ESIC Digital Economy and Innovation Journal


Maximizing the benefits of machine learning: enhancing data network effects theory to improve value creation and appropriation

Maximización de los beneficios del aprendizaje automático: mejora de la teoría de efectos de redes de datos para mejorar la creación de valor y la apropiación

Ricardo Costa Climent 1

Uppsala University


Received: 03/12/2022
Accepted: 25/05/2023


The recently proposed theory of Data Network Effects aims to account for how user value is created from the use of machine learning technology. The theory accounts for the unique learning ability of machine learning, which uses large data sets to make predictions and enhance decision-making. This paper offers an assessment of the theory of Data Network Effects, identifying some of its strengths and limitations. Regarding the strengths, it contributes to the success of companies, accounts for the unique characteristics of ML technologies and is an advancement of the body of the theory of network effects. Their limitations are then transformed into a set of interrelated research questions that focus on the relationship of the use of machine learning and issues such as: value capture, a co-evolutionary view, a multi-actor perspective, and database dynamics. This paper outlines a multi-theoretical approach to study the value creation and capture enabled by the use of machine learning technologies.


artificial intelligence; machine learning; business model; network externalities; value creation from IT; Data Network Effect theory




La teoría recientemente propuesta de los efectos de la red de datos tiene como objetivo explicar cómo se crea el valor del usuario a partir del uso de la tecnología de aprendizaje automático. La teoría explica la capacidad de aprendizaje única del aprendizaje automático, que utiliza grandes conjuntos de datos para hacer predicciones y mejorar la toma de decisiones. Este artículo ofrece una evaluación de la teoría de los efectos de la red de datos, identificando algunas de sus fortalezas y limitaciones. En cuanto a las fortalezas, contribuye al éxito de las empresas, explica las características únicas de las tecnologías de ML y es un avance del cuerpo de la teoría de los efectos de red. Sus limitaciones luego se transforman en un conjunto de preguntas de investigación interrelacionadas que se centran en la relación del uso del aprendizaje automático y cuestiones tales como: captura de valor, una visión co-evolutiva, una perspectiva de múltiples actores y la dinámica de bases de datos. Este artículo describe un enfoque multiteórico para estudiar la creación de valor y la captura que permite el uso de tecnologías de aprendizaje automático.


inteligencia artificial; aprendizaje automático; modelo de negocio; externalidades de red; creación de valor a partir de TI; teoría de efectos de redes de datos




This article develops an ongoing research proposal that aims to assess the validity of the proposed Data Network Effects (DNE) theory and to improve its ability to shed light on value creation and value capture in firms using ML technologies. In doing so, we aim to offer novel perspectives on the creation and acquisition of value arising from the application of IT in firms.


To address the research questions raised in this research programme, we plan to adopt a qualitative research approach within the interpretative paradigm. This approach will allow us to understand the social reality that precedes the theoretical development. It will guide both the formulation of hypotheses and the application of methodology and scientific research tools.


This article does not incorporate empirical results as it is at an early stage of the research.


The resulting findings will advance the development of the novel theory of Data Network Effects that explains value creation through the use of machine learning technologies, as well as provide valuable insights for managers considering the integration of ML technologies into their business model.


Studies show that managers delay the adoption of artificial intelligence (AI) and machine learning (ML) because they are unsure of how these technologies can help their companies (Bughin et al., 2017). AI Governance enables senior executives to attain their objectives by implementing reliable systems that automate processes and improve tasks that previously relied on intuition or basic data analysis while avoiding any adverse effects on employees (Papagiannidis et al., 2023). Several of the most profitable firms in the world use ML technologies to customise their offerings, including Amazon, Apple, Google and Facebook (Al Dakhil & Bayoumi, 2020). Despite the significant importance of AI implementation, practitioners and researchers are not yet aware of the success factors (Merhi, 2023). As say Enholm et al. (2022, p. 1709) “A recent study by MIT Sloan Management Review found that more than 80% of organizations see AI as a strategic opportunity, and almost 85% see AI as a way to achieve competitive advantage (Ransbotham et al., 2017)”. The navigation app Waze, for example, provides real-time route recommendations based on the continuous real-time collection of massive sets of various kinds of data. One example is drivers’ data, which is analysed for pattern identification and then used to derive rules to generate customised route recommendations for a given user at a given time and place. However, not all firms that use ML technologies succeed in creating value (Joshi et al., 2021). Managers, therefore, wonder what characterises the uses of ML that create value (Bughin et al., 2017).

Scholars have long investigated how the use of information technologies gives rise to value creation and appropriation (Kohli & Grover, 2008) in different countries (Deichmann et al., 2016), industries (Davidovski, 2018) and organisations (Tallon et al. 2000). One key message is that it is not the information technology (IT) per se that generates value but the specific uses of IT and the governance of such uses that unlock latent value (Chesbrough & Rosenbloom, 2002). Another key message is that value creation from IT use is a matter of IT complementarities, meaning that the use of IT must be aligned with other factors (e.g. operational, cognitive and HR) to generate value (Chesbrough & Rosenbloom, 2002). Throughout this theoretical field, IT is depicted as a homogeneous concept (Kohli & Grover, 2008; Bharadwaj, 2000; Brynjolfsson & Hitt, 1996; Kohli & Devaraj, 2003; Pandey & Mishra, 2021). Accordingly, the IT and information systems (IS) literature conceives the artefact of IT as a homogeneous whole, ignoring its idiosyncrasies.

The objective of this paper is to evaluate the validity of the proposed theory of Data Network Effects (DNE) and to enhance its capacity to clarify the generation and acquisition of value in companies utilizing ML technologies. By doing so, we aim to offer novel perspectives on the value creation and acquisition that arise from the application of IT in firms, providing valuable insights for managers considering the integration of ML technologies into their business model.

