Value ecosystem models for social media services

Technological Forecasting & Social Change 107 (2016) 13–27

Contents lists available at ScienceDirect

Technological Forecasting & Social Change

Value ecosystem models for social media services Dohoon Kim School of Business, Kyung Hee University, 26 Kyungheedae-ro, Dongdaemoon-gu, Seoul, 130-701, Republic of Korea

a r t i c l e

i n f o

Article history: Received 6 August 2012 Received in revised form 25 January 2016 Accepted 16 March 2016 Available online xxxx Keywords: Social media service Service platform Value ecosystem Ecosystem architecture SDP (Service Delivery Process) Value-creation capability Stage game. Equilibrium analysis Numerical simulation.

a b s t r a c t This study presents the architecture as well as the SDPs (Service Delivery Processes) of two ecosystems—CVE (Collaborative Value Ecosystem) and LVE (Leadership Value Ecosystem)—for social media services, and analyzes and compares these two ecosystems. CVE and LVE respectively represent the traditional social media service ecosystem and a new (future) ecosystem governed by service platforms. In our stylized model, the architecture of the social media platform ecosystem (to be briefly referred as the ‘value ecosystem’) is composed of user group, SP (Service Platform) and CP (Contents Provider). Specifically, users are uniformly populated over [0,Δ], where Δ represents the span of users' preference for the social media services. We also emphasize the differences in the SDP and resulting service provisioning of the value ecosystems. Accordingly, the roles of players are different in two ecosystems. Multi-stage game models are employed for analyses and comparisons of players' optimal behaviors and the value-creation capability in each value ecosystem. In particular, the latter is measured by the amount of information created in the respective ecosystem. Our first analysis indicates that the user group is segmentized differently across the value ecosystems. Subsequently, following analysis presents equilibrium strategies of SP and CP. According to our analyses and experiments, it is more likely that the overall level of the social media services in LVE is higher than that in CVE. However, the total amount of information created throughout CVE is larger than that in LVE. These findings imply that the leadership of SP may boost or constrain the value ecosystem. We also demonstrate that the user diversity measured by Δ is another key parameter for value creation in these ecosystems. © 2016 Elsevier Inc. All rights reserved.

1. Introduction The worldwide popularity of SNS (Social Network Service) clearly indicates that the social media services have become a major driver for the evolution of the Internet industries. YouTube, GoogleMaps, Wikipedia, Facebook, Twitter, LinkedIn, and Pinterest are just a few examples of social media services. They share a common property that these services are provided on the basis of the platform-based technologies. Furthermore, they are organizing their own business ecosystems, where a wide variety of participants join and make transactions with each other by the way of a loosely coupled connection with a platform.1 The social media services are rapidly changing the way of using ICT (Information and Communication Technologies), not only in usual businesses but also in our everyday life. Many organizations are able to take advantage of these new technologies to design and support new ways of communication, collaboration, coordination and learning. Employing these services, organizations expect to make existing managerial administrations more effective and efficient, and even open up completely new ways of conducting their businesses. At the same time, these new

E-mail address: [email protected]. More clear definitions about the notions such as the platform and the ecosystem will be given in Sections 2.1 and 2.2, respectively. 1

http://dx.doi.org/10.1016/j.techfore.2016.03.010 0040-1625/© 2016 Elsevier Inc. All rights reserved.

emerging practices and technologies pose a challenge to the existing theories and approaches employed in research on the ICT industries. A new research framework should be able to capture and reflect these changes in the competitive landscape of the ICT industries. For example, the competition to dominate the contents distribution using platforms is becoming increasingly fierce. As the number SNS users sharply increases and third party providers incorporate various services into SNS with related APIs (Application Program Interfaces), the attempts to utilize SNS as a platform are picking up steam. Facebook and Twitter have already become a great platform; as of the 3rd quarter of 2014, the number of monthly active users of Facebook reached more than 1.2 billion and Twitter more than 280 million (The Statistics Portal, 2014). Driven by the pace of growth, SNS providers are about to run a full-scale business as a service platform. Other best examples are Pinterest and Tencent's WeChat (Wu, 2014) which generate diverse contents using external resources and naturally builds its own ecosystem at the same time. The primary concern of this study is to develop systematic framework to capture and compare different types of service production processes for the social media services. Due to the complicated nature in the current practice,2 however, effective analysis of this phenomenon 2 For an example of complications regarding the social media services and their platforms, see footnote 8 and Section 2.2.

14

D. Kim / Technological Forecasting & Social Change 107 (2016) 13–27

requires a conceptual abstraction, which will produce a simplified model representing some essential features of the practice. In particular, since we focus on the nature of social media services provisioned through the platforms like Facebook and Twitter (sometimes to be called ‘service platform’ in our study), we designate a notion of the ‘ecosystem (Adner, 2006; Basole and Karla, 2011; Ceccagnoli et al., 2012; El Sawy et al., 2010; Gawer and Cusumano, 2014; Iansiti and Levien, 2004; Kim et al., 2010; etc.)’ as our research subject. In this context, this study provides stylized models of two ecosystems for social media services. We will call these ecosystems CVE (Collaborative Value Ecosystem) and LVE (Leadership Value Ecosystem).3 CVE and LVE respectively represent the traditional social media service architecture and a new possible platform of these services, each of which will be dealt with under the notion of ecosystem. In our stylized models, the basic architecture of the social media service ecosystems (to be briefly called the ‘value ecosystems’ later) is composed of three elements: that is, user group, SP (Social Media Service Provider) and CP (Contents Provider). We also emphasize the differences in SDP (Service Delivery Process) and resulting configuration across the value ecosystems.4 Accordingly, the roles of players and the amount of information created will vary between two ecosystems. We employ the amount of information generated throughout an ecosystem as the measure of the value-creation capability of the corresponding ecosystem. Throughout this paper, we are trying to answer the following questions: ■ Is the level of social media services in the future ecosystem such as LVE higher than that in the current ecosystem represented by CVE? ■ Is the value-creation capability (i.e., the amount of information created) of the future ecosystem such as LVE greater than that in CVE?

To find answers to these questions, we need to address the interactions among the main players of the corresponding ecosystem, each of which has its own decision variable(s) and moves in order to optimize its payoff. As a result, modeling and analyzing players' behaviors and ecosystem's performance incorporate the game theory for effectively dealing with the strategic interactions. We will also identify the key parameters and/or components that determine the service level and the amount of information created. For example, it is a wide-spread belief that the wider the user spectrum is, the more likely the ecosystem succeeds and grows. We will test this hypothesis through a series of experiments based on our models. This paper is organized as follows. We first briefly describe the evolution of social media services in the next section. Some key elements and ingredients to construct a value ecosystem are also introduced in this section. Section 3 presents the demand model as well as the supply models, and describes two architectures of the value ecosystems, focusing on the structural differences in their SDP configurations. At the end of this section, we also introduce stage game models to analyze and evaluate the two value ecosystems. In Section 4, main analytical results of game models are provided. In the following section, we present some examples and experiments to supplement our analysis, and discuss the implications of our findings. The last section concludes this article with suggesting future research directions. 2. Social media services: evolution, scenarios and ecosystems This section describes the evolution path of social media services and compares the current status with a possible scenario reflecting a new 3 The continuously evolving nature of the ICT industries and services makes it harder to develop a conceptual and stylized model that perfectly captures the reality. Thus, we admit that our stylized approach based on an abstraction of reality might be likely to miss some aspects of practical cases. 4 The concepts of the architecture and the SDP will be explained soon (refer to footnotes 5 and 6 in Section 2).

trend. In the course of our discussion, one can figure out the core elements and the basic architecture of the ecosystems, which will be employed as building blocks of the stylized models referred as the ‘value ecosystems’ in Section 3. Our approach, in particular, focuses on the SDP (Service Delivery Process), which determines configuration and service operations of the ecosystems.5

2.1. Brief overview of social media services YouTube, GooglePlus, Twitter, Facebook, LinkedIn, Pinterest, Sina Weibo, and Tencent QQ are just a few examples of so-called social media services based on web2.0 technologies (Anderson and Wolff, 2010; O'Reilly, 2005). Social media services are rapidly changing the way we use ICT, not only in our work but also in our everyday life (Anderson and Wolff, 2010; Kim and Ko, 2012; van Noort et al., 2012). Many organizations are able to take advantage of these new technologies to design and support new ways of communication, collaboration, coordination and learning. Using these new social media services, organizations expect to make existing process more effective and efficient, and even open up completely new ways of conducting their businesses (Berthon et al., 2012; Kumar and Mirchandani, 2012; Michaelidou et al., 2011; Soares et al., 2012). In particular, the platform-based service provision has formed a mainstream in the social media services. Here, the platform means a ‘service platform (Cusumano, 2011; Evans et al., 2005; Gawer and Cusumano, 2014; Yoo, 2011),’ which provides infrastructure and technological environment that accommodate multiple services in a constantly reusable fashion.6 More specifically, since this study deals with the social media services, the platform (provider) represents a ‘social media service platform’ or SP as its abbreviation. Such a trend is bringing change to the competitive landscape of the ICT service industry. As the smart device environment starts to take shape, the competition to dominate the contents distribution channels by utilizing the service platforms is becoming increasingly fierce. Service providers recognized the need to turn their asset into a platform. For example, open API was the first step they tried. As web2.0 gained the spotlight and successful cases of mash-up started to appear, open API drew much attention from developers (Kwak, 2001). The interest in the open API remains strong but in fact it has not been as activated as expected except for the map API like GoogleMap. While developers point out that there is no useful API, service providers complain it's getting harder to find proper developers. The resulting vicious cycle significantly reduced the attractiveness of open API. On the contrary, as the number of SNS users sharply increases and third party providers incorporate various services into SNS with APIs invented for SNS, the attempts to utilize SNS as a platform pick up steam. Facebook and Twitter have already become a great platform. Furthermore, new social media services like social network games, social commerce, mobile advertising, mobile payment, etc., accelerate this power shift. Driven by the pace of growth, SNS providers are about to run a fullscale business as a service platform. You can find the best cases in Pinterest, LinkedIn and Instagram. The growth of these platforms deserve attention not just because of the size of its installed-base 5 The architecture means a structural configuration of a complex body. Here, the architecture means a set of structural relations that the major elements of an ecosystem form. The SDP, on the other hand, means the supply procedures which describe service flows going through entities in a system. Here, the SDP represents the way that the social media services are provisioned and the associated service flows go through SP and CPs in an ecosystem. Therefore, the architecture presents a snapshot including both supply and demand in an ecosystem; meanwhile, the SDP focuses on the supply side of the ecosystem. 6 For various points of view on the platform, refer to Evans et al. (2006); Gawer (2011) and Gawer and Cusumano (2002); Parker and van Alstyne (2014). Since this study focuses on the service platform, other types of platform such as smart phone (a platform for various apps) and network (a platform for various communication services), will be excluded in the following modeling and discussions.

