Data-Driven Health Services: an Empirical Investigation on the Role of Artificial
Intelligence and Data Network Effects in Value Creation
Waad Fadul
Field of study: Information Systems Credits: 30 credits
Presented: Spring 2020 Supervisor: Darek Haftor
Department of Informatics and Media
1
Abstract
The purpose of this study is to produce new knowledge concerning the perceived user’s value generated using machine learning technologies that activate data network effects factors that create value through various business model themes. The data network effects theory represents a set of factors that increase the user’s perceived value for a platform that uses artificial intelligence capabilities. The study followed an abductive research approach where initially found facts were matched against the data network effects theory to be put in context and understood. The study’s data was gathered through semi-structured interviews with experts who were active within the research area and chosen based on their practical experience and their role in the digitization of the healthcare sector. The results show that three out of six factors were fully realized contributing to value creation while two of the factors showed to be partially realized in order to contribute to value creation and that is justified by the exclusion of users' perspectives in the scope of the research. Lastly, only one factor has limited contribution to the value creation due to the heavy regulations limiting its realization in the health sector. It is concluded that data network effects moderators contributed differently in the activation of various business model themes for value creation in a general manner where further studies should apply the theory in the assessment of one specific AI health offering to take full advantage of the theory potential. The theoretical implications showed that the data network factors may not necessarily be equally activated to contribute to value creation which was not initially highlighted by the theory. Additionally, the practical implications of the study’s results may help managers in their decision-making process on which factors to be activated for which business model theme.
Keywords
Data Network Effects, Artificial intelligence, Machine Learning, Value creation, Business model themes, Digital health, Data-driven health services
2
To my father Mohamed Abdelhalim Fadul – to whom I promised to dedicate this dissertation before he left this world This is for you.
3
ACKNOWLEDGMENT
First and foremost, I would like to thank the Swedish Institute (SI) for awarding me the Swedish Institute Scholarship for Global Professionals (SISGP). My master's studies in Sweden would not have been possible without this opportunity.
Furthermore, I would like to thank my supervisor Darek Haftor for his unwavering support and excellent guidance throughout the process of writing this thesis. I would also like to convey my heartfelt gratitude to Pascal Fahrni who has greatly assisted me by sharing his profound knowledge and experience.
Finally, my enormous gratitude for my Mother, Salwa Osman, whose prayers were with me every step of the way. To my siblings, Sami, Tamir & Weam, this would not have been done without your unconditional support from miles away. And to my friends, your motivation was the fuel that kept me going.
4
Content
CHAPTER ONE ... 6
INTRODUCTION... 6
1.1 Introduction ... 6
1.2 Background ... 6
1.3 The problem overview ... 8
1.4 Research Question ... 9
1.5 Delimitations ... 9
1.6 Disposition ... 10
CHAPTER TWO ... 11
THE THEORETICAL FRAMEWORK... 11
2.1 Introduction ... 11
2.2 Artificial Intelligence ... 12
2.3 AI and Health ... 13
2.4 AI Value Creation ... 13
2.5 Network Effects... 15
2.6 Data Network Effects ... 18
2.7 Business model themes ... 27
CHAPTER THREE ... 32
METHODOLOGY ... 32
3.1 Research strategy... 32
3.2 Research methods ... 34
3.3 Analysis Method ... 37
3.4 Quality Improvement Measures ... 39
3.6 Ethical consideration ... 41
CHAPTER FOUR ... 43
RESULTS AND ANALYSIS ... 43
4.1 Results of the thematic Analysis ... 43
4.1.1 Data Stewardship ... 46
5
4.1.2 Platform Legitimacy ... 50
4.1.3 User-Centric Design ... 54
4.1.4 Value Creation Themes ... 58
CHAPTER FIVE ... 64
DISCUSSION ... 64
5.1 Theoretical Implications ... 64
5.2 Practical Implications ... 68
5.3 Limitations and future research ... 70
CHAPTER SIX ... 71
CONCLUSION ... 71
REFERENCES ... 73
APPENDICES ... 79
Appendix 1: Interview Guide ... 79
List of Figures FIGURE 1NETWORK EFFECTS WITH PHONES ... 15
FIGURE 2UBER'S NETWORK EFFECTS DUCHATEAU (2017) ... 16
FIGURE 3TYPES OF NETWORK EFFECTS ... 17
FIGURE 4MODEL OF DATA NETWORK EFFECTS ... 20
FIGURE 5BUSINESS MODEL THEMES ... 27
FIGURE 6RESEARCH METHODOLOGY ... 32
FIGURE 7DATA QUANTITY ACTIVATION LEVEL ... 48
FIGURE 8DATA QUALITY ACTIVATION LEVEL ... 50
FIGURE 9PERSONAL DATA USE ACTIVATION LEVEL ... 53
FIGURE 10PREDICTION EXPLAINABILITY ACTIVATION LEVEL ... 54
FIGURE 11PERFORMANCE EXPECTANCY ACTIVATION LEVEL ... 56
FIGURE 12EFFORT EXPECTANCY ACTIVATION LEVEL ... 58
List of Tables TABLE 1LIST OF RESPONDENTS ... 36
TABLE 2THEMATIC ANALYSIS -PART-1... 43
TABLE 3THEMATIC ANALYSIS PART-2 ... 44
TABLE 4THEMATIC ANALYSIS PAT-3 ... 45
TABLE 5STUDY FINDINGS ... 65
TABLE 6THEMES AND MODERATORS MATRIX ... 69
6
CHAPTER ONE INTRODUCTION
1.1 Introduction
The first section is the background of the subject for the thesis. Following this, the research problem, the motivation for the thesis as well as thesis limitations, are defined, described, and presented. Lastly, the outline of the thesis.