The paper outlines the main concepts related to the value creation and capture theory. Then, it offers a brief description of the theory of DNEs, followed by an assessment of this theory. The strengths of the theory of DNEs are briefly described, along with its possible limitations. Those limitations are transformed into research subquestions that constitute the proposed research programme. This research programme is aimed at discovering how to use ML technology for value creation and capture. The proposed methodological approach is described, and the document ends with a discussion of the conclusions.


The notion of IT’s business value pertains to the influence of IT utilization on a company’s performance, encompassing metrics such as productivity, profitability, costs, competitive advantage, inventory, and other performance indicators. The most fundamental and disputed question addressed is which value approach for the firm best contributes to overall (aggregate) social welfare (Windsor, 2017). Scholars such as Devaraj and Kohli (2003), Hitt and Brynjolfsson (1996), and Melville et al. (2004) have examined this concept. By leveraging IT capabilities to create and capture value, organizations can improve their business performance, generating a competitive advantage that leads to superior performance, as highlighted by Porter (2001), Bharadwaj (2000), and Chae et al. (2014). In this context, the term “value creation” refers to the coordination of activities required to produce a product or service that aims to address specific issues or meet recipients’ expectations, ultimately enhancing their perceived usefulness (Chesbrough, 2007). On the other hand, “value capture” is defined as the mechanism that ensures the value generated delivers a tangible or intangible return above the opportunity cost (Dubosson-Torbay et al., 2002; Sjödin et al., 2020).

Studies investigating the impact of information technology (IT) consistently reveal a positive correlation between IT and various aspects of firm or organizational value. IT, in this context, encompasses the hardware and software components of a system and can be perceived as a broader concept, such as a “digital option,” “infrastructural capability,” or an IT management variable such as “business-IT alignment” (Kohli & Grover, 2008). This change relates to organisations and involves the individuals who use and adopt digital technologies, participants in the ecosystems who co-create their value propositions, and geopolitical frameworks that regulate the industries in which organisations and individuals are embedded (Dąbrowska et al., 2022). However, IT alone cannot generate value but must be integrated into a value creation process alongside other IT and organizational factors that synergistically work together (Melville et al., 2004). These factors could pertain to the IT-based system, which comprises IT personnel and management, routines, and policies, or the organizational system, which includes non-IT personnel and management (Brynjolfsson et al., 2021a).

The dimensions and magnitude of the business value that IT can deliver are contingent on various factors, including the type of IT, management practices, and organizational structure, as well as the competitive and macroeconomic environment (Brynjolfsson et al., 2002; Dewan & Kraemer, 2000). Despite the acknowledgement that IT’s value hinges on multiple factors, current theories often portray IT as a homogeneous entity within a system. Such an approach assumes that any type of IT is interchangeable.

The exploration of the worth of IT utilization is a crucial information resource to understand how implementing IT, whether it is a payroll program, fingerprint recognition, or an ML technology system, can enhance a company’s value. Nevertheless, each IT system possesses unique traits that differentiate its capacity to generate value. Current research on IT’s potential to create and capture value in organizations often overlooks these distinct characteristics, at best characterizing IT as a tool that processes information in multiple ways.

In the realm of Information Systems (IS), there exist conflicting perspectives on the role and conceptualization of computer artefacts. Some scholars posit that the computer is the core element of research on IS, while others view it as on par with the social elements of IS, as highlighted in theories such as sociomateriality and social structuring. However, some scholars argue that the central nature of computer artefacts in IS research limits the development of the field, given the constant evolution of technology and business landscapes and the emergence of novel challenges. Despite ongoing debate, there is a lack of consensus regarding conceptualising computer artefacts within IS.

According to Chesbrough (2007), the adoption of emerging or disruptive technologies like ML often requires a transformation of the business model to fully utilize their potential. However, the unique capabilities of ML and the lack of understanding on how to convert technological progress into business benefits require further research for successful implementation. Lee et al. (2019) also argue that more research is needed to understand how emerging technologies, like machine learning, can be commercialized using different business model archetypes.

Shaw et al. (2019) highlight that machine learning, a subfield of AI, has a significant impact on all sectors. Machine learning involves learning, reasoning, and acting on data through computer programs that process data, extract valuable information, make predictions, and suggest actions. These ML technologies allow companies to leverage the massive increase in data, the ability to store large data sets, and the improved speed of computers to develop patterns of use or behavior and make predictions (Agrawal et al., 2018). This helps firms to obtain early information about customers or users, their interests, or possible environmental changes, enabling them to personalize offers, improve their position, and gain a competitive advantage.

Machine learning algorithms are useful tools that optimize tasks, maximize and automate processes, improve efficiency in firm activities, save costs, and optimize time and resources (Shaw et al., 2019; Arel et al., 2010). Regardless of the activity and context, machine learning is likely to improve operation results compared to other technologies if enough data is available (Agrawal et al., 2018).

Furthermore, ML systems are valuable allies in decision-making, as they learn from data, extract patterns, make predictions, reduce the cognitive costs of decision-making, and accelerate the execution of high-quality decisions, improving the performance of digital information (Brynjolfsson et al., 2021).

The review highlights that specific IT applications in business contexts can create and capture value for firms. Successful business projects demonstrate the valuable capabilities of ML technology. However, other firms that use ML technology fail to create enough value, resulting in significant losses or even closure. The theoretical framework on value creation and capture through ML technology is incomplete, as it does not explain the difference in outcomes between these cases.

Current research on the creation and capture of value from IT use in firms tends to overlook the different types of IT and their capabilities. In this review, we focus on ML’s unique characteristics and its ability to extend learning in the current context, which has been ignored until the recent proposal of the theory of DNEs. We will review this theory in the next section.


3.1 A new kind of network effect

Extensive literature demonstrates that certain applications of IT can facilitate the activation of network effects, which serve as a means of value creation. Network effects arise when a product or service’s value to its users depends not only on the product’s inherent benefits but also on access to the network of users employing that product. Economides (1996), Farrell and Saloner (1986), Katz and Shapiro (1985, 1992), and Liebowitz & Margolis (1994) have all discussed this concept.