D. Kim / Technological Forecasting & Social Change 107 (2016) 13–27

(i.e., the number of active users)—more than 700 million (combined size) as of December 2014, The Statistics Portal (2014)—but also because of the value propositions that they propagate through their platforms. With growth of traffic volume, Pinterest naturally increases the inbound traffic into its partner sites that have original images. Rosenbaum named this kind of social platform ‘social curation (Rosenbaum, 2011)’ as a solution to the problem of overflow of information. In fact, the breakdown of the U.S. referral traffic as of September 2014 shows that the amount of Pinterest traffic (5.5%) far exceeded the combined total of GooglePlus, YouTube, LinkedIn, etc. (less than 1%). Pinterest creates derivative services using external resources but naturally builds its own ecosystem while generating traffic to its partners' website (Kim et al., 2012).7 It takes a wide spectrum of user types for a platform to grow and sustain its scale of growth. If a platform is limited to certain interest groups, it will eventually lose its attractiveness as a platform. Pinterest serves not only general users but also helps companies like Time Magazine to easily promote their products and services through the accounts at Pinterest or to generate revenue through direct traffic inflows to their websites. However, the traffic growth of Pinterest appears to stagger (Media Culpa, 2012) and its business model fell under suspicion. Even Facebook experienced the similar situation right after its IPO. These signs of stagnation of the service platforms imply that simple connection and aggregation of partners' services and contents may not be sufficient for constructing and maintaining a sustainable business (Cusumano, 2011; Karkkainen et al., 2010; Sugiyama et al., 2012). Therefore, despite an optimistic view on the social curation as a next generation model of social media services (Kim et al., 2012; Rosenbaum, 2011), we need to rigorously analyze the potential of these platform businesses. In particular, we need to examine whether they provide a solid basis for value creation through the platforms or not (Ceccagnoli et al., 2012; Cusumano, 2011; Sedghi, 2014). 2.2. Social media ecosystems There have been many studies that introduce and describe the trend around social media services and explain the technical features and key components of the services. Few literatures, however, deal with the social media services or the service platform from the perspective of ecosystem that we briefly mentioned in Introduction. First, the ‘ecosystem’ indicates a complex body composed of interrelated entities and players. Furthermore, a successful platform business is highly likely to evolve into an ecosystem. In fact, the notion of ecosystem has been widely employed in the business areas after the astonishing successes of global platforms like Apple, Google, Amazon, Facebook, etc. However, the traditional notions in economics and business fall short in capturing the essential features of those successes.8 In the course of building the conceptual frameworks for this phenomenon, the analogical notion of ecosystem emerged and began to be extensively adopted in field (Basole and Karla, 2011; Hanna et al., 2011; Tee and Gawer, 2009) as well as in academia (Adner, 2006; Ceccagnoli et al., 2012; El Sawy et al., 2010; Gawer and Cusumano, 2014; Iansiti and Levien, 2004; Jansen and Cusumano, 2013; Tiwana et al., 2010). In particular, we address a platform ecosystem, where a platform such as SP has the initiative to organize the ecosystem. In this regard, our

7 For example, Pinterest exposes ‘Pin it’ and ‘Follow’ button to external sites to let the contents and users naturally flow into its service platform, serving as a sort of SDK (Software Development Kit). 8 For example, these business models entail a network of diverse participants (or ‘complementors’ in terms of Cusumano (2011)), which is hardly captured in the traditional notions such as vertical integration, strategic alliances, and even value chain. The notion of ecosystem fits into this circumstance and reflects the complexity that arises from the interactions among the participants loosely coupled with each other. Indeed, one can easily find lots of news articles mentioning the notion of ecosystem. Some selective titles read ‘ecosystem war,’ ‘ecosystem competition,’ ‘platform ecosystem,’ etc.

15

ecosystem modeling is differentiated from those in the prior studies.9 In sum, the ecosystem here means the ‘social media platform ecosystem,’ to be called the ‘value ecosystem’ for ease of reference, hereafter. This study not only introduces the notion of the value ecosystem but also presents a unique approach to modeling the value ecosystems in order to systemically analyze it. And we also aim at finding the properties and characteristics of the ecosystem that could affect its evolutionary path. For example, analyzing the value-creation capability of the value ecosystems requires a stylized model to systemically represent the corresponding ecosystem. In the context of the social media services in Section 2.1, the basic architecture of a value ecosystem is composed of three key components: user group, SP (Service Platform) and CPs (Contents Providers). Our streamlined architecture cannot capture some aspects of the current practice of SP; for example, the value ecosystems do not explicitly consider the P2P (Peer to Peer) social media services, where the users also play the role of CPs. The key difference between distinct types of value ecosystems, however, does not only come from the components that constitute the value ecosystems, but also from the way of configuring the service flows throughout the ecosystem. That is, we emphasize the differences in the SDP across the value ecosystems. Accordingly, the roles of players are different between the two ecosystems. In sum, the construction of the value ecosystem and its typology are carried out on the basis of the structural and procedural characteristics in the production of social media services: the structural properties represented by the architecture of the ecosystem and the procedural characteristics by the corresponding SDP. As a result, we propose two conceptual models of the value ecosystem: CVE (Collaborative Value Ecosystem) and LVE (Leadership Value Ecosystem).10 The former (CVE) represents the traditional social media platform ecosystem and the latter (LVE) a new possible system of delivering the social media services on the basis of a service platform. Since CVE and LVE are conceptual, stylized models, however, it's not really easy to pick perfect examples in practice for them. Furthermore, the dynamic nature of the ICT industries is continuously shaking the landscape and makes it harder to develop a stable stylized model. For example, the most representative cases such as Facebook and (recently) Alibaba show the characteristics of SP not only in CVE but also in LVE. In fact, our implicit hypothesis for this phenomenon is that the value ecosystems is on the way of evolving (or at least trying to advance themselves) from CVE to LVE. Upon considering the original business model of Facebook, Twitter, and Google+, however, the architecture and the SDP of CVE (refer to Fig. 2 in Section 3.2.1) are closer to the partnership network formed around one of those global platforms (with its users and partners). Note that in CVE, SP and CP have similar market power and cooperate with each other. For example, Facebook and Twitter in early 2010s as well as in the most parts of their current service provisions collaborate with other services or contents such as online newspapers and bloggers. Conversely, online newspapers allow readers to post articles on their websites to their Facebook or Twitter. Though CP and SP need each other, they take charge of different aspects of the entire service provisioning. That is, CP presents basic services (e.g., contents and applications) and SP provides social media platforms on top of the basic service module (also refer to our delineation of services in Section 3.1). On the other hand, in LVE as the most promising value ecosystem model in the near future (refer to Fig. 3 in Section 3.2.1), SP takes the leadership to govern the ecosystem; CP follows the leader and lets its services be integrated into the leader's platform. Here, SP takes the initiative, and together with CP, offers the integrated services for its users. Thus, the architecture and the SDP of LVE are more than the recent features of the ecosystem of Facebook and Twitter. Indeed, it may not be an 9 The research subject most similar to ours can be found in Hanna et al. (2011) though the analytical approach employed there is totally different from ours. 10 These notions are rigorously developed in Section 3.2.1.

16

D. Kim / Technological Forecasting & Social Change 107 (2016) 13–27

exaggeration to take Facebook and Tencent's QQ into account as the SP in LVE. Users will first access their SPs to enjoy basic services like emails and messages, as well as other social media services like mobile games and shopping through the same channel that the respective SP provides. To our best knowledge, service providers which are currently closest to the ideal SP in LVE are Daum-Kakao and Naver Line, global service platforms headquartered in Korea. One could find at least some clue of LVE in their current moves. For example, they not only provide SNS together with typical SNS-linked transaction services but also serve as a gateway or portal for a wide range of Internet services including games, shopping, e-payments, online banking, and even taxi service (similar to Uber). In the next section, we will show that the SDP in CVE is quite different from that of LVE. For example, CVE entails ‘prosuming (= producing + consuming)’ efforts for service customization. On the other hand, the service platform in LVE plays a role as a portal for the entire services thanks to, for example, virtualization techniques. In addition to such a technological breakthrough for the purpose of service platform, a pricing scheme (either flat rate or usage-based charging) becomes another feature that distinguishes SP as the service platform in LVE from that in CVE. Our value ecosystem models to be proposed in detail will capture these differences. We will also employ multi-stage game models for analysis and comparison of players' optimal behaviors in the competing ecosystems. 3. Models Our study focuses on two different scenarios of the value ecosystems: collaborative social media service and social media services provisioned and integrated by a service platform. Each scenario emphasizes distinctive aspects in the SDP of social media service: for example, prosuming for service customization vs. pricing for service platform. The collaborative model (CVE) assumes users' own efforts or prosuming in provisioning the services with separate service tracks, one from CP and the other from SP. On the other hand, the service platform in LVE combines both services and provides the integrated services through a single channel, thereby taking an initiative to govern the entire SDP. We also show that the differences in the SDPs together with diverse needs of users determine a unique configuration of each value ecosystem. Introduced first is a demand model for the social media services, which specifies how users evaluate the value from the services and determine their service consumption. Next, two different value ecosystem models describe the way of provisioning the entire services. Specifically, this section emphasizes the SDPs and presents the way that both SP and CPs determine the service level in the respective value ecosystem. Based on the structural differences in participants' decisions across the value ecosystems, we can analyze and compare the outcomes from the two models. Table 1 summarizes the terminology and symbols employed in our models. Presented also are brief explanations and definitions for the terms. 3.1. Demand model for social media services Users' demand for the social media services comes from access capability to information sources and convenience in customizing information. This convenience and easiness can be defined as avoiding the opportunity costs or technical efforts typically incurred when searching for appropriate information and tuning the basic services from CP without the help of social media tools. Users' needs for the service level may well be different from user by user, thereby characterizing user type. We assume that user type (i.e., user's needs for the overall services) is uniformly distributed over a finite range [L, U]. We set L = 0 and U = Δ, which simplifies the variance of user population (into just Δ) and enhances readability without harming the quality of model outcomes.

Table 1 Terminology and symbols. Terms

Symbols Descriptions

Users' diversity

Δ

User k's service consumption

xk

Users' efforts (in CVE) e σC# σS# Service levels σ#

Installed-base

G# QC1

Amount of information created

QS1 Q2

Marginal benefit of information created Unit cost for service provision SP's service fee (in LVE)

m# h# p

The span of users' preference for the social media services. User type k is spread over [0, Δ]. The service level that the user k chooses. The subscript ‘k’ is sometimes suppressed. Users' time and efforts (i.e., non-pecuniary costs) when customizing one unit of the social media services (if necessary). The service level that CPs decide to provide. σC1 for CPs in CVE and σC2 for CPs in LVE. The service level that SP decides to provide. σS1 for SP in CVE and σS2 for SP in LVE. The overall service level in a value ecosystem. σ1 = σC1 + σS1 for CVE and σ2 = σC2 + σS2 for LVE. In particular, the latter (σ2) is called integrated service level. The number (size) of the active users who participate in the value ecosystem. G1 for CVE and G2 for LVE. The amount of information generated by users and flowed back to CPs in CVE. The amount of information generated by users and flowed back to SP in CVE. The amount of information generated by users and flowed back to both SP and CPs in LVE (i.e., SP and CPs share Q2) The marginal benefit of information created for the respective provider. mC for CP and mS for SP. The unit cost of service provisioning. hC for CP and hS for SP. The unit price charged for the integrated services provided by SP in LVE.

