population average of the belief parameter (θ ) is presented in table 6. According to these estimations, the consumers are more likely to use the Mobile E-mail PLUS feature than the Happy Roamer feature.

28

Attributes: Estimated Uncertainty parameter θ Variance of uncertainty parameter θ

Happy Roamer 0.33 0.07

Mobile E-mail PLUS 0.47 0.11

Table 6. Estimated Uncertainty Parameter θ These estimations reflect that the consumer expect to use the Mobile E-mail feature more frequently than the Happy Roamer feature. This agrees with our interviews with a few subjects. This leads to the difference between ex-ante valuation and ex-post valution, which we present in table 7. Ex-ante Valuation Mobile E-mail PLUS 1.67 Happy Roamer 1.49

Ex-post Valuation 5.13 6.65

Table 7. Difference in Ex-ante and Ex-post valuations We present the population averages of the ex-ante valuations and ex-post valuations (consumption value). The consumption values are higher than the ex-ante valuations. The difference is significant (p < 0.001) for both Mobile E-mail PLUS and Happy Roamer. Interestingly, due to a higher likelihood to be used, the Mobile E-mail PLUS feature is more important ex-ante, but consumers value Happy Roamer more when they do need it. The results presented in Table 6 and Table 7, taken togather, point to the validity of our first model assumption: • There exist consumer uncertainty about feature need. The ex-ante willingness-to-pay is significantly smaller than the ex-post willingness to pay when the feature is offered as an upgrade-on-demand. Our second assumption concerns the heterogneity of feature need uncertainty. The variances of uncertainty parameter θ are presented in Table 6. The consumers have more heterogeneous beliefs on the usage of the Mobile E-mail PLUS feature. In fact, their belief about the 29

Figure 3: Population Distributions of the θ Parameter

Mobile E-mail PLUS pattern exhibits a multi-modal pattern, while the belief about Happy Roamer is unimodally distributed wherein most consumers think it’s unlikely to be useful. The histograms which represent the empirical distributions of θ are plotted in Figure 3: The above results support our second utility function assumption: • The consumers have heterogeneous uncertainty about their feature need. The above results, although preliminary, provide some support for our chosen utility function. They support our central assumption of feature need uncertainty and the heterogeneity in such uncertainty. Clearly, the empirical results also show that the distribution of belief parameter θ is not uniform, which is a simplifying assumption made for technical reasons. The empirical context (mobile service plan) was chosen because optional upgrade was considered nature and easier to comprehend in this category. In future works, we plan to examine feature need uncertainty more systematically across more product categories.

6

Conclusions and Limitations

In this paper, we developed a theory of ’feature overload’ to explain the widely observed practice of including features in the products that consumers never use. We argue that even it is common knowledge that none of the consumers ex-post know all the features, offering products with all the features still merge as an equilibrium outcome. The option value of additional features appeals to the uncertain consumers, and this leads to both higher valuation and better market segmentation. In competition, offering feature rich products can happen both as a prisoner’s dilemma outcome and as an optimal differentiation outcome. Heterogeneous feature need uncertain stands as a central assumption in our theory and 30

drives most of our results. We develop a methodology that explicitly measures consumer uncertainty in feature need. This enables us to provide some preliminary evidence on our modeling assumptions in a field study. Our study has a number of limitations. First, our model assumes a simple utility function. While being consistent with the literature on preference uncertainty, this utility function doesn’t capture several important factors: the potential richness of consumer belief, such as the correlation between feature need; heterogeneity in complexity cost which depends on consumer expertise; and heterogeneous ex-post valuations when a feature is needed. These factors are left out to simplify exposition, and incorporating them represents fruitful directions for future research. Second, we did not compare our theory with alternative theories on feature overload. While our theory addresses the demand side, the feature overload phenomenon can be attributed to the cost structure of the firm, the biases in firm level marketing research or the behavior of the new product design engineers. Comparing these various theories will lead to a more comprehensive understanding of the feature overload phenonmen. The explosion of product features is a recent phenomenon especially proununced in the high technology markets. This paper serves as a first study on this topic in an equilibrium framework. A number of interesting issues remain unexplored. For example, the penetration of the Internet leads to wide availability of both product and usage information. Consumers discuss their usage experience in the on-line social sphere such as blogs and social networks, and professional consumer education agencies such as ConsumerRerport.com is increasingly accessible. How consumers feature need uncertainty, partly derived from the lack of consumption experience, interact with this abundance of information? On the supply side, the advent of design modularity and mass customization has enabled firms to offer much greater flexibility in terms of product feature provision, such as offering features as on-demand upgrades. How will such practices interact with the firm’s product design approaches? We left these questions for future research.