1.2 Background
The idea of the Apple store that was founded about 10 years ago has revolutionized business models and created a new generation of business models that are based on platform concept and network effects. The concept of network effects has disrupted a wide range of traditional businesses in several industries. The main ones are Apple in the mobile phone industry, Amazon in e-commerce, Airbnb in hotels, Uber in transport, LinkedIn in professional interactions, and many others. Marshall et al., (2016), Clemons (2009) and Eisenmann et al. (2006) have produced literature to provide a better understanding of the mechanisms behind this phenomenon.
Network effects can be defined as the influence of a platform's number of users on the value it generates for each user (Marshall, 2017). Since there are more connections to leverage, more value is given to users who use the platform. A type of network effect is the data network effect (DNE).
According to Gregory et al. (2020), data network effects occur as the platform learns more from the accumulated data of its users, and the platform's value increases with each of its users and this is referred to as user’s perceived value. A user is defined as everyone who used the platform for one or more times to receive the service that the certain platform offers (Marshall, 2017). On the other hand, user’s value is the value that the user adds to the platform through the contribution of his/her data and usage. Gregory et al. (2020) developed a data network effects model to complement network effects theory, and this model is to answer the question of how value is derived from data with artificial intelligence and machine learning (AI/ML) for each user of the platform. Gregory et al. (2020) propose a model to explain the role of AI/ML in the creation of user value, which is described as the value perceived by platform users. According to the model,
7 three moderators moderate this relationship: 1) platform legitimacy, 2) data stewardship, and 3) user-centric design.
Artificial Intelligence (AI) is a term used to describe a technology that allows businesses to create more value for their customers. This stems from advancements in AI techniques that enable us to imitate our cognitive actions while also automating the processes of recognizing and solving complex problems (Zhuang et al., 2017; Lee et al., 2019). AI offers opportunities to increase operational performance and drive innovation by offering information from vast data sets and forecasting unpredictable events, according to Lee et al. (2019), which piques interest in a wide range of industries. Artificial intelligence (AI) technologies and creativity have been used in a variety of business industries, and they have recently begun to be used in the healthcare field. As artificial intelligence (AI) applications and innovation were applied in different business sectors and recently started to be implemented in the healthcare sector as well. Recent studies have shown the transformational impact of artificial intelligence and machine learning in patients care, the pharmaceutical industry, and even the administrative processes of healthcare providers. These studies suggest that AI capabilities have the potential to outperform humans at key healthcare tasks such as diagnosing analysis. Nowadays, algorithms are providing more accurate results than radiologists in identifying tumors as well as guiding researchers in their clinical trials (Davenport
& Kalakota, 2019). Machine learning and Artificial intelligence solutions can be used in a variety of health-related applications, including assisting physicians in developing more customized medications and procedures for patients, as well as assisting patients in determining when and whether they can schedule follow-up appointments (Bhardwaj et al., 2017).
Artificial Intelligence (AI) is a technology that, when incorporated into healthcare apps and smart wearables like Fitbits, can predict the occurrence of health conditions in users by collecting and analyzing their health data (Bhardwaj et al., 2017). The widespread integration of AI with wearable devices and smartwatches can be proved to be the solution to both reduce the rising costs of healthcare and help in improving the relationship between doctors and their patients (Bhardwaj et al., 2017). Menictas et al., (2019) discussed how smartphones and wearable health devices assist in enacting health behaviors through messages or notifications. The implementation of AI provides a tool for effective data use to address the users’ health behaviors. In mobile Health, this means learning from users’ data in order to solve problems, suggest solutions and predict certain
8 outcomes. AI in mHealth aims to augment human capacities to improve self-management of healthy habits.
1.3 The problem overview
In this era of digital technology and transformation, Telemedicine, and electronic health (eHealth) are introduced to reshape and redefine the delivery of health services through telecommunication.
According to Stubberud (2018), little research has been done about how e-Health can be used for delivering better health services but there is a knowledge gap of how eHealth and mobile applications can be designed, developed, and delivered to create new values for patients as well as helping the healthcare providers deliver better services. In addition, there are several developed applications and platforms but still, there is a gap between commercially available applications and scientifically validated and developed applications (Duchateau & Lejeune, 2017).
Several eHealth services in the Swedish market are implementing Artificial Intelligence to create value for their users. The primary outcome of the data network effects (DNE) model is the
“perceived user value” and the model suggests that the AI capabilities have a positive impact in increasing the perceived user value. The DNE model suggests that for an AI capability to increase the user’s perceived value through high speed and accuracy outcomes, three moderators are governing this connection: 1-Data Stewardship: which mainly focusing on the data quantity and quality that the AI capability is using. 2- Platform Legitimacy: which represent the trust of the user’s data privacy and security and clarity of the AI capabilities results, and 3- User-centric design: that focus on the user’s commitment to use the platform. Each of these three moderators has two subfactors that the DNE model suggests if realized, will increase the user’s perceived value.
However, the DNE model is a new concept that was not empirically tested and validated in AI- driven health offerings. Therefore, this research has identified to main knowledge gap that this study aims to fill. Firstly, the DNE model suggested by Gregory et al. (2020) did not discuss if these factors are equally important or the possibility of increasing user’s perceived value through the realization of only part of the factors should be considered. The model was not empirically tested in health services, therefore the partial realization of the DNE factors was not investigated.