Thus, if k denotes the location index of a particular user then k’s are uniformly distributed over [0, Δ]. Now, we employ the following utility function of a representative user (indexed by k). Note that user k's utility is maximized at k when there is no constraint in his/her consumption. The inverted-U shape of the utility function (refer to Fig. 1) implies the circumstances of deluge of information and superfluous media service channels. Thus, too much information or service offering is rather harmful to users and eventually diminishes the utility levels of some users.11 U k ðxÞ ¼ x ð2k−xÞ ðx: service level consumedÞ

ð1Þ

A user determines his/her optimal service level x⁎ which maximizes his/her utility with the consideration of two reference values: 1) σC, the basic service level from CP like NFL (National Football League) and 2) σS, the social media service level from SP like Facebook. It will be useful to define the overall service level σ≡ σC + σS. That is, in our delineation of services, users' final services are assumed to be composed of two parts: one from CPs and the other from SPs. Accordingly, the term ‘basic’ here is associated with the former and the term ‘social media’ is linked to the latter. However, the entire service provisions are different across the ecosystems (as to be shown in Section 3.2). Our demand model also assumes that users' decisions do not depend directly on providers' service contents but on the levels of services (σC and σS), each of 11 Chellappa and Shivendu (2003) suggested a similar approach to modeling users' needs on personalization services. In fact, this type of utility function is not a new one. For example, Han et al. (1996); Levy (2006), and Philipson and Posner (1999) employ similar utility functions. Many standard textbooks in the fields of microeconomics and decision science (e.g., Rubinfeld and Pindyck, 2012) usually focus on the increasing portion of a concave utility function. However, the inverted-U-shaped utility function [Eq. 1] implies that not only zero consumption of the social media service but also too much consumption of the service (the particular level varies from user by user) produce 0 utility. Since information goods and media services are (even excessively) abundant nowadays, this shape of utility is more appropriate to the context of this study.

D. Kim / Technological Forecasting & Social Change 107 (2016) 13–27

17

Fig. 1. Demand model: representative utility functions.

which stands for typical capability of the corresponding providers' representative services.12 The user group is then segmentized on the basis of his/her location and the gap between σC and his/her type k, at which the corresponding user's utility is maximized. Note that the demand model separates the benefit side (represented by [Eq. (1)]) from the cost factor (see Section 3.2). This technique is devised to deal with the situation where users choose whether they enter or exit the market.13 For example, the user with high service needs—bigger than σC (i.e., k N σC)—may have a willingness to put extra efforts using specialized social media service tools from SP so that he/she can fill the gap between σC and k.14 Fig. 1 describes the behaviors of users given a basic service level σC from CP. Note that users' utility function is concave with its maximum at k, which is uniformly distributed over [0, Δ]. When CP provides the basic service level σC (∈ [0, Δ]), users are classified into two groups: one whose optimal usage k locates below σC and the other whose k lies beyond σC. In the former segment, users are satisfied with the current level of the basic services since σC exceeds their optimal usage of service (i.e., xk⁎ (=k) ≤ σC for all k's in that segment). On the other hand, users in the latter segment have a need to increase their utility by employing additional social media services from SP since the current level of basic service falls short of their optimal usage of overall service (i.e., k ≥ σC for all k's in that segment). Though this fundamental behavioral logic of users responding to the service levels remains unchanged across the ecosystems, the actual responses is likely to be different due to the differences in the SDPs and the relative market power between CP and SP. In the next section, we introduce service supply models, each of which corresponds to the respective value ecosystem. After presenting new stylized models for provisioning the social media services, we compare the new value ecosystem with the current one. 3.2. Supply models and value ecosystems 3.2.1. Service Delivery Processes The way for users to address the gap depends on the characteristics of a specific SDP for the social media services. Here, the SDP describes 12 For example, high level of CPs' services (i.e., high σC) implies that CPs' services (representative contents and applications) provide many functionalities, which some users could utilize without SPs' platform services. In this regard, it could be acceptable to define the overall service level σ by adding two service levels. 13 Thus, [Eq. 1] does not capture all the elements constituting the ‘net utility,’ and the maximization procedure is also dependent on the corresponding value ecosystem (particularly, its SDP). That is, because of this separation, we need to develop an extra step to reflect the cost factor, which will be completed by incorporating the supply side (refer to Section 4.1 and the corresponding optimization models: in particular, [Eqs. (4) and (6)]) 14 An example of these efforts (in particular, in CVE) is to install and manage a plug-in software or app to watch video clips provided by CPs (e.g., NFL).

the procedure and the role of players in the value ecosystem, and represents the supply model associated with the ecosystem. Combining the supply model and the demand model, we complete the architecture of the corresponding value ecosystem. With this perspective of value ecosystem, we can specify how the social media service are produced and delivered to appropriate users and how users utilize these services and create information that will be fed back to providers (SP and CPs). We also assume that there is a structural difference between the current value ecosystem and the one in the near future. The key distinction lies in the role of SP. While SP collaborates with CP in the current value ecosystem, it is supposed to lead the participants in another value ecosystem in the foreseeable future. For ease of reference, we call the former (the current situation) ‘Collaborative Value Ecosystem (‘CVE’ in short)’ and the latter ‘SP Leadership Value Ecosystem (‘LVE’ in short).’ As explained earlier, LVE is a promising scenario in the next generation social media services. We assume that users freely use the contents from CP without any further processing of the relevant programs. On the other hand, users may have to pay a considerable amount of time and effort to configure and maintain the social media services. In particular, we assume that users in CVE pay a fixed amount of non-pecuniary costs (time and efforts) to use the social media services. However, SP in LVE is supposed to charge service fee to users for its integrated services. Thus, the core difference in the cost structure for users is embedded in the value ecosystem and comes from the way of incurring costs in provisioning the social media services. Specifically, the major cost component in CVE is nonpecuniary and determined by technological characteristics of social tools. We assume that this cost is given from the outside of CVE.15 On the other hand, the large portion of the cost incurred in LVE is due to the price that SP charges for its services (i.e., pricing as one of SP's decision variables). Thus, the overall cost will be endogenously determined in LVE. Different cost structures reflect the fundamental operational differences between those two value ecosystems. For example, the SDP in LVE works based on virtualization of ICT resources as in cloud computing, which entails a change in pricing scheme for resource usage (Armbrust et al., 2010; Brynjolfsson et al., 2010; Cusumano, 2010; Rodero-Merino et al., 2010). Virtualization of resources makes it possible for users to enjoy the social media services in an integrated manner and as much as they want at any place and time. Figs. 2 and 3 depict the architecture and the SDP of CVE and LVE, respectively. Note that two value ecosystems share the basic components regarding the social media services. However, the way of provisioning 15 This means that parameter e in CVE is given and fixed. Though e may vary user by user (depending on user's computing skills), we do not incorporate this feature in our model since it will greatly increase model complexity and the main focus of our study does not lie in this feature. We also believe that the variation of users' technical skills is smaller than that of users' preferences as shown in the demand model.

18

D. Kim / Technological Forecasting & Social Change 107 (2016) 13–27

Fig. 2. Architecture and SDP in CVE (Collaborative Value Ecosystem).

final overall services is different from each other. For example, from the visualization of each ecosystem, we can infer that it will be difficult in CVE to customize social media services to the extent as in LVE. Two value ecosystems share the common features. First, note that service provisioning in the value ecosystem is determined by both SP and CP. That is, the overall service level is a composition of two service channels, each of which is supplied by either SP or CP. Thus, it is assumed for analysis that each provider—SP or CP—has its own portion (i.e., its own service level in an abstract sense) in the whole service. On the other hand, information is generated by the active users participating in the ecosystem. Thus, the amount of information created is assumed to be proportional to the respective installed-base.16 In sum, the service level σ is the notation applied to the supply side; meanwhile, the amount of information Q is the notation to the demand side (though Q flows back to the suppliers). In CVE, CP plays a role of a service encounter, whose payoff is determined by the amount of information generated by users, Q(σC, σS) and marginal value of information, m. Since the unique characteristics in the SDP also affect Q(⋅) in the corresponding value ecosystem, we will distinguish it by Q1(⋅) and Q 2(⋅) for CVE and LVE, respectively. We also denote the installed-bases (i.e., the number of users who actively participate in the value ecosystem) by G1(⋅) and G2(⋅) in CVE and LVE, respectively, which will have an effect on the amount of information created in the corresponding value ecosystem (refer to Figs. 2 and 3 for visualizing these concepts).17 SP provides the social media services together with some relevant tools for users and collects information from them. In LVE, SP also collects monthly service fees from users. Configuration and customization of the overall services will be accomplished either by users' direct provisioning in CVE or by SP's integrating services from CPs in LVE. SP does not charge for their services in CVE just like Facebook and Twitter in their earlier days as well as for most users of today; on the other hand, they charge service fees to users in LVE.18 Subsequently, the cost terms (e and p) represent

16 It is still difficult to find an exact instance in point for those terms such as the service level and the amount of information created. Without a well-defined index, it is also impossible to define these terms for empirical test. However, there is no such index to our best knowledge, but this circumstance does not damage the utility of these terms. In fact, we truly sense that the ‘level’ of services is changing as the service platforms continuously develop technologies, introduce new services, and upgrade infrastructure. Moreover, one can devise a surrogate measure for Q by metering the data traffic into SP and CPs. 17 The active users are those who join the ecosystem and use the services and constitute the base of the amount of information that will be fed back to the suppliers (Huang et al., 2009). Thus, the role of the installed-base G is to link the response from the demand side with the benefits Q in the supply side. 18 Again, there is not yet a good example for SP in LVE. However, we can discover a clue for the future form of SP in the charged services that some SNS are trying to launch: e.g., epayment, premium messaging, etc. Moreover, in 2014, there was a rumor that Facebook would charge a monthly fee of $2.99 (Evon, 2014). This rumor has turned out to be untrue, but this potential business model could present a good example of SP in LVE.