In the context of stakeholder networks, the use of IT enables the continued, dynamic collection of large sets of diverse data that are analysed with various ML techniques to identify patterns. Such patterns may in turn be transformed into rules for predictions and decision making. This kind of network-based learning enables an increase in the quality and effectiveness of services (e.g. driving route recommendations updated with traffic). This improvement of quality and efficiency can satisfy existing users of the service, thereby retaining them, whilst attracting new users. Both of these outcomes lead to increased service usage, which, in turn, generates new user data. These data then feed the ML algorithms, which provide refined and updated patterns and thus rules for prediction and decision making, consequently offering even higher quality and efficiency. Moreover, as the loop continues, it deters other service providers from entering the market and competing with similar services because they lack both the user base and the data sets. For instance, Facebook uses ML to identify and eliminate toxic content, which increases user value whilst retaining existing users and attracting new users to the detriment of other similar social platform offerings.

3.2 Factors that co-condition user value

The proposed theory of DNEs, as introduced by Gregory et al. (2021), outlines the various factors and their interrelationships that jointly contribute to the success of a learning loop. According to the theory, the value derived from a product or service that utilizes DNEs extends beyond the network’s size. The scale of data-driven learning and improvements made through ML technology plays a critical role. DNEs suggest a direct and positive correlation between a platform’s AI capabilities (or its provider) and its users’ perceived value. The theory further asserts that this relationship is moderated by three factors: data stewardship, user-centred design, and platform legitimation, as illustrated in Figure 1. We elaborate on these moderators below.

Figure 1. Factors that moderate the relationship between AI capacity and value creation

Source: Gregory et al. (2021)

A key driving force of DNEs is the AI capability of the ML computations that analyse large sets of data. Exponential growth in data and computing power has fuelled rapid advances in data and computational technologies like artificial intelligence (AI) (Chhillar & Aguilera, 2022). The main way that a platform’s AI capability can improve users’ perceived value is by improving prediction (Meinhart, 1966). Prediction refers to the ability of a system to use existing data about the past and present to generate information about the future (Churchman, 1961). The calculations enabled by AI gives it a capability in terms of speed and accuracy of predictions. Depending on the specific application, users’ perceived value of a service depends on whether the service is provided with enough speed and accuracy (Agrawal et al., 2018). For example, a driver stopped at an intersection wishes to take the quickest route (not necessarily the shortest) from Berlin to Paris. It be of little help if the route recommendation takes a week, a day or an hour to calculate, nor will it be helpful if the route recommendation states that the drive will take 11 hours when the actual duration is 20 hours.

Data stewardship conditions the relationship between AI capability and users’ perceived value. To increase the speed and accuracy of predictions, ML models rely heavily on the quality and quantity of the data used for AI capabilities to enhance users’ perceived value. Firms must refine and extract value from data through data stewardship, defined as the holistic, enterprise-wide management of a firm’s data assets to help ensure adequate data quantity and quality (Baesens et al., 2016; Cooper et al., 2000). If data quality is low due to inaccurate measurements or if there is not enough data, then the identified patterns will be of low quality and so will any predictions. An example is provided by IBM’s inaccurate predictions of cancer diagnosis (Khoury & Ioannidis, 2014).

The subsequent set of moderating factors pertains to user-centered design, which can serve as a vital success factor for companies adopting consumerization, i.e., the widespread adoption and diffusion of consumer digital technologies throughout society (Gregory et al., 2018). User-centered design helps companies gain a better understanding of user needs, which enhances the performance and capabilities of products and services. This approach empowers users to contribute feedback and personal data to co-create value, continuously improving and fine-tuning a platform’s models and features (Gabriel et al., 2015). Achieving this outcome is contingent on performance expectancy, which denotes the extent to which users anticipate gains in task performance by using the offering (Venkatesh et al., 2003), and effort expectancy, which refers to the degree to which a user anticipates that using the offering will not pose a burden (Venkatesh et al., 2003). As such, successfully activating DNEs hinges on minimizing effort expectancy while maximizing performance expectancy to engage users in using the offering and thus the DNEs loop.

The final set of moderating factors relates to platform legitimacy. The theory of DNEs holds that platform legitimacy is a balancing act performed by platform owners to meet the interests of diverse stakeholders by mitigating any potential risks to personal data use (Kroener & Wright, 2014) and prediction explainability (Coglianese & Lehr, 2019). With regard to personal data use, platform owners must show that the data collection and use practices are morally acceptable, whereas higher prediction explainability by the algorithms increases users’ perceived value. Interestingly, the recent debates about Facebook’s use of its ML algorithms reflects the broader legitimacy challenges (Yang & Ji, 2019).

Overall, the proposed theory of DNEs describes a new category of network effects that offer a unique account of user value creation from the use of ML technology. The theory proposes a set of factors and their relations, specifying the conditions under which DNEs generate significant perceived value for users. This theory explains, for example, why offerings based on Google’s search engine are highly popular for online search use. These offerings generate data from each usage, and those data are analysed to find patterns that enable an improvement of subsequent individual search services. This service satisfies users because the search process and outcome are perceived as accurate and fast, the performance is perceived as good and the effort needed to use it is perceived as low. It seems that its users mostly find its use legitimate, in terms of both the handling of personal data and the expandability of its algorithms.


Next, we evaluate and analyse the theory of DNEs. We consider its strengths and challenges, in light of its relatively early stage of development. This analysis allows us to propose lines of research derived from these limitations.