Secondly, The DNE model has focused on big platforms with huge user’s installed base, hence
9 providing little information about how data-driven learning can help medium and small-size platforms to create user value.
This research aims to fill in this knowledge gap by investigating how existing data driven e-health offerings are creating value for its users by the full or partial realization of the DNE factors.
1.4 Research Question
• What are the factors that condition value creation using AI/ML technology in order to deliver health care services?
• What are the factors hindering the value creation of AI-enabled health services?
1.5 Delimitations
It is important to mention that there exist several delimitations in this study. Firstly, this study delimited itself to investigate the role of AI and data network effects in value creation in the Swedish digital health sector and investigating which of the business model themes are activated in the value creation process. This research does not intend to pinpoint the evolution of business models or AI adoption in the digitalized health field, or the improvements that have occurred over time.
Additionally, the organizations that were selected to participate in this study are delimited to either digital health service providers or digital health regulatory agencies which are operating to create value. In this study, only the perspective of service providers and digital health public agencies was considered even though the researchers are aware that the value creation is influenced by the patients and end users too. In that matter, the patient’s perspective should be considered in the possible future research.
Additionally, this study will use two theoretical frameworks, firstly, the data network effects its factors to increase value creation, and secondly, the theory of business model themes that categorize value creation to four different themes. As a result, this study will not address how digital health service providers capture value, nor will it discuss theories relevant to AI-driven value capture activities.
10 Given the research questions and the theoretical models, the research will address and discuss the theoretical frameworks and thematic analysis will be performed to process the collected data. An abductive approach will be followed in the research as it is a good fit when encountering surprising and anomalous observations that do not fit the chosen theories.
1.6 Disposition
Chapter 2 Theoretical frameworks
This chapter consists of the theoretical frameworks that have been used in this study. Artificial intelligence is introduced first. Then, network effects and its types were highlighted to help introducing the primary model of the Data Network Effects is presented, and the second model of Business model themes is introduced. Finally, an analysis of how these theories will be used to answer the research question will be showed.
Chapter 3 Methods
The methodology of this study is presented, starting with the research strategy, followed by the data collection methods and the data analysis methods.
Chapter 4 Results & Analysis
This chapter presents the empirical findings from the interview. The data is also analyzed based on the theoretical framework presented in chapter 3.
Chapter 5 Discussion & Implications
The research findings are discussed. Theoretical and practical implications are presented, findings were summarized, and further research is suggested.
Chapter 6 Conclusion
Concluding the research and show how the results answered the research questions.
11
CHAPTER TWO
THE THEORETICAL FRAMEWORK
2.1 Introduction
This research is focusing on the Swedish digital healthcare sector and its existing practices and methodologies to create value through the implementation of AI and data-driven solutions. This choice was based on the observation that this sector is offering an interesting empirical ground for research purposes for four reasons, 1- the fast-growing economy of Sweden, which allows for the establishment and operation of new companies in historically monopolized sectors such as health care. 2- High levels of digitalization and the implementation of new and cutting-edge technology across sectors, as well as the health sector's steady yet evolving approach to embracing AI and machine learning, and 3- The digital health industry is heavily regulated by a variety of government entities, and many laws and regulations must be followed. 4- the DNE theory was not empirically investigated in the health care that has AI-enabled services and the DNE factors was not tested given the high regulated industry.
The health sector adoption of data-driven technologies such as AI will introduce the data network effects as a model for value creation. The current institutional structure, on the other hand, heavily regulates the realization of data network effects and its value creation. Therefore, this study would look at how e-health providers are realizing data network effects and which of the business model themes were activated to ensure the DNE value creation.
In this section, conceptual background and theoretical frameworks related to Network Effects and Data network effects model, and business model themes will be reviewed. The first part will focus on defining Artificial intelligence as primary technology and its implementation in the health sector, then and the prior studies of Network effects and data network effects (DNE) theories is detailed. Furthermore, the theory of the business model themes (BMT) will be introduced. These two theoretical models (DNE and BMT) were chosen as they are related in the concept of value creation. The DNE moderators increase the user’s perceived value, which can be realized through the activation of different strategies that represented by the BMT theory. The two theories will be
12 utilized to answer the research question through investigating how the DNE moderators contribute to the activation of the different BMT which create value for users.
2.2 Artificial Intelligence
Ertel (2011) provides a list of historically important AI definitions. First, there's John McCarthy's 1955 definition: " The goal of AI is to develop machines that behave as though they were intelligent" (Ertel 2011, p. 13), which Ertel considers inadequate. He also criticizes encyclopedias' definitions of AI, which he believes have a number of flaws. Finally, he settles on Elaine Rich's definition from 1989: “Artificial [i]ntelligence is the study of how to make computers do things at which, at the moment, people are better” (Ertel 2011, p. 14), which he describes as "elegant and timeless".
Corea (2017, p. 1–2), on the other hand, defines AI as follows: “[a] system that can learn how to learn, or in other words a series of instructions (an algorithm) that allows computers to write their own algorithms without being explicitly programmed for” (Corea 2017, p. 2). Corea's definition is excessively limited for the sake of this thesis. This concept focuses primarily on machine learning (see chapter 2.1.2), however AI as applied to an organizational environment requires a broader definition.
The distinctions between strong and weak artificial intelligence are becoming more generally used definitions (Etzioni & Etzioni ,2017). Weak artificial intelligence is defined as AI that assists users but is not self-sufficient, lacks ethics, is standardized for its function, and cannot pass the Turing test (i.e. cannot be mistaken for a person) (Etzioni & Etzioni ,2017). Strong artificial intelligence is expected to be autonomous, adaptive, cognitive and be able to pass the Turing test (Etzioni &
Etzioni 2017).