users' effort to configure and maintain their social media services in CVE and users' payment of the service fee to SP in LVE, respectively (refer to Figs. 2 and 3 for visualizing and comparing the service consumptions). We may consider other relevant variable costs incurred during implementing and operating basic contents that constitute a part of the standardized service level σC. However, those variable costs associated with the number of subscribers will not be included in our model since it is a convention that the marginal cost of producing a unit of information goods is small enough to be taken as zero (Brynjolfsson et al., 2010; Rifkin, 2014). By the same token, SP is assumed to incur costs associated only to the level of social media services σS. Thus, both CP and SP are assumed to bear the implementation, operation and maintenance costs directly relevant and proportional only to the service levels σC and σS, respectively. 3.2.2. Suppliers' payoffs As for the payoffs of service providers (SP and CPs), we first note that the actual information inflows into CP (denoted by Q C(⋅)) and SP (denoted by Q S(⋅)) constitute the fundamental revenue sources for respective players. Q #(⋅)19 is determined not only by the basic service level (σC) from CP but also the social media service level (σS) by means of SP's application tools (in CVE) or SP's integrated services (in LVE). Another factor that determines Q #(⋅) is the SDP configuration in the respective value ecosystem. For example, since SP in CVE provides the standardized social media services on the top of the basic services from CP, the costs involved in implementing and operating SP services at the level of σS1 may not have a direct effect on σC1. The SDP of each value ecosystem further specifies the payoff structures of SP and CP. First, in CVE, the revenue stream of CP is positively correlated with the amount of information created and designated for CP, Q C1(⋅), which is, in turn, proportional to the installed-base of CP, GC1(⋅). To simplify our models (but not to the extent to undermine the results), the amount of information created from a user and the selected level of service usage of the user are assumed to coincide.20 Similarly, SP's revenue stream is proportional to Q S1(⋅) and GS1(⋅) in CVE. 19 From now on, ‘#’ will be used as a wild character to indicate the cases meant by the context. For example, ‘#’ in Q # represents ‘C’ or ‘S.’ For another example, G#i(⋅) represents the installed-base of provider # (either ‘C’ for CP or ‘S’ for SP) in the ecosystem i (either ‘1’ for CVE or ‘2’ for LVE). We may sometimes suppress some subscripts when it is clear what they mean according to the context. These conventions of using indices will be maintained throughout the paper unless there is a specific comment. 20 This assumption confines the applicability of our models within a specific relation among the parameters (though this restriction is based on the fact that the former (Q #) is ‘non-decreasing’ as the latter (σ#) increases). However, we need to assume this relation for simplifying the mathematical derivation in the following Propositions (in particular, Propositions 3 and 4). Also note that the installed-base G# provides a link that connects the response from the demand side, which is fed back to the benefits Q # in the supplier side. However, G# ≠Q # in general as shown in the functional forms of [Eq. (7) and (8)].

D. Kim / Technological Forecasting & Social Change 107 (2016) 13–27

19

Fig. 3. Architecture and SDP in LVE (Leadership Value Ecosystem).

We also assume that the net benefits or the profits of providers are linear in the respective amounts of information created by users and inversely proportional to the service levels they produce. Let m# and h# denote the marginal benefit of information created and the cost coefficient for service provision, respectively.21 For example, hS denotes the cost coefficient of SP, which is arising when implementing and operating the social media services in either ecosystem. It is typically assumed that as the level of social media services increases, the entire costs for service provisioning increase. Then, the profit functions of SP and CP are represented by the respective equation below: ΠSP1 ¼ mS Q S1 ðσ S1 ; σ C1 Þ−hS σ 2S1

ð2  1Þ

ΠCP1 ¼ mC Q C1 ðσ S1 ; σ C1 Þ−hC σ 2C1

ð2  2Þ

Here, the subscript i indicates the ecosystem type; that is, i = 1 and i = 2 mean CVE and LVE, respectively. In the following section, we will derive Q #(⋅) in terms of G#(⋅) when necessary, which specifies the relation between these two terms and clarifies the concept of installedbase. The SDP in LVE shows that the social media services are delivered through SP on a pay-as-you-go basis. Thus, the main source of revenue for SP is the service fee charged to users. On the other hand, CP's payoff still depends on the information flow generated with the level of combined services, σ2 (= σC2 +σS2). Since SP and CP share the information flow, CP expects a large amount of information created. However, SP's top priority lies in maximizing its profit, not in maximizing the installed-base or the amount of information created. Along this line, we set the payoffs of SP and CP in LVE as in [Eq. (3)]. A specific form for Q2 will be introduced in the following section. ΠSP2 ¼ p Q 2 ðσ S2 ; σ C2 ; pÞ−hS σ 2S2

ð3  1Þ

ΠCP2 ¼ mC Q 2 ðσ S2 ; σ C2 ; pÞ−hC σ 2C2

ð3  2Þ

3.3. Stage games Before introducing game models, we summarize the user model as well as the supply model in terms of decision variables in the value ecosystems. First, it is the standardized service level σC that constitutes the 21 This linearity assumption (particularly, constant return to scale of information created) needs to be verified by empirical study. Without a clear evidence of this relationship, however, our model has no choice but to resort to its simplest form for tractable analysis. We also fix the coefficients (m#'s) across the value ecosystems for ease of comparisons. However, they may vary in practice as the market evolves. That is, they may be endogenously determined in practice. This possibility presents another complication to our stylized approach and remains a challenge for future work.

key decision variable of CP. For SP, the decision variables are different in CVE and LVE. While SP in CVE determines only the social media service level σS1, it decides not only the social media service level σS 2 but also the service charge p in LVE. The cost that users bear in CVE is denoted by e and given exogenously under the context where users configure and maintain their social media services by themselves. Lastly, with given σC#, σS#, and p (or e), users decide whether to join the social media services (in CVE) or buy the social media services and tools from SP to enhance their customized level of social media services (in LVE). Users try to consume the social media services at the level that maximizes their utility under the constraints set by the providers (i.e., the strategy profiles given by SP and CP): {σC1, σS1 | e, Δ, mC, mS, hC, hS} in CVE and {σC 2,σS2,p | Δ,mC,mS,hC,hS} in LVE. Now, xk⁎ denotes the optimal service level of user k. Note that players' decisions are interrelated with each other in our value ecosystem models. Their various interactions are best modeled by employing a multi-stage game. For example, the installed-bases GS1(⋅) and GC 1(⋅) in CVE are to be simultaneously determined once SP and CP choose σS1⁎ and σC 1⁎, respectively; while GS2(⋅) and GC 2(⋅) in LVE are to be sequentially determined when SP first chooses σS 2⁎ and CP responses to σS 2⁎ by choosing σC 2⁎ in turn. The overall game is composed of two or three stages, and the context governing the game process at the upper stages can be postulated in various ways based on the relative market power between SP and CP. We will consider and compare the following two scenarios, each of which is suitable for one of our ecosystems. First, one scenario describes that both CP and SP at the top stage of the game determine their service levels σC and σS at the same time. We call this scenario ‘Cournot competition,’ which will be applied to CVE. In the other scenario entitled ‘Stackelberg competition,’ SP takes the initiative to lead the market and moves first at the top stage; then, CP responds to the leader's decision at the second stage. Considering the SDP in LVE, we will apply this game scenario to LVE, where SP decides the strategy profile {σS2, p} first, and CP as a follower determines σC2 after observing the strategy profile of SP. At the final stage (the second stage in Cournot competition and the third stage in Stackelberg competition), users select their consumption level of (social media) services; for some users, it may be the best to give up using the social media services and accept only the basic service level (i.e., σC). The specific payoffs of users are calculated from the utility function [Eq. (1)]. And the service levels provisioned at the upper stage(s) will affect the consumption patterns in various segments of the user group (refer to Propositions 1 and 2 in the next section). Table 2 summarizes the game scenarios described above. 4. Analysis Typical flow of analysis in a multi-stage game starts from the bottom layer (here, the users' optimal decisions). Then, we conduct the analysis

20

D. Kim / Technological Forecasting & Social Change 107 (2016) 13–27

for the upper stages with the profit functions of CP and SP. The analysis scheme of the upper layer(s) depends on the interactions between CP and SP. In the following subsections, we specify and analyze the SDPs of two stylized ecosystems—CVE and LVE—under the respective game scenario introduced in the previous section and Table 2.

The optimization problem (4) is a mixed non-linear programming and known to be difficult to be exactly solved in general. Our optimization problem, however, is a special case of mixed non-linear programming, and easily solved for the users belonging to [σC 1, σ1]. That is, they will choose x⁎ = σC1 if k ≤ k1 and x⁎ = k if k1 ≤ k ≤ σ1 since the

4.1. Users' optimal decisions

user at k1 is indifferent between two options (x⁎ = σC 1 and x⁎ =k) and the utility is an increasing function of k. One can find the indifferent

Users' optimal decisions on the service consumption will result in a partition of the entire user space [0, Δ]. First, note that user behavior will vary across the value ecosystems due to the differences in the SDPs. For example, we can distinguish two separate portions in the user space around σC1 in CVE (see Fig. 1). One whose optimal level of service usage is below than σC1 (i.e., k ≤ σC 1) determines his/her service usage at k since x⁎ =k. On the other hand, one with the optimal service level larger than σC1 (i.e., k N σC 1) may need more service provisioning than σC1 with the help of SP's social media tools. In LVE, however, users show different behavior since they now have to pay the service fee p for the integrated services. Proposition 1. Given σC1 and σS1 from CP and SP, respectively in CVE, the entire user space [0, Δ] is partitioned as shown in Fig. 4-(a). Under pffiffiffi the assumption of e b σ1 − σC 1, the segments in the partition are well defined as follows:

(a) users who belong to the section {k| k b σC1} consume web service at k, at which their utilities are maximized (i.e., x⁎ =k), pffiffiffi (b) users whose optimal service level (k) is below k1 ≡ σ C1 þ e (i.e., σC1 ≤ k b k1 ) hold their service at σC1 (i.e., x⁎ = σC1), (c) users whose index k lies between k1 and σ1 (i.e., k1 ≤ k b σ1) subscribe to SP and use social media services (with paying some efforts e) in order to achieve their maximum utility (i.e., x⁎ =k), and (d) remaining users (i.e, k ≥ σ1) subscribe to SP and consume the total service at its optimal level (i.e., x⁎ =σ1) despite paying the effort (costs) e.

Proof. First, consider two intervals that partition the entire user space into [0, σC1] and [σC1, Δ] in CVE. It is clear that the level of optimal consumption x⁎ is k for users whose index belong to the lower section (i.e., k b σC1). However, users who belong to the upper section (i.e., k ≥ σC1) behave differently since they have to consider whether to employ the social media services from SP or not. If one chooses to use the social media services then the user should put an effort e to configure these services. Thus, users in [σC1, Δ] face the following optimization problem:

Maximizex ;y

xð2k−xÞ−ey

ð4  1Þ

Subject to

x ≤ σ 1 ð ≡ σ C1 þ σ S1 Þ ξy−x þ σ C1 ≥0 y ∈ f0; 1g

ð4  2Þ

where ξ is a sufficiently large number. The second inequality in Eq. (4-2) represents the condition for the effort e to be incurred (i.e., y= 1) when employing the social media services (i.e., x N σC1).

user k1 , whose utilities from choosing either option remains the same: 2

i.e., σ C1 ð2k1 −σ C1 Þ ¼ k1 −e. Solving the equation with respect to k1 pffiffiffi under the conditions of e ≤ σ1 −σC 1 identifies such k1 as follows: k1 ¼ σ C1 þ

pffiffiffi e:

Note that the condition in the above proposition guarantees thatk1 is less than σ1, and the partition of the user space is thus well defined. Similar arguments hold for users in [σ1, Δ] except x⁎ = σ1 instead of k. Q.E.D. Proposition 2. Given σC2 and σS2 from CP and SP, respectively in LVE, the entire user space is partitioned as shown in Fig. 4-(b). Let's definek 2 ≡ p p 2 and k2 ≡ σ 2 þ 2 . The segments in the partition are well defined as follows:

(a) users who belong to the segment fkjk b k 2 g do not subscribe to any service (thus, not paying the service fee p),  g consume the total (integrated) ser(b) users in fkjk 2 ≤ k b k 2 vices at x ¼ k− 2p, where their utilities are maximized, and (c) users whose indices are larger than k2 consume the total service at its upper bound (i.e., x⁎ = σ2).