4.1 Strengths of the theory of DNEs

The theory of DNEs has several strengths. First, it offers a non-trivial theoretical account of real-life active data networks used by firms to contribute to their success, in some instances providing an unprecedented magnitude of value creation (Rosenblat, 2018). One example of recent success is ChatGPT. This is a chat system developed by OpenAI and based on the GPT-3 AI language model. As a language model, ChatTGPT has been trained using large amounts of textual data and has demonstrated a strong ability to generate consistent and accurate responses to a variety of natural language questions and tasks. ChatGPT’s success can be partly explained by DNEs, which are a key feature of many information technologies and artificial intelligence. DNEs refer to the idea that the more a technology or platform is used, the more valuable it becomes to users. This is because the continuous use of the platform generates an increasing amount of data, which in turn allows the algorithm to learn and improve its performance. In addition, the continuous use of the platform also attracts more users, which further increases the amount of data available and thus the quality of the model. In the case of ChatGPT, the DNEs are evident in its ability to generate accurate and consistent answers. The more ChatGPT is used, the more data is generated and the more accurate and contextualised the response becomes. Furthermore, as ChatGPT has been trained on a large amount of data, it can answer a wide range of questions and topics.

A second strength is the theory’s account of the unique characteristics of ML technologies: the ability to learn. The theory of DNEs thus opens the black box of the concept of IT in research on the use-value of IT. Thus, the theory provides a more fine-grained conceptualisation of the IT artefact beyond the black box, which can guide managerial decision making as to the design of uses of machine leaning technology. Third, the theory of DNEs seems to offer a high degree of practical usefulness in providing clear and relevant practitioner guidance for those who design the uses of ML technology by specifying what factors require specific attention to create user value. Finally, from a theoretical point of view, the theory of DNEs contributes to the advancement of the body of theory of network effects, by articulating a powerful mechanism that networked actors can activate. In this sense, the theory of DNEs shifts scholarly attention from a network’s size as a key determinant of value creation to a network’s magnitude of leaning based on user data. The more learning there is, the more value may be created. DNEs are a recent phenomenon, so the theory of DNEs (Gregory et al., 2021) does not yet have solid empirical support. We expect large studies of companies employing DNEs to confirm the underlying operating mechanism concerning the co-conditional factors of value creation enabled by the use of ML technologies.

Next, we summarise some of the criticisms of the theory of DNEs and then add our critical assessment. This review leads to a set of fundamental questions that form the basis of the expanded proposed programme of research on data networks effects.

4.2 Some challenges to the theory of DNEs

Despite its newness, the theory of DNEs has already attracted criticism. Clough and Wu (2022) argue that the theory of DNEs does not take into account the difference between (a) the centrality of the platform’s database (which is legally part of a focal firm and contains data generated from the use of an offering, good or service for use by ML algorithms), and (b) the decentralised nature of the user base that legally does not belong to the focal firm and is therefore free to participate or refrain from participating in the use of the offering. For Clough and Wu (2022), not considering such differences has consequences related to (i) the capture of value from the activated DNE, and (ii) the understanding of DNEs as a new kind of network effect.

In relation to the capture of value from the activated DNE, the decentralised nature of the user base requires special consideration on how some value can be captured by the focal firm that creates user value. The varying interests of a platform’s stakeholders (users, owners, complementors, employees, etc.) in the use of the network, together with other external factors (changes in the environment, competitors, etc.), can create tensions between the focal firm and users and can thereby challenge the focal firm’s ability to capture enough value aside from the value that is created for users. For example, platform owners may use aggressive revenue-sharing systems to improve value appropriation, which may reduce incentives for complementors to contribute to the platform and have implications for value creation (Chen et al., 2022).

In relation to the understanding of DNEs as a new kind of network effect, the user-generated database has different dynamics compared to the installed user base. The dynamics of the database derive from the fact that the database is created and evolves over time. The accuracy of the predictions generated by the ML algorithm is not the same initially with a small database of training data as a later situation with a more extensive database and with established learning patterns. According to Clough and Wu (2022), these dynamics differ from those of the user base, whose potential is linked to the size of the network at any given time. It is well established that the value created by network effects derives from the size and structure of a company’s or platform’s installed base (Afuah, 2013; Katz & Shapiro, 1985). Therefore, the very name of DNEs implies that there must be a close relationship between data-driven learning and the installed base. However, according to Clough and Wu (2022), the differences between the installed base and the corresponding database mean that the relationship between the two is not so close.

Reflecting on this criticism, Gregory et al. (2022) subsequently highlighted some nuances in a more recent study. Considering the database as company property is not necessarily accurate today as we witness a shift from the centralised use of data by a firm to the decentralised use of data enabled by data-sharing flows between multiple organisations and individuals and access to public databases, as well as greater access to numerous ML technology applications already trained with data from various sources (Gregory et al., 2022). As we understand it, this observation dismantles the main argument of Clough and Wu (2022) about the necessary distinction between a centralised database and a decentralised user base.

As for the criticism about the lack of specification of value capture by an activated DNE, Gregory et al. (2022) explained that both value creation and value capture processes are part of a single, dual, co-constitutive process within activated DNEs. However, our position is that value creation and value capture should be understood as two distinct notions, even though they are sometimes closely interrelated in practice, such as when value is co-created (Vargo & Lusch, 2008). This idea is illustrated by the example of two very similar social media platforms, namely Facebook and Google Plus. Despite the technologies used by both, the fact that the value product they offer is practically the same and the fact that both have large amounts of data, Facebook has been successful, whereas Google Plus has not.