Finally, Bataller & Harris (2016) suggest that artificial intelligence (AI) “consists of multiple technologies that enable information systems and applications to sense, comprehend and act. That is, computers are enabled (1) to perceive the world and collect data; (2) to analyze and understand the information collected; and (3) to make informed decisions and provide guidance based on this analysis in an independent way” (Bataller & Harris 2016, p. 6). Their definition strikes a good blend of narrowness and breadth. It is a modern definition intended for usage in a business setting.
13
2.3 AI and Health
As artificial intelligence (AI) applications and innovation were applied in different business sectors and recently started to be implemented in the healthcare sector as well. Recent studies have shown the transformational impact of artificial intelligence and machine learning in patient care, pharmaceutical industry and even the administrative processes of the healthcare providers. These studies suggest that AI capabilities have the potential to outperform humans at key healthcare tasks such as diagnosing analysis. Nowadays, algorithms are providing more accurate results than radiologists in identifying tumors as well as guiding researchers in their clinical trials (Davenport&
Kalakota, ,2019). Machine learning and Artificial intelligence solutions can be used in a variety of health-related applications, including assisting physicians in developing more customized medications and procedures for patients, as well as assisting patients in determining when and whether they can schedule follow-up appointments (Bhardwaj el al. ,2017).
Artificial Intelligence (AI) is a technology that, when incorporated into healthcare apps and smart wearables like Fitbits, can predict the occurrence of health conditions in users by collecting and analyzing their health data (Bhardwaj el al. ,2017). The widespread integration of AI with the wearable devices and smartwatches can be proved to be to be the solution to both reduce the rising costs of healthcare and help in improving the relationship between doctors and their patients ((Bhardwaj el al. ,2017). Menictas et al., (2019) discussed how smartphones and wearable health devices assist in enacting health behaviors through messages or notifications. The implementation of AI provides a tool for effective data use to address the user’s health behaviors. In mHealth, this means learning from user’s data in order to solve problems, suggest solutions and predict certain outcomes. AI in mHealth aims to augment human capacities to improve self-management of healthy habits [30].
2.4 AI Value Creation
Value creation is defined by Chesbrough (2007) as the sequence of activities that is required to be followed in order to solve a problem and meet customer’s expectations. Barney (1999) stated that value creation originated from identifying customers' needs and issues and how certain technologies are utilized and which activities to be performed in order to fulfill these needs and solve these pain points. Value creation also includes the resources and capabilities that are required
14 to generate a value that is desired by customers through the cooperation with firms, providers and partners.
According to Agrawal et al., (2019) and Cockburn et al., (2018), AI value creation process can be categorized into two types of activities, 1) Activities focused on increasing efficiency and reducing costs which refers to improvements and maintenance of existing operations. In these activities, the output of these operations will not change and the customer will not realize the transformational refinements’ impact on his perceived value. Furthermore, AI can substitute parts of the decision making and do repetitive tasks that are done by labor which result in reducing the activities costs (Agrawal et al., 2019). and 2) activities that focus on revenue growth through the value AI creates in the decision making that improve the outcomes (Cockburn et al., 2018). This essentially means that data is fed into AI algorithms, which then produce knowledge that was not previously available, as AI is capable of solving complex problems and providing more accurate insights.
This allows humans and technology to take more accurate decisions with the help of more accurate data. From a market standpoint, AI assistance allows managers to make more accurate decisions in the face of uncertainty, thus complementing decision-making processes and improving their outcomes.
However, AI should not be seen as a stand-alone innovation that generates value. To fulfill its function, Pisano and Teece (2007) argue that any innovation needs complementary products, technologies, or services. Understanding how AI generates value is critical, as is ensuring that complementary products, technologies, or services are accessible within the value creation network. Because of the lack of understanding of AI applications, how value is generated, and how it can be captured, further research is needed.
15
2.5 Network Effects
Researchers in industrial economics introduced several definitions for the network effects concept. The definitions from frequently cited papers are:
1. “The benefit that a consumer derives from the use of a good often depends on the number of other consumers purchasing compatible items” Katz and Shapiro, (1986).
2. “A good is often more valuable to any user, the more others use compatible goods” Farrell and Saloner (1986); and
3. “A network externality exists when the value of consuming a particular product or service increases in the number of consumers that use compatible products or services” Gandal (1995).
Marshall et al., (2017) has defined network effects as the impact a platforms’ number of users has on the value the platform produces for each user. Figure-1 illustrates the network effects in the case of phone lines. More users who use the platform mean more value offered to these users because there are more interactions to leverage for the platform’s managers. The same phenomenon is observed in platform-based companies.
Figure 1 Network Effects with Phones
Another example is Uber, where the larger size of the platform user base results in a better matching mechanism between the demand and offer. This represents the network effects which only focus on the number of the platform’s users. (Liu et al., 2016) have highlighted another effect that is called the “data-driven network effect” which refers to the data collection on the platform
16 that contributes to the enhancement of the efficiency of the connection between drivers and passengers through Uber’s ranking system.
Duchateau (2017) states that the platform's transaction volume grows as a result of the virtuous circle. Figure 2 depicts the ecosystem's dynamic as a result of users’ network effects. Each line represents the value creation between consumers and producers. If there are more drivers or passengers, it can directly influence the service given to each new customer by shortening or lengthening pickup times or changing rates.