Proof. Users face the following concave optimization problem:

Maximizex

xð2k−xÞ−px

ð5  1Þ

Subject to

0 ≤ x ≤ σ 2 ð ≡ σ C2 þ σ S2 Þ

ð5  2Þ

One can solve this problem easily if 0 b x⁎ b σ2 (i.e., when x⁎ is an interior solution), where x ¼ k− 2p. Note that x⁎ satisfies the FONC(First Order Necessary Condition) and the SOSC(Second Order Sufficient Condition) for global optimality, and x⁎ b σ2 for k b k2 . Since the optimal usage level is increasing as k increases, x⁎ = σ2 when the right inequality of the constraints (Eq. (5-2)) is binding at σ2. This situation occurs first when k ¼ k2. Users beyond k2 (i.e., k ≥ k2) have no choice but to consume the integrated service at σ2. On the other hand, for users whose indices are lower than k 2, it is better not to join the integrated service (i.e., x⁎ = 0) since even a tiny amount of consumption results in a negative value of Eq. (5-1). Q.E.D. Fig. 4 describes and compares the patterns of user segments in CVE and LVE. Propositions 1 and 2 show that the optimal responses vary across the value ecosystems due to differences in the SDPs. A key decision issue in CVE is whether to put additional effort e to enjoy the overall service level at his/her maximum (k) or pause at the basic service level from CP (σC1). Unlike in CVE, the users in LVE determine their optimal usage against the unit price p when they decide the consumption level

Table 2 Game scenario plots. Descriptions Scenario title (abbreviation)

Number of stages

Cournot in CVE (CU) Stackelberg in LVE (SB-L)

2 3

Role of CP

Role of SP

Users' responses

Determine σC1 (simultaneously) Determine σC2 after looking at {p, σS2}

Determine σS1 (simultaneously) Determine {p, σS2} first

Respond to {σC1, σS1} Respond to {p, σC2, σS2}

D. Kim / Technological Forecasting & Social Change 107 (2016) 13–27

21

Fig 4. User segments and level of services consumed. (a) User Segments in CVE. (b) User Segments in LVE.

of the integrated service through SP. As a result, there is no jump in the service consumption over the user space in LVE (see Fig. 4-(b)). Accordingly, the partition of the user space in each value ecosystem is distinct as shown in Fig. 4-(b). Before examining the user group segmentations, we note that these partitions are well defined in the sense that the providers' payoffs are conformable to and go well with the model assumptions. Right-hand sides of the equations for k1 , k 2 and k2 show different aspects of the corresponding value ecosystem due to the differences in their SDPs. For example, from the outcomes of Proposition 1, GC1(⋅)and GS1(⋅) become k1 ðσ C1 ; eÞ and Δ−k1 ðσ C1 ; eÞ , respectively, thereby G1(⋅) = GC1(⋅) + GS1(⋅) = Δ; that is, the overall installed-base in CVE covers all the user space. On the other hand, some users whose indices are below k 2 do not join the integrated service, and the overall installed-base in LVE G2(⋅) reduces to Δ−k 2 . This reduction comes from the charging scheme in LVE, which results in differences in the user group segmentation and the overall installed-base. The amount of information generated will be proportional to the total size of service consumed. We assume that those two quantities are identical for ease of analysis (also refer to footnote 18); that is, the size of information created and the total size of the corresponding service consumed are the same. Accordingly, Q C1(⋅) and Q S1(⋅) in CVE can be specified in terms of the service levels set by SP and CP as follows: Q C1 ðσ C1 ; σ S1 Þ ¼

  1 2 1 σ C1 þ k1 −σ C1 σ C1 ¼ k1 σ C1 − σ 2C1 2 2

  1  σ 1 −k1 σ 1 þ k1 þ σ 1 ðΔ−σ 1 Þ 2  2 1  2 ¼ σ 1 Δ− σ 1 þ k1 2

Q S1 ðσ C1 ; σ S1 Þ ¼

ð6  1Þ

Similarly, we specify the information generated from the integrated service in LVE as follows:     1 σ 2 k2 −k 2 þ Δ−k2 σ 2 2  σ C2 þ σ S2 þ p ¼ ðσ C2 þ σ S2 Þ Δ− 2

Q 2 ðσ C2 ; σ S2 ; pÞ ¼

ð7Þ

4.2. Cournot competition in CVE In this section, we analyze the Cournot competition in CVE, where both SP and CP have compatible market power and they are deemed moving simultaneously. This scenario may reflect the situation around 2010 (or early 2010s), when SNS providers just entered the ICT service market and their services were recognized to add new features on the top of the typical Internet services like web browsing and video streaming (e.g., YouTube). Until now, the roles of service providers have been separated and the SNS providers complement the basic services provided by CPs. Given the equations for Q#'s as in [Eq. (6)′], the payoffs of SP and CP take specific forms in terms of their corresponding decision variables as follows:   pffiffiffi  1 ΠSP1 ¼ mS ðσ C1 þ σ S1 Þ Δ− 2σ 2C1 þ σ 2S1 þ 2σ C1 σ S1 þ 2 e σ C1 þ e −hS σ 2S1 2

ð2  1′Þ

ð6  2Þ



pffiffiffi 1 2 σ C1 þ e σ C1 −hC σ 2C1 2

pffiffiffi Finally, we get [Eq. (6)′] below by plugging k1 ¼ σ C1 þ e and σ1 = σC1 + σS1 into [Eq. (6)]:

ΠCP1 ¼ mC

pffiffiffi pffiffiffi  1 1 Q C1 ðσ C1 ; σ S1 Þ ¼ σ C1 þ e σ C1 − σ 2C1 ¼ σ 2C1 þ e σ C1 2 2

Let's suppose that the following conditions hold in order for SP and CP to make optimal decisions. These conditions guarantee that the payoff functions [Eq. (2)′] to have positive maximum points. In particular, the second condition in [Eq. 8] implies that the range of user population Δ is large enough for a standardized service level σC1 to fall short of satisfying all the users. Thus, this condition is not too strict, but rather in line with the real world. The following

Q S1 ðσ C1 ; σ S1 Þ ¼ ðσ C1 þ σ S1 Þ Δ−

ð6  1′Þ

pffiffiffi  1  2 2σ C1 þ σ 2S1 þ 2σ C1 σ S1 þ 2 e σ C1 þ e ð6  2′Þ 2

ð2  2′Þ

22

D. Kim / Technological Forecasting & Social Change 107 (2016) 13–27

Proposition presents a Nash equilibrium {σ S1⁎ , σC1⁎ } under the Cournot competition.

2hc N mC

pffiffiffi mC 2hS and Δ N þ þ1 e 2hC −mC mC

ð8Þ

Proposition 3. In CVE with the conditions in [Eq. 8], SP and CP determine their Nash equilibrium strategies, σS1⁎ and σC1⁎, respectively as follows:

follows:  σ C2 þ σ S2 þ p ΠSP2 ¼ p ðσ C2 þ σ S2 Þ Δ− −hs σ 2S2 2

ð3  1′Þ

 σ C2 þ σ S2 þ p ΠCP2 ¼ mC ðσ C2 þ σ S2 Þ Δ− −hC σ 2C2 2

ð3  2′Þ

Let's define the following terms to simplify the expressions in the proposition below: C , J = J(σS2) = ζ (Δ − σS2)+ σS2, K ≡ J⋅ð1− 12Þ, and L ≡ ζ (2 Δ − J). ζ ≡ mCmþ2h C

σ S1 ¼

mS

pffiffiffi

mC e pffiffiffi mC e 2hC −mC and σ C1 ¼ mS þ 2hS 2hC −mC

Δ−

ð9Þ

Proof. First, note that (σS1⁎, σC1⁎) is the solution of simultaneous equations d σd S1 Π SP1 ¼ 0 and d σd C1 Π CP1 ¼ 0. Thus, both σS1⁎ and σC1⁎ satisfy FONC. The first condition in pffiffi[Eq. (8)] guarantees that SOSC is also satisfied. Finally, since Δ N 2hmCC−meC if the second condition in [Eq. (8)] holds, both σS1⁎ and σC1⁎ are non-negative. Furthermore, the second condition pffiffiffi makes the former no smaller than e, which guarantees the existence of the segment between k1 and σ1 in Fig. 4-(a). Q.E.D. We can derive the following properties from the proposition above. First, from the shape of σC1⁎ and σS1⁎, we easily know that the effects of the parameters representing marginal benefits (m#) are different from those of the parameters for marginal costs (h#): that is, and

∂σ 

# ∂h#

 ∂σ# ∂m#

N0

b 0 as expected. In particular, while both σC1⁎ and σS1⁎

reduce to 0 as its corresponding m# goes to 0, σS1⁎ converges to a finite pffiffi value ðΔ− 2hmCC−meC Þ under the conditions in [Eq. (8)] as mS increases; σC1⁎ explodes (i.e., diverges to infinity) as mC increases and gets close to 2hC. Thus, we can say that σC1⁎ is far more sensitive to its parameter corresponding to the marginal benefit than σS1⁎. Assuming a very large Δ, this difference in sensitivity to the respective marginal benefit implies that CP is at a better position to compete with SP in CVE. For example, a small technological or quality improvement in CP services brings about a big jump in the standardized service level (σC1⁎), which eventually reduces the room for effective enhancement of SP services. One may also find that the parameter Δ (the scope of diversity in users' preference) affects only σS1⁎, but leaves σC1⁎ intact. This asymmetric influence of Δ on the service levels at the equilibrium also implies the advantageous position of CP in CVE. Without a large Δ enough to meet the conditions in [Eq. (8)] by a wide margin, there will not be much room for SP's move in the game.

Moreover, let J⁎, K⁎, and L⁎ denote J(σS2⁎), K(σS2⁎), and L(σS2⁎), respectively. Note that ζ, J, K, and L are all positive if Δ ≥ σS2, which is actually required for SP's decisions to be feasible. Then, Proposition 4 provides a Stackelberg equilibrium {p⁎, σS2⁎, σC 2⁎} under the scenario SB-L. Proposition 4. Suppose first that Δ is large enough for players' decisions  to be feasible: that is, σS 2⁎, σC 2⁎, p⁎ N 0. Specifically, Δ ≥ max fσ S2 ; p g þ p2 satisfies this requirement. Then, in LVE, there is a such that for all ζ ≥ , players' decisions described in [Eq. (10)] provide a Stackelberg equilibrium. That is, while CP determines their equilibrium strategy σC 2⁎ as in [Eq. (10-1)], SP's optimal strategy profile {σS2⁎,p⁎} is the solution of the simultaneous equations in [Eq. (10-2)].