Casual empirical experience informs us that DNEs can be a key factor in a company’s value capture. For example, Facebook and Instagram create user value by connecting people and capture value through personalised advertising. However, everyday experience also informs us of cases that have failed, such as Google Plus mentioned above. It is important to note that value creation can also result in its destruction (Rai & Tang, 2014; Lepak et al., 2007; Canhoto & Clear, 2020). Value destruction occurs when users perceive a decline in the usefulness of a product or service (Canhoto & Clear, 2020). Therefore, companies must assess the potential costs of utilizing ML technologies, including the risk of reputational damage due to ethical issues associated with algorithmic decisions in self-driving cars (Goodall, 2016), balancing calculation speed with trust issues (Cormen et al., 2022), and reconciling accuracy with algorithm inoperability (Lee & Shin, 2020). This situation underscores that value creation and capture are not always a sufficient and binary process. Whilst the literature frequently depicts value creation and value capture as a dual and co-constitutive process (Vargo & Lusch, 2008; Gregory et al., 2022), a focal firm that instigates a DNE for value creation may not always be able to appropriate enough value, or even any value, to provide a return on investment. If a platform that offers free content such as YouTube wants to capture more value and bombard its users with too much advertising, users would leave the network, and the platform would lose the opportunity to capture the value created. Each stakeholder (owners, users, etc.) has its subjective perception of what is valuable and has different purposes regarding the use of the network. Therefore, some tension arises in the creation and capture of value because all stakeholders try to appropriate as much value as possible. The key here is not to choose between creating and capturing value but to strike a balance between the two to achieve the desired return of investment for the focal firm and thus achieve a competitive advantage. Therefore, value creation and value capture are two distinct yet typically interrelated processes, where each ought to be also given individual consideration. In its current form, the theory of DNEs neither differentiates between the two nor focuses on factors that moderate value capture for the focal firm through DNEs. Therefore, a core research question is: “What factors co-condition value capture from the use of ML technology?”

A second critical point here focuses on the data gathered and analysed with ML technology to establish the DNE loop. We recall that user value creation is moderated by the quantity and quality of the data being analysed. The theory does not consider other factors related to the data and its dynamics such as network size, expiration, source of origin and combination that could affect the creation and capture of value through DNEs. One issue is the role of database size and whether a specific data threshold (i.e. a boundary line on database size) affects or even determines the activation of DNEs. The literature explains that there are at least three data threshold possibilities: critical mass (minimum amount of data needed to activate network effects; Afuah & Tucci, 2003); maximum data threshold (maximum amount of data beyond which more data may not lead to more learning; Cennamo, 2020); and network saturation (the suitability of a given size of a network where a new offer has higher visibility and better adaptation to each user; Cennamo, 2020). All three thresholds are related to the amount of data needed to activate network effects and thus to co-determine their ability to capture the value that is created. Therefore, a core question is: “What are the dynamics of the database that activate and sustain DNEs?”

A third critical point here focuses on the notion of a user that perceives value from the service being provided by an activated DNE. Currently, the theory conceives the users as a single actor (or stakeholder), understood as a distinct group of individuals that use the service and have similar user profiles, needs, interests and legitimacy profiles. This conception implies that the theory does not consider the possibility that there are several user groups, or actors, with functionally disparate ways of perceiving and legitimising the use of the network. The use by each actor may lead to a situation where the provided service or product differs. Therefore, the value that each actor perceives may differ too. One example is an app on a health issue or a specific pathology (Costa-Climent & Haftor, 2021). The user network may consist of patients who enter their data and receive personalised information on treatment, suitable products and advice. Additionally, it may also include physicians who use the app to monitor patients and avoid transfers and queues. Meanwhile, pharmaceutical companies use the app to disseminate their products and receive customer reviews. All actors may benefit from activated DNEs that use the same database. Each user and user group uses the app differently and for different purposes. Therefore, the value that different users perceive may differ, as may the way they appropriate this value. Likewise, the value that legitimises the use of the app is different for each actor depending on their reason for using it. For patients, it may be sufficient if the app offers benefits in terms of reducing the effects of their illness. For doctors, the app can help them ensure the necessary medical rigour. These different forms of legitimation of the product or service and the different uses of each actor ought to be considered when designing mechanisms for creating and capturing value, but they are currently ignored by the theory. Therefore, a core question is: “What are the dynamics of multiple user actors in activating DNEs?”

A fourth critical point here focuses on the theory’s static view of data network activation, which disregards the role of temporal changes of a data network loop and its context. The implementation of most kinds of non-trivial IT in a business model is not immediately reflected in the company’s productivity (Brynjolfsson et al., 2021). For example, there is a period of latency until the value provided from the use of new IT is no longer intangible and can be measured as tangible capital (Brynjolfsson et al., 2021b). The use of a new technology such as ML requires the acquisition of the necessary technology and the training of workers, as well as the adaptation of the workforce and other organisational strategies. Furthermore, the DNE activation process is not immediate and requires a series of steps: launching the offer, initiating data collection, learning with training data, achieving sufficient learning to improve products successfully, observing user reactions and materialising the network effect to increase value. This dynamic process may involve the design of an evolving business model that adapts its value creation and capture mechanisms according to the phase it is in or the need to change at any given time. Therefore, a core question is: “What are the temporal dynamics in activating and sustaining DNEs?”


To address the research questions posed in this research programme, we plan to adopt a qualitative research approach within the interpretative paradigm. This approach will allow us to understand the social reality that precedes theoretical development. It will guide both the formulation of hypotheses and the application of the methodology and scientific research tools.

Given the non-existence of a body of theory on how the use of ML creates and captures value for companies, this research proposal is based on an abductive approach (Tavory & Timmermans, 2014). When existing theories cannot account for a given phenomenon (in this case, the creation and capture of value through the use of ML), a deductive approach may limit the research by imposing predefined categories and relationships. In contrast, a purely inductive exploration mode ignores prior knowledge that can help guide the research (Behfar & Okhuysen, 2018). An abductive approach, which offers a combination of induction and deduction, aims to overcome these weaknesses. It does so by allowing the discovery of new features and relationships whilst linking the findings to some pre-existing body of theory (Tavory & Timmermans, 2014). To this end, we propose an exploratory case study methodology for theory building (Eisenhardt & Graeber, 2007; Eisenhardt, Graeber, & Sonenhein, 2016). This approach is particularly suitable for the present research question.

Exploratory case studies

In a first stage, it has been necessary to develop additional constructs and theoretical relationships that fit with the constructs that exist in the theory of DNEs. This allows the formulation of measures for the constructs, which will allow the theory to be tested.

This can be done through empirical case study research, as this methodology allows for the elaboration of theoretical constructs and their measurements and the relationships between them.