Figure 2Uber's Network Effects Duchateau (2017)
Belleflamme (2016) states another interesting approach: “Network effects are present if users think about the engagement and decisions of other users”. The volume of users and the quality of each user are two dimensions of other users' participation on the platform. In certain platforms, the users’ volume is the primary incentive for a new user to join the platform; in others, the identity or quality of the participants is the primary incentive for a new user to join that platform. Reillier et al., (2017) gave an example of The Diners Club platform that facilitates transactions between restaurants and consumers. The amount/volume of users is critical in this case. Indeed, the motivation for a restaurant owner to participate in the platform is largely determined by the number of potential customers on the platform, rather than the identity of these customers. Network effects, on the other hand, aren't just about the number of people on a platform; they're also about the strength of interactions between those people. Network effects are reduced if users are not ‘active' enough on the platform.
17 To better understand the dynamics, the relationship between the size of the network and the value generated has been divided into several sub-categories. The literature distinguishes between two forms of network effects: direct network effects and indirect network effects, as seen in the diagram below. The sum of those two distinct effects is the total network power.
Figure 3Types of Network Effects
Church & Gandal (1992), Katz & Shapiro (1992) and, Rochet & Tirole (2006) explored how increases in the network size of one user group can create a virtuous cycle with increases in the network size of either the same user group (direct network effects) or another user group, providing complements to the platform (indirect network effects).
2.5.1 Direct Network Effects
Direct network effects, also known as same-sided network effects, occur when the importance of a technology to a user is proportional to the number of other users. Those effects, according to Marshall (2016), are caused by the impact of users from one side of the market on users from the same side of the market. In other words, it's the effect of one customer on another, or one producer on another. The advantages or costs that users receive from the involvement of other users on the same side, rather than those on the opposite side, are measured by direct network effects. Direct network effects, according to Gawer (2014), shape demand sides of the economy of scale. The economy of scale allows for a decrease in cost per unit as the number of output units increases.
Because of direct network effects, the platform approach is concerned with the ability to produce cost optimization as more users are introduced to the platform. It is critical for a platform company to prioritize economies of scale by adding users rapidly and regularly.
18 According to Rochet & Tirole (2003), The benefit that users gain from a network comes from their ability to communicate directly with one another in the case of direct network effects. On social media networks, for example, network effects are largely caused by users communicating with one another, according to Gregory et al. (2020). However, negative same-sided effects with a threat to the number of consumers on one side of the market are possible. That means. the global value of each user decreases as new users joins the platform. Upwork, for example, is a website that links freelancers and self-employed individuals. It would draw jobs as the number of freelancers increases. However, Gregory et al. (2020) mentioned that as the number of freelancers outnumbers the number of employees, it becomes more difficult for workers and freelancers to find each other.
According to Gregory et al. (2020), there would be a reduction in the number of freelancers, resulting in a negative same-sided network impact. If the impact becomes too high, there will be insufficient freelancers on the site, and the number of jobs will begin to decline.
2.5.2 Indirect Network Effects
Marshall (2016) defines it as the effect that users on one side of the market have on users on the opposite side of the market. For a two-sided or multi-sided platform market, indirect network effects are critical. Cross-group or cross-sided network effects are other names for them.
According to Belleflamme (2016), we can examine a particular category of users whose activities affect the value offer of other groups on a variety of platforms. Reillier (2017) uses the example of operating system platforms with two distinct groups: software developers and consumers. The impact is the result of a virtuous circle in which more developers build more apps, which attracts more customers, who then encourage more developers to join in.
Another interesting definition from Hagiu and Wright (2011) is that indirect network effects develop when members of side A's decision is influenced by the number of users on side B's side, and vice versa. Platforms and network effects thus show a pre-existing fundamental interdependency and complementarity between two or more groups of users.
2.6 Data Network Effects
Data network effect is considered a type of network effect. Gregory et al. (2020) describe that data network effects are exhibited when the more the platform learns from the collected users’ data, the
19 more the value of the platform increases for each of its users. An example given by Gregory et al.
(2020) is Google as a research engine and the more it learns from users and their searches the more it can provide an individualized experience for each user which makes Google more valuable for the users. Gregory et al.’s (2020) framework that will be used in this research focuses on the role of AI and data network effects in value creation. Gregory et al. (2020) developed a model of data network effects that complements the network effects theory and this model is to address the question of how is value for each user of the platform created from data with AI.
Gregory et al. (2020) have developed the data network effects model that extends the theory of network effects. They propose that the platform’s ability to learn from the data generated by its users for the continuous improvement of the platform will result in creating new platform externalities. The manifestation of the improvements that are driven by the user utility through the utilization of AI would be in the form of greater product functionality, higher quality, and better user experience. The model of Gregory et al. (2020) of data network effects aims to explain this novel phenomenon.
The model proposed by Gregory et al. (2020) aims to explain the role of AI in creating user value which is defined as the value perceived by the platform’s users. The model suggests that this relationship is moderated by three moderators: 1) platform legitimation, 2) data stewardship, and 3) User-centric design. The model has proposed several propositions that explain how these moderators impact the relationship between the AI capability and the perceived user value. The structure for explaining Gregory et al’s. (2020) DNE model for the role of AI and data network effects in generating user value is shown in Figure 4 below.
20
Figure 4Model of Data Network Effects
According to Gregory et al. (2020) the DNE model was designed based on four main assumptions:
Assumption 1: The “computer in the middle of every transaction” transforms AI-enabled platforms into adaptable infrastructures that are capable of learning. For example, social networking platforms like Facebook use machine learning algorithms to help find and remove toxic content.