1 σ C2 ¼ ζ Δ−σ S2 − p 2

ð10  1Þ



qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi p 2  p ð 1−ζ Þ Δ− 2   2 2K þ L − 3K þ ðK −L Þ 2

p ¼ and σ S2 ¼  ð1−ζ Þ2 þ 2h ζ p S 3ζ 1− 2 



ð10  2Þ

Proof. The proof in the Stackelberg games starts from the follower's decision. FONC for CP with given σS2⁎ and p⁎ is summarized as follows: i.e.,  σ C2 ¼ ζ ðΔ−σ S2 − p2 Þ. Note that SOSC for CP's optimal decision is also satisfied. By solving the following two equations from FONC for SP's op∂ Π SP2 ¼ 0, one can find an optimal detimal decision, ∂σ∂ Π SP2 ¼ 0 and ∂p S2 cision of SP. Unfortunately, however, we cannot solve these non-linear simultaneous equations to derive an exact solution. Thus, it will be the best to summarize two equations as in [Eq. (10-2)] after a long series of arithmetic calculations and define implicitly the equilibrium as a solution of those two equations. Then, what remains is to show SOSC for SP's optimal decision. We first construct a Hessian matrix H as follows for that purpose:

4.3. Stackelberg competitions in LVE In a Stackelberg competition, one of providers has the leadership and exercises the market power on the other. Under the scenario titled ‘SB-L in LVE’ in Table 2, SP is at the leading position and moves first at the top of the game. Though the current situation may evolve along various paths where either SP (e.g., Facebook) or CP (e.g., YouTube) seizes the leadership first, we will consider only the SB-L scenario in this paper since the current situation and the prospect for technology development seem to support this scenario. As a matter of fact, SB-L is devised to envision LVE in detail, and it gets more promising that SP will have more market power than CP in the near future. Thus, the Stackelberg competition scenario seems practically more appropriate for our research context and background. Given the equations for Q 2(⋅) as in [Eq. (7)], the payoffs of SP and CP take specific forms in terms of their corresponding decision variables as

2 H¼4

−pð1−ζ Þ2 −2hS ð1−ζ Þ2 ðΔ−σ S2 −pÞ

3 ð1−ζ Þ2 ðΔ−σ S2 −pÞ

5: 3p ζ ð J−2ΔÞ−ð2−ζ Þ J− ζ 2

Note that H1 = −p(1−)2 − 2hS b 0 (for any positive p) and | H |ζ = 1 = 2  hS ð2Δ− 3p2 Þ N 0 if the condition on Δ in the proposition holds. Since | H | is a continuous function of ζ and ζ takes only values from a compact set [0, 1], there must exist a ζ such that | H | N 0 for all ζ ∈ [ ζ, 1]. Thus, σS2⁎ and p⁎ from FONC truly maximize ΠSP2 for ζ ∈ [ ζ, 1]. Q.E.D. The optimal behaviors of players in LVE are quite different from those in CVE. According to Proposition 4, not only SP but also CP consider Δ (the user diversity) in their optimal decisions. Thus, SP is now a better position than in CVE, and has an initiative to govern the ecosystem. Since two service tracks (one from CP and the other

D. Kim / Technological Forecasting & Social Change 107 (2016) 13–27

23

from SP) are integrated by SP, two service levels σC 2⁎ and σS2⁎ are interrelated each other; for example, σC 2⁎ decreases as σS 2⁎ increases. Such optimal behaviors in LVE sharply contrast with those in CVE. Increasing p⁎ also affects negatively σC2⁎, which reflects the fact that two service tracks do not complement each other; rather, they move together in LVE. Note that this proposition presents only a sufficient condition. There may be no small possibility that an equilibrium which satisfies FONC but fails SOSC remains a true equilibrium. Furthermore, if p is not subject to SP's decision but set by a third party (e.g., a policy authority) then the outcomes in Proposition 4 are dramatically simplified. The following corollary summarizes this result.

performance of the value ecosystems such as providers' total profits, overall services levels, and value-creation capability (e.g., the total amount of information created). Note that the providers' total profits are also connected to suppliers' surplus in terms of social welfare. Though other measures cannot precisely represent consumers' surplus (the other half of social welfare), they are somehow linked with this surplus since they are derived from the installed-bases: that is, the larger the installed-base is, the more information is generated and the higher chance of increase the total utility gains have.

Corollary 5. Suppose that p is given and Δ N 2p in LVE. Then, players' decisions described in [Eq. (11)] provide a Stackelberg equilibrium.

 p   ð1−ζ Þ2  Δ−

p  p 2 ð11Þ σ C2 ¼ ζ  Δ−σ S2 − and σ S2 ¼ 2   ð1−ζ Þ2 þ 2hS p

We first simulate specific outcomes with all the parameters fixed at certain values except the ratio between mS and mC. In particular, h#'s equal to 1 and Δ equals to 10. The experiment results in Fig. 5-(a), (b), and (c) demonstrate changes of the overall service levels and the amount of information created as the relative benefit m between mS mS and mc ðm ≡ m Þ varies. On the other hand, Fig. 5-(d) depicts another senC

5.1. Performance comparisons: service level and information created

Our models and analyses in the previous section show the distinctive natures of the two value ecosystems. Now, we conduct some experiments and compare CVE and LVE so that we can evaluate the analytical results and derive practical implications for the current situations and the future.22 The outcomes from the experiments together with following discussions focus on some key measures representing the

sitivity analysis (i.e., comparative statics), where the user diversity Δ changes with the other parameters fixed: h#'s and m#'s all fixed at 1. While Fig. 5-(a) and (b) demonstrate the changes in magnitudes, Fig. 5-(c) and (d) display the relative changes between two ecosystems. Since we focus on comparing the performances of two social media ecosystems, we present only relative changes in the following figures and interpretations. Fig. 5-(a) shows that the overall service levels in LVE are always higher than those in CVE, which results in the upper curve in Fig. 5-(c); that is, the ratios are greater than 1. Fig. 5-(b), however, demonstrates that the actual amount of information created in CVE is larger than that in LVE. As a result, the lower curve in Fig. 5-(c) lies below 1. Furthermore, these figures demonstrate that not only the overall service level but also the amount of information created in LVE decrease as the marginal benefit for SP (mS) increases (or, the relative benefit m increases). Accordingly, beyond a sort of threshold, the relative performances of LVE drawn in Fig. 5-(c) diminish as the relative benefit of SP increases. These outcomes, which seem contradictory on the first glimpse, actually come from the underlying optimization behavior of SP. SP in LVE increases not only the service level (σS2⁎) but also the price (p⁎) as partner's (CP's) benefit coefficient (mC) decreases.23 Accordingly, the number of users joining the integrated services in LVE decreases, thereby reducing the amount of information created. On the other hand, even though the rate of increase of SP's service level (σS1⁎) in CVE may be lower than that in LVE, σS1⁎ keeps increasing and attracts more users to its service. Furthermore, the increasing benefit coefficient (mS) of SP in CVE generates sufficient incentives for SP to enhance its service level. In sum, while growing benefits of SP in CVE (i.e., increasing m in Fig. 5-(a)–(c)) raise the overall service level σ1⁎ (Fig. 5-(a)) and eventually lead to a sizable increase of information created (Fig. 5-(b)), decreasing benefits of CP in LVE (i.e., increasing m) may erode the benefits for users (Fig. 5-(b)) and even for leading SP. Note that SP's service in CVE is basically optional to users. And also note that m# is the coefficient that represents the marginal benefit from the amount of information created for a corresponding provider. Thus, this coefficient eventually represents the strength of impacts of the corresponding installed-base. When mS increases thanks to the network externality (Burtch, 2011; Henderson and Bhatti, 2001),24 SP will try to enhance its service level, which in turn results in a sizable increase in information created throughout the ecosystem. It is not just a

22 Since we consider two ecosystems, each of which represents different stage of the evolution of social media services, the following comparisons and interpretations may seem to deal with a dynamic model. However, all the results in this section are based on two static equilibriums (they vary as parameters change, though), and our explanations and interpretations often address possible migration from one state (usually, an equilibrium in CVE) to the other (eq. in LVE).

23 Even though we omit a graph that describes the changes of p⁎ against the relative benefit (m), the graph shows a positive relationship between those two terms. 24 Their studies show that the network externality increases the traffic (a surrogate measure of the amount of information created) into a specific website (e.g., crowd funding in the former and online game in the latter). Thus, the network externality can be thought of as a major factor that affects the marginal benefit parameter (m#).

Proof. The proof follows a similar (but much simpler) procedure to that of Proposition 4. Indeed, the decision variable of SP is now only σS2 since p is assumed to be given due to some reasons (e.g., price regulation set by policy authority). Also note that FONC for CP with given σS2⁎ and p remains the same as in Proposition 4. That is, σ C2 ¼ ðΔ−σ S2 − 2pÞ, which establishes the first part of [Eq. (11)]. FONC for SP's optimal decision reduces to ∂σ∂ Π SP2 ¼ 0, which in turn yields the following equation: S2

  p p 2 pð1−ζ Þ Δ−σ C2 −σ S2 − −2hS σ S2 ¼ pð1−ζ Þ Δ−σ S2 − −2hS σ S2 ¼ 0: 2 2

Solving the above equation in terms of σS2 produces the second part of [Eq. 11]. Furthermore, SOSC also reduces to checking out the condi2

tion ∂σ∂

S2

2

Π SP2 ¼ −pð1−Þ2 −2hS b 0, which is clearly satisfied. This com-

pletes the proof. Q.E.D. One can compare the outcomes from the above Corollary with those from Proposition 4 to investigate possible consequences of price regulation in LVE. Unfortunately, however, it's impossible to rigorously analyze SP's behavior in LVE with its more general form of the profit function since the expressions in [Eq. (10)] are too complicated to be explicitly examined. Though our analysis so far makes it possible to compare the players' behaviors across the value ecosystems, the applicability and implications of these comparisons are limited since we cannot derive an exact form of the equilibrium in LVE. As a remedy for this problem, we have no choice but to proceed with a couple of special cases where some parameters are fixed or restricted within certain ranges. In our experiments in the next section, we will solve some examples by means of a numerical calculation tool such as Matlab, demonstrate the outcomes with various parameters, and conduct some sensitivity analysis. 5. Experiments and discussions

24

D. Kim / Technological Forecasting & Social Change 107 (2016) 13–27

Fig. 5. Comparison of Service Levels (σ#) and Information Created (Q#) across the ecosystems. (a) Relative benefit ðm ≡ ðm ≡

mS mC Þ

varies: information created in magnitude. (c) Relative benefit ðm ≡

mS mC Þ

mS mC Þ

varies: overall service level in magnitude. (b) Relative benefit

varies: ratios. (d) User diversity Δ varies: ratios.