In this paper, we rely on Eisenhardt’s methodology when using a research strategy related to theory building based on case studies (Eisenhardt & Graebner, 2007). Theory building from cases is research that involves the use of one or more cases to create theoretical constructs, propositions and/or theory from empirical evidence based on cases (Eisenhardt, 1989). In doing so, we aim to test the theory of DNEs and advance its empirical construction.

Therefore, in this first phase of the empirical research, we will implement exploratory case studies of one or more companies with relevant experiences in the use of data through ML and their management of learning and predictive capabilities to create value. This will allow us to identify dependent variables (such as value creation and capture) and drivers (amount of data, ML capabilities, innovation processes, etc.) in the case studies. Thus, we are able to develop patterns in the business models related to the key factors of the network effects theory of data. Through this method, we can formulate new constructs and their measurements and the relationships between the new constructs to situate them in the theoretical body of DNEs. The theory of DNEs, within which the elaborated constructs have been related, will then be tested

Multi-case study

The empirical test of the theory can be carried out using different methodologies. This research proposes a qualitative methodological approach such as the Qualitative Comparative Analysis Methodology (QCA) (Rihoux & Ragin, 2008; Ragin, 2014).

The QCA method is considered adequate because it allows identifying which factors (or independent variables) are associated with the presence of a given result (Ragin, 2014). Identify the necessary or sufficient conditions that occur in relation to the causality of the event studied.

The reason is that the proposed DNEs theory is built through a set of factors that can interact with each other, allowing a series of configurations of these factors (prediction speed, ability to explain predictions, and the rest). The QCA methodology does not require a large sample, so it offers an excellent approach to perform the one of the first empirical test of the theory.

In this study, the selection criteria of the cases is not random and does not seek representativeness but is based on purposive sampling based on relevance to the research question (Eisenhardt, 2021). The sample selection criteria are related to the use of ML technologies by the firm in creating and delivering value to the end-user or customer, the size of the database, the process of implementing ML technology, among others.

The data to be collected from the cases will come from different sources, such as interviews, archival data, survey data, ethnographies, and observations (Eisenhardt & Graebner, 2007). These cases will be analyzed through a qualitative data methodology such as QCA (Rihoux & Ragin, 2008; Ragin, 2014), which identifies necessary or sufficient conditions that explain the causal facts studied.


These critical issues, each captured by a subquestion developing the central research question, constitute the proposed programme for research on creating and capturing value through DNEs. The research programme’s objective is to advance our theoretical understanding of the structure and dynamics of successful DNEs. The aim is to provide managerial guidance for the activation and continuance of such effects. The research questions posed in this paper are interrelated and thus contribute to an enhanced understanding of DNEs, value creation, and capture.

Whilst each of these subquestions targets a key knowledge gap, we also recognise that there are relations between these gaps that are important to address because they may provide new theoretical knowledge and underpinnings for the conceptualisation of DNEs and for guiding empirical research in the field of value creation and capture through ML technology.

Briefly, although distinct, value creation and value capture are typically highly interrelated, particularly in the context of ML technology use. The question is whether any interdependency between the kind of value that is created conditions value capture. Second, the dynamics of the database that activate and sustain DNEs co-constitute the domain of temporal dynamics of DNEs, both of which co-condition value creation and value capture. Additionally, a DNE loop may be started with one main user actor and then evolve to several user actors, as is the case with Facebook, which relates to the temporal dynamics of the DNE. Additional relations may be spotted and articulated, which is part of the proposed research program.

IS is a discipline whose main objective is to study the applications of technology by organisations and society (Avison & Elliot, 2006). It is multidisciplinary and has a long history of incorporating theories from other disciplines. Following this tradition, this research programme is based on a multi-theoretical research approach to study the factors that co-condition the creation and capture of value from the use of ML technologies. To address the main research question, we must investigate how companies generate and acquire value, considering all stakeholders within and beyond the organization. In doing so, we draw upon the theories of business models for creating and capturing value within companies and DNEs. Business model theory encompasses the content, structure, and governance of transactions intended to generate value by exploiting business opportunities (Amit & Zott, 2001). This theory elucidates how an organization engages with external and internal stakeholders and participates in economic transactions to create value for all exchange stakeholders, consistent with our multi-stakeholder user perspective. Drawing on business model theory will help us identify the dimensions of the ML-based business model under study (its offerings, the system of activities, the actors involved, the transaction system and the governance of all of them) and understand how they are orchestrated to create and capture value.

We consider business model theory appropriate because it also demonstrates the capacity of IT to create value by activating the four business model themes that trigger value creation: novelty, efficiency, complementarity, and lock-in (Amit & Zott, 2001; Zott & Amit, 2008). Although the ML technologies that form the basis of our research are advanced and differ from those described in business model theory, the underlying theoretical construction regarding how novelty, efficiency, complementarity and lock-in can explain value creation through the use of IT is, in our view, equally applicable to the study of more advanced technologies such as ML. This theoretical lens will allow us to identify the activation of these value-creation factors in certain cases. We can also explore whether some of them are activated individually or in combination and whether they remain stable or evolve over time, thus responding to our proposal for a dynamic perspective of DNEs. Therefore, this theoretical construction can enlighten us on the role of business model themes in creating and capturing value through ML technologies.


The increasing use of ML technology has made the use and exploitation of data indispensable in many sectors. An increasing number of organisations invest substantial resources in developing and using ML technology to provide users with novel value. Despite such investments, some organisations fail whilst others succeed. This circumstance implies that there are factors that jointly contribute to the generation and acquisition of value through ML technology and warrant further examination. The newly proposed theory of DNEs is instrumental in identifying the factors that influence the creation and capture of value resulting from the application of ML technology. This theory accounts for the unique characteristics of ML technology, namely its data-driven learning, its use to generate predictions and decisions and its framing within a social and psychological context. This paper unearths some of the merits, strengths and limitations of the recently proposed theory of DNEs. These observations are translated into a set of research questions that constitute their proposed research programme on value creation and capture from the use of ML technology. We propose the study of the questions raised in this research programme from a multi-theoretical approach. Such an approach enables identification of the factors that co-condition the success of the use of ML technologies by companies.