Assumption 2: Machine learning's strategic position in today's platforms emphasizes data as a key input into learning and value creation, transforming data into a valuable asset. Assumption 3:
Consumerization concept that was introduced by Gabriel, Korczynski, & Rieder (2015) has blurred the distinction between consumption and production and that resulted in transforming consumers into prosumers who co-create value. Gregory et al. (2020) provided an example of YouTube, where content creators consume and create marketing content at the same time, essentially co- creating value with brands and other content creators. Assumption 4: For a platform to make a success that lasts for a long term, the owners of the platform must ensure balanced and diverse stakeholders’ interests. This was derived from the fact that AI-enabled platforms can indeed affect the actions, perceptions, expectations, and emotions of people taking part in elections, demonstrations, and education, among other things, influencing the interests of a wide variety of stakeholders in sometimes contradictory ways.
Based on these four assumptions Gregory et al. (2020) were able to build the illustrated DNE model in figure 4 that will be explained in the following sections.
2.6.1 Platform AI Capability
21 Gregory et al. (2020) suggest that the platform AI functionality can be the engine driving data network effects. Derived from assumption 1, known as a platform's ability to learn from data in order to enhance its products and services for each user on a continuous basis. Meinhart (1966) identified that the key mechanism in which a platform’s AI capabilities can improve perceived user value is by enhancing its prediction capabilities. Churchman (1961) has defined prediction as the capacity of a certain system to produce information about the future by using current data from the past and present. An example Gregory et al. (2020) provided was about how a lending platform’s creditworthiness decision is based on estimating the probability that someone will repay a loan using current data on users and previous transactions.
Computations enabled by a platform AI capability will effectively result in two main improvements: prediction speed and prediction accuracy (Agrawal et al., 2018). Both improvements and their impact should be considered to understand how the platform AI capabilities can impact user-perceived value (Gregory et al. ,2020).
Speed of Prediction
Gregory et al. (2020) mentioned that the participation of the platform’s users to use its products and services ensures that they are free agents where they can autonomously decide when to do certain actions. Taking Facebook and Uber as examples where users get to decide when to post, accept requests, and reject trips. This user’s autonomy resulted in considering time as a bounding factor for the exchange relationship that was affected by the actions made by the users at a specific time. Each user’s activity on the platform can have an impact on the network’s structure, which can potentially create new interactions between the users.
Gregory et al. (2020) claim that a platform AI capability responds to a change can help produce value-enhancing interaction through the minimization of time between the occurrence of a network structure change and the network detection of that change to produce recommendations and new interactions. The best-case scenario is for a network to carry the ability of instant prediction of any network dynamics that may cause a destruction of user value. An example of this case is Uber’s mechanisms to detect any fraudulent activities for prearrangement which directly impact the open competition which reduce the user value. Another example is the machine learning algorithms
22 implemented by Facebook to detect false news and misinformation that are being shared in the platform which increase the perceived value of the platform for its users (Gregory et al. (2020).
Based on the previous understanding, Gregory et al. (2020) made a proposition that says” the greater the speed of prediction, the higher the perceived user value is likely to be.”.
Prediction Accuracy
As Gregory et al. (2020) illustrated above, the platform AI capability learning is not limited to the data collected by the network but also the influence that it has on the network and its users’
interactions. This influence occurs by linking the prediction capability and machine learning algorithms with the platform’s product and services. This influence has a significant impact on main network characteristics, such as network users' perceptions of confidence. Gregory et al.
(2020) explained the impact of the networks influence of prediction on the users trust through the example of Facebook’s algorithm that works on filtering, curating and ranking of the information has generated a perception of trust (and sometimes weaker perception of trust) depending on the prediction accuracy.
A platform AI capability that ensures increased prediction accuracy can minimize the difference between what has been recommended or forecasted and what users actually want which results in increasing the perception of trust of the network’s users. An example discussed is the gap between Uber’s driver pick-up time prediction and the actual time for a driver to pick-up the rider, if that gap is large, that may result in decreased perception of trust and vice versa (Gregory et al. ,2020).
2.6.2 Data Stewardship
Based on Assumption 2 that suggests data is a valuable asset, especially if it is utilized for user value creation through helping a platform's AI capability. Gregory et al. (2020) suggest that the effect of a platform’s AI capability on the user perceived value is directly impacted by the data quality and the data quantity. In addition, Gregory et al. (2020) define data stewardship as a DNE mechanism by helping a platform more valuable for its users through higher speed and accuracy of prediction by filling it with high data quality and quantity.
For full understanding of the moderating role of data quantity and quality, Agrawal et al., (2018) discussed how machine learning models require training data sets to iteratively build and modify
23 its prediction models to help produce more accurate and speedy results. Simon (1996) highlighted that the great amount of the data used for training machine learning algorithms leads to better prediction models which leads to benefiting the users of these prediction models. Accordingly, Gregory et al. (2020) suggests the consideration of both data quality and data quantity as moderating factors of the relationship between a platform AI capability and the perceived user value.
Data Quantity
For an increased prediction speed and accuracy, machine learning models are depending highly on the quality and quantity of data that is used in order for a platforms AI capability to positively impact the perceived user value (Gregory et al. ,2020). This is derived from a common reason that AI sometimes failed to accurately predict is the reliance on singular information typically produced from a single user or a small group of users with similar cases and to underweight distributional information. Kahneman & Tversky (1977) suggest that to avoid the prediction bias, a group of distributed cases of the same class to be used in the comparison of the case in hand. Accordingly, Gregory et al. (2020) suggested that the volume of past cases data has an immense impact on increasing the machine learning models’ prediction abilities in terms of speed and accuracy.