31

References Bakos, Yannis, Erik Brynjolfsson. 1999. Bundling information goods: Pricing, profits, and efficiency. MNSC 45(12) 1613–1630. Biyalogorsky, E., E. Gerstner, D. Weiss, J. Xie. 2005. The Economics of Service Upgrades. Journal of Service Research 7(3) 234. Brander, J.A., J. Eaton. 1984. Product Line Rivalry. American Economic Review 74(3) 323–334. Cabral, L.M.B., M. Villas-Boas. 2005. Bertrand Supertraps. Management Science 51(4) 599–613. Chernev, A. 2007. Jack of all trades or master of one? Product differentiation and compensatory reasoning in consumer choice. Journal of Consumer Research 33(4) 430–444. Della Vigna, S., U. Malmendier. 2006. Paying Not to Go to the Gym. American Economic Review 96(3) 694–719. Gilbert, R.J., C. Matutes. 1993. Product Line Rivalry with Brand Differentiation. JOURNAL OF INDUSTRIAL ECONOMICS 41 223–223. Guo, L. 2006. Consumption flexibility, product configuration, and market competition. Marketing Science 25(2) 116–130. doi:10.1287/mksc.1050.0169. Hanemann, W.M. 1984. Discrete/continuous models of consumer demand. Econometrica: Journal of the Econometric Society 541–561. Hauser, J.R., B. Wernerfelt. 1990. An evaluation Cost Model of Consideration Sets. Journal of Consumer Research 16(4) 393. Iyengar, R., K. Jedidi, R. Kohli. 2008. A Conjoint Approach to Multipart Pricing. Journal of Marketing Research 45(2) 195–210. Klemperer, P. 1992. Equilibrium product lines: Competing head-to-head may be less competitive. The American Economic Review 740–755. Lambrecht, A., K. Seim, B. Skiera. 2007. Does Uncertainty Matter? Consumer Behaviour Under Three-Part Tariffs. Marketing Science 26(5) 698–710. Matutes, Carmen, Pierre Regibeau. 1992. Compatibility and bundling of complementary goods in a duopoly. The Journal of Industrial Economics 40(1) 37–54. McAfee, R. Preston, John McMillan, Michael D. Whinston. 1989. Multiproduct monopoly, commodity bundling, and correlation of values. The Quarterly Journal of Economics 104(2) 371–383.

32

Meyer, Robert J., Shenghui Zhao, Jin K. Han. 2008. Biases in Valuation vs. Usage of Innovative Product Features. MARKETING SCIENCE 27(6) 1083–1096. doi:10.1287/mksc.1080.0367. Mukherjee, A., W.D. Hoyer. 2001. The Effect of Novel Attributes on Product Evaluation. Journal of Consumer Research 28(3) 462–472. Nalebuff, BJ. 2000. Competing Against Bundles. Tech. rep., Working Paper Series, Yale School of Management. Narayanan, S., P.K. Chintagunta, E.J. Miravete. 2007. The role of self selection, usage uncertainty and learning in the demand for local telephone service. Quantitative Marketing and Economics 5(1) 1–34. Shin, J., D. Ariely. 2004. Keeping doors open: The effect of unavailability on incentives to keep options viable. Management Science 575–586. Simonson, I., Z. Carmon, S. O’Curry. 1994. Experimental Evidence on the Negative Effect of Product Features and Sales Promotions on Brand Choice. MARKETING SCIENCE 13 23–23. Stremersch, S., G.J. Tellis. 2002. Strategic Bundling of Products and Prices: A New Synthesis for Marketing. Journal of Marketing 66(1) 55–72. Thompson, D.V., R.W. Hamilton, R.T. Rust. 2005. Feature fatigue: when product capabilities become too much of a good thing. Journal of Marketing Research 42(4) 431–442. Verboven, F. 1999. Product Line Rivalry and Market Segmentation-with an Application to Automobile Optional Engine Pricing. Journal of Industrial Economics 47(4) 399–425. Villas-Boas, J.M. 2004. Consumer learning, brand loyalty, and competition. Marketing Science 23(1) 134–145. Xie, J., S.M. Shugan. 2001. Electronic Tickets, Smart Cards, and Online Prepayments: When and How to Advance Sell. Marketing Science 20(3) 219. Xie, Jinhong, Eitan Gerstner. 2007. Service Escape: Profiting from Customer Cancellations. Marketing Science 26(1) 18–30.

33