The general contributions are as follows. For theory, this research will be an essential contribution to the theoretical body related to the value of the use of ML technologies. The main contribution of this research is that it will be one of the first empirical test of the theory of DNEs, which can benefit from four developments. First, the research will expand the existing theory by considering factors that affect value capture through DNEs. Until now, DNEs theory focuses on how the use of data-driven learning technologies creates value for the user. Our contribution involves considering in a differentiated way how different actors can capture the value created, or part of it, through the use of ML technologies. This incorporates into the theory the need to consciously align the processes of creation and capture of value to achieve the balance between both approaches and thereby improve the firm’s competitive advantage.

Secondly, with this research the theory will benefit from extending the notion of a single human user of the offer generated with the help of machine learning techniques, to the notion of multiple actors, which can be individuals or organizations. Furthermore, the findings of this research will provide new factors that will complement the existing moderators of the theory of DNEs related to data management and platform legitimacy. We broaden the consideration of data quality and quantity as the only factors related to data management that condition the creation of value through the use of DNE, identifying other conditioning factors such as the role played by database size thresholds in the creation of value through the use of ML technologies.

Finally, this research takes a dynamic view of the use of ML technologies. This means moving away from the current static view of the theory of DNEs since the value created by the use of ML can be the result of the coevolutionary development of events, which accounts for the temporality and dynamics of the effects of the data network.

To practice, the contributions from this work will provide valuable knowledge for business practice by assisting firm managers in implementing the use of ML technologies. This means decreasing the risk of using ML due to the significant investment involved in its implementation in the company. Therefore, our results will have clear implications for company data management, business strategy and business model design, and the development of regulatory policies for data management.


Afuah, A. (2013). Are network effects really all about size? The role of structure and conduct. Strategic Management Journal, 34(3), 257–273.

Afuah, A. & Tucci, C. L. (2003). Internet business models and strategies: Text and cases (Vol. 2). McGraw-Hill.

Agrawal, A., Gans, J., & Goldfarb, A. (2018). Prediction Machines: The Simple Economics of Artificial Intelligence. Harvard Business Press.

Al Dakhil, S. & Bayoumi, S. (2020). Reviews Analysis of Apple Store Applications Using Supervised Machine Learning. In R. Agrawal, M. Paprzycki & N. Gupta (eds.), Big Data, IoT, and Machine Learning (pp. 51–78). CRC Press.

Amit, R. & Zott, C. (2001). Value creation in e-business. Strategic Management Journal, 22(6-7), 493–520.

Arel, I., Rose, D. C., & Karnowski, T. P. (2010). Deep machine learning-a new frontier in artificial intelligence research [research frontier]. IEEE Computational Intelligence Magazine, 5(4), 13–18.

Avison, D. & Elliot, S. (2006). Scoping the Discipline of Information Systems. In J.L. King (ed.), Information Systems: The State of the Field (pp. 3–18). Wiley.

Baesens, B., Bapna, R., Marsden, J. R., Vanthienen, J., & Zhao, J. L. (2016). Transformational Issues of Big Data and Analytics in Networked Business. MIS Quarterly, 40(4), 807–818.

Behfar, K. & Okhuysen, G. A. (2018). Perspective—Discovery within validation logic: Deliberately surfacing, complementing, and substituting abductive reasoning in hypothetico-deductive inquiry. Organization Science, 29(2), 323–340.

Bharadwaj, A. S. (2000). A resource-based perspective on information technology capability and firm performance: an empirical investigation. MIS Quarterly, 24(1), 169–196.

Brynjolfsson, E. & Hitt, L. (1996). Paradox lost? Firm-level evidence on the returns to information systems spending. Management Science, 42(4), 541–558.

Brynjolfsson, E., Hitt, L. M., & Yang, S. (2002). Intangible assets: Computers and organizational capital. Brookings Papers on Economic Activity, 2002(1), 137–181.

Brynjolfsson, E., Jin, W., & McElheran, K. (2021a). The power of prediction: predictive analytics, workplace complements, and business performance. Business Economics, 56, 217–239.

Brynjolfsson, E., Rock, D., & Syverson, C. (2021b). The productivity J-curve: How intangibles complement general purpose technologies. American Economic Journal: Macroeconomics, 13(1), 333–72.

Bughin, J., Hazan, E., Ramaswamy, S., Chui, M., Allas, T., Dahlström, P., & Trench, M. (2017). Artificial intelligence: the next digital frontier? [discussion paper]. McKinsey Global Institute.

Canhoto, A. I. & Clear, F. (2020). Artificial intelligence and ML as business tools: A framework for diagnosing value destruction potential. Business Horizons, 63(2), 183–193.

Cennamo, C. (2020). Value Preserving Platform Regulation: Network Effects, Platform Value and Regulatory Remedies. Platform Value and Regulatory Remedies.

Chae, H. C., Koh, C. E., & Prybutok, V. R. (2014). Information technology capability and firm performance: contradictory findings and their possible causes. MIS Quarterly, 38(1), 305–326.

Chen, L., Tong, T. W., Tang, S., & Han, N. (2022). Governance and design of digital platforms: A review and future research directions on a meta-organization. Journal of Management, 48(1), 147–184.

Chesbrough, H. (2007). Business model innovation: it’s not just about technology anymore. Strategy & Leadership, 35(6), 12–17.

Chesbrough, H. & Rosenbloom, R. S. (2002). The role of the business model in capturing value from innovation: evidence from Xerox Corporation’s technology spin-off companies. Industrial and Corporate Change, 11(3), 529–555.

Chhillar, D. & Aguilera, R. V. (2022). An Eye for Artificial Intelligence: Insights Into the Governance of Artificial Intelligence and Vision for Future Research. Business & Society, 61(5), 1197–1241.

Churchman, C. W. (1961). Realism in management science: A report. Management Science, (3), 63–81.