Based on the previous analysis, Gregory et al. (2020) made a proposition that says: “The higher the quantity of data for the training of machine learning algorithms on the platform, the stronger is the relationship between platform AI capability and perceived user value”.
Data Quality
Gregory et al. (2020) consider data quality of the data sets used to train machine learning models as a key pillar to increased prediction speed and accuracy. In addition, four main characteristics that describe the quality of data were highlighted by Gregory et al. (2020) which are: 1- data truthfulness which was defined as a degree of conformity between the recorded value of data and the actual value, 2- data completeness which refers to the existence of the record values for all the instances, 3- data consistency which explained as the level of which a certain value was measured with the same measurements across all recorded cases and 4- data timeliness which represents the speed which data is updated after the occurrence of a change.
24 According to Kahneman & Tversky (1977), the reduction of the likelihood of prediction overconfidence bias can be achieved through greater data quality which consequently strengthen the impact the platform AI capability has on the user value.
Gregory et al. (2020) concluded with a proposition that describes the relationship between data quality and perceived user value as follows: “The higher the quality of data for the training of machine learning algorithms on the platform, the stronger is the relationship between platform AI capability and perceived user value.
2.6.3 User Centric Design
Gregory et al. (2020) considered that a platform AI capability that is trained with high quantity and quality of data can help produce greater prediction speed and accuracy, but for better results, these trained AI models are represented through highly-designed products and services which users will interact with the AI model through. This engagement will result in a stronger perceived value for the platform AI capability. Based on Assumption 3 Gregory et al. (2020) made, the platform's consumerization has blurred the line between consumption and production which turned users into prosumers. Customer-centric design can be considered a key factor for the success of firms that adopt consumerization as it provides better understanding of users’ needs which will help increase the performance and efforts capacities of the products and services.
By understanding the users’ needs better and designing the platform’s products and services accordingly, this will result in better user engagement to co-create value in the platform. This engagement will contribute to improving the AI model and the platforms features. Therefore, Gregory et al. (2020) considered the user-centric design as a key mechanism of DNE by helping users have better experience with the platform and its accurate and speedy prediction. To conceptualize the user’s engagement, one factor to be considered is the intensity of which users interact with the platform’s products and services. In today’s platform business, the user’s engagement is usually measured by reporting the quantity of active users on a weekly or monthly basis which refers to the level of users’ commitment to use a certain product or service [15].
Gregory et al. (2020) suggest that both performance and effort expectancies of designed products and services must be considered as driving factors of the impact of platform AI capability on perceived user value [15]. Both will be explained and described in the following sections.
25 Performance expectancy
Gregory et al. (2020) defined the performance expectancy as the level of which a user trusts that the system will help them achieve gains in job performance. Based on Assumption 3 that mentioned above, the term “job” refers to the series of activities carried out by users by using the platform’s products and services. The users will measure the performance by evaluating to what extent the platform’s product and services helped achieve the task and fulfilled their needs. The performance expectancy is an indicator of the users willingness and commitment to use the platform’s products and services in both voluntary or mandatory settings.
Accordingly, Gregory et al. (2020) concluded that a higher performance expectancy can contribute to strengthening the impact of the platform AI capability on the user perceived value.
Effort expectancy
Gregory et al. (2020) highlighted that the level of which a user engages with a platform’s product or service is also affected by effort expectancy. Gregory et al. (2020) defined the effort expectancy to refer to the degree that the user thinks that the usage of the system requires minimal effort.
Similar to performance expectancy, the effort expectancy also plays a main role in the user's intention to adopt a certain system and his/her commitment to use it. The level of commitment and frequent usage is linked to how easy the platform’s product or service to use. The easier the use, the more committed the users become which leads to data-driven improvements for the AI models and its prediction speed and accuracy.
Based on this analysis, Gregory et al. (2020) proposed that: “The higher the effort expectancy of the platform’s products and services, the stronger is the relationship between platform AI capability and perceived user value.”
2.6.4 Platform Legitimation
Suchman, (1995, p. 574) has defined the term legitimacy as a generalized assumption that the action carried out by a certain entity complies with certain social constructs that are built based on norms, beliefs and definitions. Gregory et al. (2020) describe the platform legitimacy as the balance that is carried by the platform owners to represent diverse interests (Assumption 4) to mitigate potential risks related to data privacy and security. Based on this assumption, Gregory et al. (2020) consider that the responsible use of data and AI explainability are strategic factors that
26 contribute in strengthening the relationship between a platform AI capability and its users perceived value by mitigating and eliminating risks of data breaches and privacy violation. The platform’s legitimacy has two main characteristics: 1- The platform’s governance model for users’
data collection, storage and use and 2- The platform machine learning transparency and explain ability.
Personal data use
One key concern of the platform's legitimacy is the platform addresses the aspects of users' data collection, storage and usage. This factor to be addressed by the platform through clear and well- communicated privacy policies as well as information security compliance documentations.
Gregory et al. (2020) discussed that privacy-by-design and security-by-design are two principles that encourage the consideration of privacy and security in early stages of the engineering and development of the platform which will be reflected in the design of the platform's products and services.