coincidence, for example, that many successful SPs started their business with a free service (e.g., free search, free messaging, etc.), which is the most effective strategy for expanding the installed-base to reach a critical mass in a short period (so-called ‘freeconomics,’ Evans et al., 2006). In the early phase of the value ecosystem, services and technologies that make the marginal benefits of SP enhanced are also beneficial to the entire value ecosystem. Therefore, SP in CVE has an incentive to develop such services and technologies, and all the other players in the ecosystem would welcome SP's behavior. SNS providers are still trying to devise services or technologies that leverage the network externality and eventually improve mS. And only those SPs that first reach the critical mass could attain the self-reinforcing feedback, which provides a great advantage over their competitors. With few exceptions, SPs pursuing the leadership position in LVE already have secured the substantial market share. On the other hand, for SPs which do not possess the compatible market share, it is hardly thinkable to develop themselves into SP in LVE.25 However, in the LVE scenario, SP may be too strong to take responsibility for the co-prosperity of the entire value ecosystem. Before a SP achieves a sizable installed-base, it maintains an equal partnership with CPs (e.g., a provider of a popular game). However, once the SP 25 Compare Facebook and Google+. The size of active users of Google+ is about 340 million as of December 2014 (The Statistics Portal, 2014), which amounts to one quarter of Facebook users. Actually, Google does not seem to put all of its efforts in Google+ (its SP) and is pursuing a platform ecosystem strategy different from SP in LVE.

secures a solid foundation of the installed-base due to some reasons (e.g., technological advancement, sudden boom of a killer service, etc.), it will try to take the initiative to lead the value ecosystem and want to change the relationship with CPs. Furthermore, SP's enhancement of its service level is likely to entail higher pricing and may (probably unintentionally) encroach on CP's benefits (i.e., reducing mC). For example, since the power of the ecosystem has been shifted to SP, advertisement sponsors who used to support CPs now desert them and move to SP, thereby mS Þ. Thus, the network externality sources which eventuincreasing mð ≡ m C ally improve mS more than mC (i.e., increase m) could be a key driver toward LVE. For example, the effects of the virtualization technologies (e.g., mobile cloud computing) on the network externalities of SP and CP are not symmetric. The boom of social games in many North-East Asian countries (Dong et al., 2014) is in line with the trend of mobile cloud services provided by Chinese global SPs (e.g., Tencent's QQ and Sina Weibo) and Korean global SPs (e.g., Daum-Kakao and Naver Line).26 SP draws more benefits from this technology than CP does since SP provides more general access to various Internet services. In this context, SP's efforts to enhance its marginal benefits may not boost LVE (unlike in CVE), but ironically could harm the entire ecosystem in the worst case as the diminishing part in the upper curve in Fig. 5(c) implies. Technologies or services that leverage the network 26 At the same time, the market share of the traditional game players such as Microsoft (Xbox), Sony (PlayStation) and Nintendo started to drop. They were not aggressive enough in incorporating the SNS platforms into their online game platform.

D. Kim / Technological Forecasting & Social Change 107 (2016) 13–27

externality, thereby enhancing mS, may be a double-edged sword for the value ecosystem. It helps growth and co-prosperity of the value ecosystem (particularly, CVE) in the early phase; but, it may also contribute to an unbalanced growth of the ecosystem (particularly, LVE) and restrict its potential value-creation capability. We will come back to this issue in Section 5.3. Fig. 5-(d) demonstrates the gains from the user diversity (Δ). Examining the equilibriums in the ecosystems ([Eq. 9] for CVE and [Eq. 10] for LVE), we realize that Δ will enhance the overall service levels in both d σ

ecosystems: for example, d ΔS1 ¼ ζ N 0. That is, the diversity positively affects both CVE and LVE. However, Fig. 5-(d) implies that increasing diversity Δ will be more effective on the enhancement of the overall service level in LVE than on that in CVE. On the other hand, as the lower curve in Fig. 5-(d) shows, the amount of information created does not seem to be greatly affected by increasing diversity. This contrast comes again from the underlying behavior of SP in LVE. That is, as the diversity increases, the optimal price (p⁎) also increases. We will soon come back to this issue when dealing with Fig. 6-(b) in the following section.

5.2. Providers' profits and user diversity Providers' profits except CP's in CVE27 usually increase with the user diversity Δ (these trends are omitted in the following figures). Accordingly, as expected from our previous analysis, we can conclude that the user diversity is beneficial to players and provides them with great incentive for innovation. Fig. 6 demonstrates more about the effects of changes in the diversity. Fig. 6-(a) and (b) respectively depict the rela

tive players' profit changes and the relative user cost ðpe Þ changes when Δ varies. Here, all the parameters except Δ are fixed at the same values in Fig. 5-(d). According to Fig. 6-(a), the positive effects of the user diversity on providers' profits do not seem symmetric though Δ is a key parameter for value creation irrespective of the ecosystem type. The effects of Δ seem much stronger on the providers in LVE than on those in CVE, and these effects in LVE are stronger on CP's profit than on SP's (CP's relative profit curve outgrows SP's in Fig. 6-(a)). Regarding this outcome, Propositions 1 and 2 present the reason why the user diversity is more beneficial to LVE than to CVE. First note that the active users in CVE span the entire user space (Proposition 1); whereas the overall installed-base in LVE does not cover the entire space. That is, some users in LVE do not join the ecosystem due to a higher service fee exceeding their utility gains in using the integrated social media services. Accordingly, if the diversity Δ in LVE increases due to some reasons (one to be provided below), then it is likely to accommodate more users with higher benefits to both SP and CP as indicated by Fig. 6-(b). On the other hand, the greater diversity in CVE does not necessarily mean a higher Π margin to the providers. As a result, the relative gains ðΠ#2 Þ keep in#1 creasing as the diversity increases. Thus, the drivers and the market forces that develop users' taste and activate potential needs will benefit the providers: particularly, those in LVE. One can find some good examples and lessons in leading SPs' efforts to cultivate the demand for customized services. For example, many SPs launched closed networking services in order to serve specific needs of some user communities (e.g., Facebook ‘group’, Daum-Kakao ‘Kakao story’, Naver Line ‘band,’ etc.). These actions were actually taken as a strategic response to increasing popularity of and needs for social curation (Rosenbaum, 2011) and vertical SNS providers (e.g., Pinterest and LinkedIn), and contributed to substantially extending the user space (i.e., increasing Δ). These service providers present intermediate forms of service platform in the course of moving toward SP in LVE. And this 27 CP's profit in CVE does not change as Δ increases since σC1⁎ is independent of Δ (see [Eq. 9] in Proposition 3).

25

way of service provisioning makes it possible to enhance the user diversity (Enders et al., 2008), thereby strengthening SP's position in LVE. However, Fig. 6-(b) shows that the optimal pricing of SP in LVE keeps increasing as the user diversity increases. For a fair comparison, we set e at 30 in CVE so that p⁎ in LVE can be smaller than e in early values of Δ. This graph confirms the previous reasoning about the immaterial increase of information created in LVE (also see Fig. 5-(d)). SP in LVE is trying to increase its price to make the most of expanded user space despite the possible loss due to the reduction of the installed-base (GS2). As a result, the entire value ecosystem in the LVE scenario may experience a reduced scale of information created. This issue triggers a debate on regulating the service fee p in LVE, which is a discussion topic in the next section. 5.3. Policy implications According to the experiment results above, it is more likely that the amount of information created throughout CVE is larger than that in LVE despite a higher level of the social media services in LVE. That is, a better performance in the overall service level does not necessarily mean that a larger amount of information would be generated. These conflicting results are brought about by the SP's optimization behavior in LVE, where not only the level σS2 but also service charge p are key decision variables, and eventually determine the amount of information created in the value ecosystem (see Fig. 6-(b)). Thus, a strong position of SP in the value ecosystem may constrain the value creation throughout the ecosystem. For a similar case in practice, after Google started charging for its map services, many developers and corporate users are leaving Google and switching over to other competing service platforms such as Apple's map API in iOS6. Faced with such a situation, it may not be a bad idea to enforce a price regulation in order to prevent the market size from shrinking and maintain the vitality of the ecosystem. Following this reasoning and the previous implication from Fig. 6(b), one may raise a question about the amount of information created if p is regulated by a policy authority as in Corollary 5. Fig. 7 shows the experimental outcomes under the scenario where p is exogenously given by a third party or a policy authority or as a result of fierce competition among competing SPs. For fair comparison with the previous examples, we set p at 7 and 15 in Fig. 7-(a) and 7-(b), respectively; all the mS at the same values as those in other parameters except Δ and m ¼ m C Figs. 5-(c) and 6-(b). It is reasonable to infer from the outcomes in Fig. 7 that a regulation on the price for SP's integrated service in LVE will increase the overall inmS Þ formation created. For example, as the relative marginal benefits mð ≡ m C increases, the relative amount of information created with p fixed shows a similar pattern to the lower curve in Fig. 5-(c), which is repeated in a magnified form from Fig. 7-(a) (refer to the lower curve in Fig. 7-(a)). However, when p is fixed at p⁎ (it corresponds to the case with m= 0.5), more information is generated throughout LVE. Moreover, when the user diversity varies as before, the effect of price cap (the upper curve in Fig. 7-(b)) contrasts sharply with the amount of information created without the price cap (the lower curve in Fig. 7-(b)). In the former case, we find that the amount of information in LVE is even larger than that in CVE thanks to wider user space. Lastly, note that the price regulation is not the only measure to control the service fee charged by SP in LVE. A policy that promotes competition among SPs is also able to lower the service fee. The recent Facebook case described in footnote 19 holds the clue to this possibility. Indeed, the competition policy usually results in a better outcome than the price regulation since the latter is highly likely to slash the service level; whereas the former is less likely to negatively affect the service level. However, the network externality is sometimes powerful enough to shape the competitive landscape into a winner-take-all system. Accordingly, it may be necessary to introduce a competition policy like the ‘market-dominating position’ to the value ecosystems in order to

26

D. Kim / Technological Forecasting & Social Change 107 (2016) 13–27

Fig. 6. Comparison of performance across the ecosystems: effects of Δ changes. (a) Relative profits change as Δ varies. (b) Relative user cost terms change as Δ varies.

keep a minimal degree of competition among SPs. In this context, the Korean Fair Trade Commission's recent attempt (i.e., screening) to impose the market-dominating position upon Naver and Daum-Kakao (simply on the basis of their sizes) established a precedent. The policy authorities in France and German seem to follow a similar path (the French Digital Council, 2014; Rosati, 2013; Wang et al., 2013). 6. Conclusion We develop two ecosystem models for the social media services. The key difference between the current ecosystem (CVE) and a new one (LVE) come from the way of service provisioning throughout the ecosystem (that is, differences in the SDPs). We also present a multi-stage game model, where the decisions of players—users, CPs, SP—are interrelated with each other and the decision process follows a certain format: for example, in LVE (or in the SB-L scenario) the decision process is totally sequential. According to our analyses and experiments, it is more likely that the overall level of social media service led by SP in LVE is higher than that in CVE. However, the total amount of information created throughout CVE is larger than that in LVE. These findings imply that the leadership of SP in a value ecosystem may boost or constrain the entire ecosystem. We also demonstrate that the user diversity measured by the span of users' preference to the social media services (Δ) is another key parameter for value creation in the value ecosystems.