This understanding of the impact of privacy and security on the design of the platform’s products and services led to this proposition suggested by Gregory et al. (2020): “The higher the moral desirability of the use of personal data by the platform, the stronger is the relationship between platform AI capability and perceived user value.”.
Prediction Explainability
The second characteristic of platform legitimacy is the interpretability and understanding of the prediction results made by the AI models. The predictions made by the AI-models have an impact on the emotions and behaviours of the platform’s users (Gregory et al. 2020). Therefore, the meaningfulness of the AI-model predictions and the ability to explain it to the users is considered as an important factor to strengthen the relationship between the platform AI capability and its users perceived value. An example that Gregory et al. (2020) discussed is the creditworthiness assessments that AI-model predicts in banks. The decision made by these predictions has an immense effect on the user who would be disappointed and question the legitimacy of the platform if they were not provided a comprehensive understanding of the factors that led to these predictions. Due to this issue, several banks have developed a scoring system and the components that impact the final score and how the AI-model calculates these components. Fostering the
27 explainability of the AI-model is likely strengthening the relationship between the platform AI capability and the perceived user value.
2.7 Business model themes
Amit & Zott (2001), have studied the evolution of e-business and have identified four main themes that represent the e-businesses source for value creation. Amit & Zott (2001) defined the word
"value" refers to the overall value generated in e-business transactions, regardless of whether the value is appropriated by the company, the consumer, or any other party involved in the transaction.
These sources of value creation are: novelty, complementarities, efficiency and lock-in. Each of these themes will be discussed below besides the linkage between them.
Figure 5 Business Model Themes
Efficiency
One of the primary value drivers for e-business is transaction efficiency, according to the data.
This result, which is in line with transaction costs theory Williamson (1975), suggests that transaction efficiency rises as transaction costs fall. As a result, the higher the transaction efficiency gains allowed by a particular e-business, the lower the costs and thus the higher the value (Amit & Zott, 2001).
28 Efficiency gains in offline businesses (i.e., those operating in conventional markets) and e- businesses can be realized in a variety of ways. One way is to reduce knowledge gaps between buyers and sellers by providing current and comprehensive information. This method is convenient and simple due to the speed and ease with which information can be transmitted and processed electronically. Costa Climent & Haftor (2020) have explained that the efficiency-centered business model is focused on some or all of the actors in the business model using relatively few resources.
This efficiency was one of Spotify's growth boosters, since it provided access to practically unlimited quantities of music anywhere, at any time, far more easily than alternatives like Apple's comparatively inconvenient file uploading transaction.
Complementarities
Complementarities exist if the total value of a package of goods is greater than the total value of each of the goods separately (Amit & Zott, 2001). Which means, Complementarity is a business model theme that relies on different forms of bundling or synchronizing products (goods or services), events, or resources, including technologies (Costa Climent & Haftor ,2020). According to Kulins et al. (2016), the emphasis here is on the business model's complementarity of offerings, with the potential to build synergies where A creates more value in the presence of B than it does on its own or in the presence of C being a key factor (Costa Climent & Haftor ,2020). These complementary products can be in the form vertical complementarities (provided by the same company) or in a form of horizontal complementarities that are provided by other partner companies (Amit & Zott, 2001).
Brandenburger and Nalebuff (1996) have emphasized the importance of delivering complementary outputs to customers in the strategy literature. Customers like your product more when they have the other player's product than when they have your product alone, according to them (Brandenburger and Nalebuff, 1996: 18).
Amit & Zott’s (2001) research also revealed the interconnectedness of value creation sources.
Information technology-enabled efficiency improvements pave the way for e-business complementarities to be exploited. When transaction costs are low, it makes sense to weave together the resources and capabilities of different companies, which is a hallmark of e-businesses.
The opposite is also true, complementarities can result in increased efficiency, at least from the
29 perspective of the consumer. Customers may be more efficient if they have access to goods and services that are complementary to the primary product of interest (Amit & Zott, 2001).
Lock-in
Amit & Zott (2001) explained lock-in as the degree to which consumers are encouraged to participate in repeat transactions (which tends to increase transaction volume) and the extent to which strategic partners have incentives to sustain and expand their relationships, enhances the value-creating potential of e-businesses and these attributes can be achieved through the lock-in business model theme. The key factor of this theme is to prevent customers, partners, suppliers and other actors from leaving the business and migrate to other competitors (Costa Climent &
Haftor ,2020).
Customer retention can be improved in many ways, according to the data review. First, loyalty programs (Varian, 1999) may be created to reward regular customers with special incentives.
Second, companies can establish dominant design proprietary specifications for business processes, products, and services (for example, Amazon's patented shopping cart) (Teece, 1987).
Third, businesses can build connection with customers by providing them with transaction security and reliability guaranteed by independent and highly reliable third parties.
Customer learning is required to become familiar with a website's interface design; after this learning has started, this will inhibit the customers from transitioning to other digital platforms, where their learning will have to begin all over again (Smith, Bailey, and Brynjolfsson,1999).
When possibilities for customization (by the customer) and personalization (by the e-business) are taken advantage of, this argument gains strength (Amit & Zott ,2001). Amit & Zott’s (2001) research findings indicate that e-businesses increase consumer loyalty by allowing them to customize goods, services, or information to their specific needs in a number of ways.
Novelty
Schumpeter articulated the value generating potential of inventions (1934). While the conventional sources of value creation by developments have been the launch of new goods or services, new methods of manufacturing, distribution, or marketing, or the tapping of new markets, Amit &
Zott’s (2001) data analysis shows that e-businesses often innovate in the ways they do business, that is, in the structuring of transaction.