As a conceptual, mathematical approach, our models have some limitations. First, we had to make some assumptions to conduct rigorous analyses, thereby reducing the applicability of the models. A couple of simplifying assumptions to note are as follows. First, we assume that service configuration efforts in CVE are the same across the users and fixed at constant e. The marginal profits m# also remain the same; they are not altered by providers' decisions. Third, user type k is assumed to follow a uniform distribution; it may be necessary to try different distributions for user type. Lastly, the overall service level and the amount of information are assumed to be identical. Without these assumptions, however, tractable analysis would become impossible and analysis would result in too complicated, long formula to be interpreted. And a stylized approach based on an abstraction of reality is likely to miss some aspects of reality. Moreover, the continuously changing nature of the ICT industries and services makes it harder to develop such a typology as carried out in this study. Since the subject is a moving target, some notions used in the modeling are not able to have a perfectly matching case or a representative feature (data) in practice. In our future works, we will first develop some measures to quantify core notions employed in the value ecosystem models (e.g., the level of social media services and the amount of information created), thereby conducting an empirical test about the models presented here. It is also necessary to investigate different evolutionary scenarios for the social media platforms. For example, the most promising path for the next

mS Fig. 7. Comparison of information created: with given p in LVE. (a) Relative information created changes as ðm Þ varies. (b) Relative information created changes as Δ varies. C

D. Kim / Technological Forecasting & Social Change 107 (2016) 13–27

generation social media services will be expanding the boundary of the value ecosystem as well as migrating into LVE. Acknowledgment “This work was supported by a grant from the Kyung Hee University in 2012” (KHU—20120757). References Adner, R., 2006. Match your innovation strategy to your innovation ecosystem. Harv. Bus. Rev. 84 (4), 98. Anderson, C., Wolff, M., 2010. The web is dead. Long live the Internet, Wired Magazine 18 (August 17, 2010), 2010. Armbrust, M., et al., 2010. A view of cloud computing. Commun. ACM 53 (4), 50–58. Basole, R.C., Karla, J., 2011. On the evolution of mobile platform ecosystem structure and strategy. Bus. Inf. Syst. Eng. 3 (5), 313–322. Berthon, P.R., Pitt, L.F., Plangger, K., Shapiro, D., 2012. Marketing meets Web 2.0, social media, and creative consumers: implications for international marketing strategy. Bus. Horiz. 55 (3), 261–271. Brynjolfsson, E., Hofmann, P., Jorddan, J., 2010. Cloud computing and electricity: beyond the utility model. Comm. of the ACM 53 (5), 32–34. Burtch, G., 2011. Herding Behavior as a Network Externality. Working Paper Series of the Department of MISTemple University. Ceccagnoli, M., Forman, C., Huang, P., Wu, D.J., 2012. Cocreation of value in a platform ecosystem: the case of enterprise software. MIS Q. 36 (1), 263–290. Chellappa, R.K., Shivendu, S., 2003. Online Personalization and Privacy Concerns: An Axiomatic Bargaining Approach. Proceedings of Information Systems & Technology, INFORMS-CIST Conference, Atlanta, GA, pp. 18–19. Media Culpa, 2012. Has traffic to Pinterest plateaued? http://www.kullin.net/2012/06/ has-traffic-to-pinterest-plateaued (Accessed on June 30th 2012) Cusumano, M.A., 2010. Cloud computing and SaaS as new computing platform. Comm. ACM 53 (4), 27–29. Cusumano, M.A., 2011. Platform wars come to social media. Comm. ACM 54 (4), 31–33. Dong, D., Sun, L., Sun, Z., 2014. Web services in China. Handbook of Research on DemandDriven Web Services: Theory, Technologies, and Applications. El Sawy, O.A., Malhotra, A., Park, Y., Pavlou, P.A., 2010. Seeking the configurations of digital ecodynamics: it takes three to tango. Inf. Syst. Res. 21 (4), 835–848. Enders, A., Hungenberg, H., Denker, H.P., Mauch, S., 2008. The long tail of social networking: revenue models of social networking sites. Eur. Manag. J. 26 (3), 199–211. Evans, D.S., Hagiu, A., Schmalensee, R., 2005. A survey of the economic role of software platforms in computer-based industries. CESifo Econ. Stud. 51 (2–3), 189–224. Evans, D.S., Hagiu, A., Schmalensee, R., 2006. Invisible Engines: How Software Platforms Drive Innovation and Transform Industries. MIT Press. Evon, D., 2014. Facebook to start charging monthly fee? Old hoax is still a hoax. Social News Daily, Sept. 22, 2014 http://socialnewsdaily.com/. Gawer, A. (Ed.), 2011. Platforms, Markets and Innovation. Edward Elgar Publishing. Gawer, A., Cusumano, M.A., 2002. Platform Leadership. Harvard Business School Press. Gawer, A., Cusumano, M.A., 2014. Industry platforms and ecosystem innovation. J. Prod. Innov. Manag. 31 (3), 417–433. Han, J.K., Han, S., Vanhonacker, W.R., 1996. A habit-formation model of brand choice. HKUSF School of Management Working Paper, No. MKTG 96.077. Hanna, R., Rohm, A., Crittenden, V.L., 2011. We're all connected: the power of the social media ecosystem. Bus. Horiz. 54 (3), 265–273. Henderson, T., Bhatti, S., 2001. Modelling user Behaviour in Networked Games. Proceedings of the 9th ACM international conference on Multimedia, pp. 212–220. Huang, P., Ceccagnoli, M., Forman, C., Wu, D.J., 2009. Participation in a Platform Ecosystem: Appropriability, Competition, and Access to the Installed Base. NET Institute Working Paper, No.09–14. Iansiti, M., Levien, R., 2004. Strategy as Ecology. Harv. Bus. Rev. 82 (3), 68–81. Jansen, S., Cusumano, M.A., 2013. Defining software ecosystems: a survey of software platforms and business network governance. In Software Ecosystems: Analyzing and Managing Business Networks in the Software Industry. Edward Elgar Publishing. Karkkainen, H., Jussila, J., Vaisanen, J., 2010. Social media use and potential in business-tobusiness companies' innovation. Proceedings of the 14th International Academic Mindtrek Conference: Envisioning Future Media Environments, October 2010, pp. 228–236. Kim, A.J., Ko, E., 2012. Do social media marketing activities enhance customer equity? An empirical study of luxury fashion brand. J. Bus. Res. 65 (10), 1480–1486. Kim, H., Lee, J.N., Han, J., 2010. The role of IT in business ecosystems. Commun. ACM 53 (5), 151–156. Kim, M., Xie, L., Christen, P., 2012. Event diffusion patterns in social media. Proceedings of the Sixth International AAAI Conference on Weblogs and Social Media, pp. 178–185. Kumar, V., Mirchandani, R., 2012. Increasing the ROI of social media marketing. MIT Sloan Manag. Rev. 54 (1), 54–61.

27

Kwak, M., 2001. Web sites learn to make smarter suggestions. Sloan Mgmt. Rev. 42 (4). Levy, A., 2006. Sincere social capital with material status sensitivity: index and an inverted U-shaped utility-wealth theory. Faculty of Commerce-Economics Working Papers, No.163. Michaelidou, N., Siamagka, N.T., Christodoulides, G., 2011. Usage, barriers and measurement of social media marketing: an exploratory investigation of small and medium B2B brands. Ind. Mark. Manag. 40 (7), 1153–1159. O'Reilly, T., 2005. What is Web2.0?: design patterns and business models for the next generation of software. http://oreilly.com/web2/archive/what-is-web-20.html (Accessed on June 30th 2012). Parker, G., van Alstyne, M.W., 2014. Platform Strategy Survey. SSRN Working Paper Series, No. 2439323. Philipson, T.J., Posner, R.A., 1999. The Long-run Growth in Obesity as a Function of Technological Change. NBER Working Paper Series, No. w7423. National Bureau of Economic Research. Rifkin, J., 2014. The Zero Marginal Cost Society: the Internet of Things, the Collaborative Commons, and the Eclipse of Capitalism, Palgrave Macmillan. Rodero-Merino, L., et al., 2010. From Infrastructure delivery to service management in clouds. Futur. Gener. Comput. Syst. 26, 1226–1240. Rosati, E., 2013. The German ‘Google Tax’ Law: groovy or greedy? J. Intellect. Prop. Law Pract. 8 (7), 497-497. Rosenbaum, S., 2011. Curation Nation: How to Win in a World Where Consumers Are Creators. McGraw-Hill. Rubinfeld, D.L., Pindyck, R.S., 2012. Microeconomics. eighth ed. Prentice-Hall. Sedghi, A., 2014. Facebook: 10 years of social networking, in numbers,. The Guardian, Feb. 4, 2014 http://www.theguardian.com/data (accessed on March 15, 2014). Soares, A.M., Pinho, J.C., Nobre, H., 2012. From social to marketing interactions: the role of social networks. J. Trans. Manag. 17 (1), 45–62. Sugiyama, D., Shirahada, K., Kosaka, M., 2012. Strategic 5Ps and their IT based service business model for corporate sustainability. Technology Management for Emerging Technologies (Proceedings of PICMET'12), July 2012, pp. 1209–1215. Tee, R., Gawer, A., 2009. Industry architecture as a determinant of successful platform strategies: a case study of the i-Mode mobile internet service. Eur. Manag. Rev. 6 (4), 217–232. the French Digital Council, 2014. Platform Neutrality: Building an Open and Sustainable Digital Environment, Opinion no. 2014–2. the Statistics Portal, 2014. (http://www.statista.com/: accessed on December 2, 2014). Tiwana, A., Konsynski, B., Bush, A.A., 2010. Platform evolution: coevolution of platform architecture, governance, and environmental dynamics. Inf. Syst. Res. 21 (4), 675–687. van Noort, G., Antheunis, M.L., van Reijmersdal, E.A., 2012. Social connections and the persuasiveness of viral campaigns in social network sites: persuasive intent as the underlying mechanism. J. Mark. Commun. 18 (1), 39–53. Wang, Y., Vasilakos, A.V., Jin, Q., Ma, J., 2013. Survey on Mobile Social Networking in Proximity (MSNP): approaches, challenges and architecture. Wirel. Netw 1–17. Wu, J., 2014. How WeChat, the Most Popular Social Network in China, Cultivates Wellbeing Mimeo. Yoo, J.H., 2011. Service Platform Strategy: Social Networking and Mobile Service Platform Perspectives (Doctoral Dissertation) Massachusetts Institute of Technology. Dohoon Kim is professor of the School of Management at Kyung Hee University in Seoul, Korea. He received B.S. with honor in Economics (with minor in Statistics) from Seoul National Univ., and Ph.D. in Management Engineering from KAIST. He served as Fulbright scholar at UC Santa Cruz (2008–2009) and Univ. of Pennsylvania (2000–2001), where he conducted interdisciplinary researches on socioeconomic issues regarding the evolution of Internet technologies and services. He also actively partcipates in consulting projects for various ICT companies and governmanet agents in Korea. He published more than 20 scholarly papers in academic journals such as IEEE Network, Electronic Commerce Research & Applications, Telecommunication Systems, Annals of Telecommunications, Asia-Pacific Journal of Operational Research, Journal of Applied Mathematics, etc. He received several academic awards for his research papers and presentations from some Korean academic institutions such as the Korean Operations Research and Management Sience Society. Areas of Interest Business ecosystem and service delivery process: modeling, analysis and empirical study Information and network technologies and services: management, strategy, and policy Economics of telecommunication and media industries Two-sided markets and platform strategy Evolutionary games and dynamics