Clough, D. R. & Wu, A. (2022). Artificial Intelligence, Data-Driven Learning, and the Decentralized Structure of Platform Ecosystems. Academy of Management Review, 47(1), 184–189.

Coglianese, C. & Lehr, D. (2019). Transparency and Algorithmic Governance. Administrative Law Review, 71(1), 1–56.

Cooper, B. L., Watson, H. J., Wixom, B. H., & Goodhue, D. L. (2000). Data Warehousing Supports Corporate Strategy at First American Corporation. MIS Quarterly, 24(4), 547–567.

Cormen, T. H., Leiserson, C. E., Rivest, R. L., & Stein, C. (2022). Introduction to algorithms. MIT press.

Costa-Climent, R. & Haftor, D. M. (2021). Business model theory-based prediction of digital technology use: An empirical assessment. Technological Forecasting and Social Change, 173, 121174.

Dąbrowska, J., Almpanopoulou, A., Brem, A., Chesbrough, H., Cucino, V., Di Minin, A., … & Ritala, P. (2022). Digital transformation, for better or worse: a critical multi-level research agenda. R&D Management, 52(5), 930–954.

Davidovski, V. (2018). Exponential innovation through digital transformation. In Proceedings of the 3rd International Conference on Applications in Information Technology (pp. 3–5). New York.

Deichmann, U., Goyal, A., & Mishra, D. (2016). Will digital technologies transform agriculture in developing countries? Agricultural Economics, 47(S1): 21–33.

Devaraj, S. & Kohli, R. (2003). Performance impacts of information technology: Is actual usage the missing link? Management Science, 49(3), 273–289.

Dewan, S. & Kraemer, K. L. (2000). Information technology and productivity: evidence from country-level data. Management Science, 46(4), 548–562.

Dubosson-Torbay, M., Osterwalder, A., & Pigneur, Y. (2002). E-business model design, classification, and measurements. Thunderbird International Business Review, 44(1), 5–23.

Economides, N. (1996). The economics of networks. International Journal of Industrial Organization, 14(6), 673–699.

Eisenhardt, K. M. (1989). Building theories from case study research. Academy of Management Review, 14(4), 532–550.

Eisenhardt, K. M. & Graebner, M. E. (2007). Theory building from cases: Opportunities and challenges. Academy of Management Journal, 50(1), 25–32.

Eisenhardt, K. M., Graebner, M. E., & Sonenshein, S. (2016). Grand challenges and inductive methods: Rigor without rigor mortis. Academy of Management Journal, 59(4), 1113–1123.

Eisenhardt, K. M. (2021). What is the Eisenhardt Method, really?. Strategic Organization, 19(1), 147–160.

Enholm, I. M., Papagiannidis, E., Mikalef, P., & Krogstie, J. (2022). Artificial intelligence and business value: A literature review. Information Systems Frontiers, 24(5), 1709–1734.

Farrell, J. & Saloner, G. (1986). Installed base and compatibility: Innovation, product preannouncements, and predation. The American Economic Review, 76(5), 940–955.

Gabriel, Y., Korczynski, M., & Rieder, K. (2015). Organizations and their Consumers: Bridging Work and Consumption. Organization, 22(5), 629–643.

Goodall, N. J. (2016). Away from trolley problems and toward risk management. Applied Artificial Intelligence, 30(8), 810–821.

Gregory, R. W., Henfridsson, O., Kaganer, E., & Kyriakou, H. (2021). The role of artificial intelligence and data network effects for creating user value. Academy of Management Review, 46(3), 534–551.

Gregory, R. W., Henfridsson, O., Kaganer, E., & Kyriakou, H. (2022). Data network effects: Key conditions, shared data, and the data value duality. Academy of Management Review, 47(1), 189–192.

Gregory, R. W., Kaganer, E., Henfridsson, O., & Ruch, T. J. (2018). IT Consumerization and the Transformation of IT Governance. MIS Quarterly, 42(4), 1225–1253.

Hitt, L. M., & Brynjolfsson, E. (1996). Productivity, business profitability, and consumer surplus: Three different measures of information technology value. MIS quarterly, 20(2), 121–142.

Joshi, M. P., Su, N., Austin, R. D., & Sundaram, A. K. (2021). Why So Many Data Science Projects Fail to Deliver. MIT Sloan Management Review, 62(3).

Katz, M. L. & Shapiro, C. (1992). Product introduction with network externalities. The Journal of Industrial Economics, 40(1), 55–83.

Katz, M. L. & Shapiro, C. (1985). Network externalities, competition, and compatibility. The American Economic Review, 75(3), 424–440.

Khoury, M. J. & Ioannidis, J. P. A. 2014. Big Data Meets Public Health. Science, 346(6213), 1054–1055.

Kohli, R. & Devaraj, S. (2003). Measuring information technology payoff: A meta-analysis of structural variables in firm-level empirical research. Information Systems Research, 14(2), 127–145.

Kohli, R., & Grover, V. (2008). Business value of IT: An essay on expanding research directions to keep up with the times. Journal of the Association for Information Systems, 9(1), 1.

Kroener, I. & Wright, D. (2014). A Strategy for Operationalizing Privacy by Design. The Information Society, 30(5), 355–365.

Lee, I. & Shin, Y. J. (2020). Machine learning for enterprises: Applications, algorithm selection, and challenges. Business Horizons, 63(2), 157–170.

Lee, J., Suh, T., Roy, D., & Baucus, M. (2019). Emerging technology and business model innovation: the case of artificial intelligence. Journal of Open Innovation: Technology, Market, and Complexity, 5(3), 44.

Lepak, D. P., Smith, K. G., & Taylor, M. S. (2007). Value creation and value capture: A multilevel perspective. Academy of management review, 32(1), 180–194

Liebowitz, S. J. & Margolis, S. E. (1994). Network externality: An uncommon tragedy. Journal of Economic Perspectives, 8(2), 133–150.

Meinhart, W. A. (1966). Artificial Intelligence, Computer Simulation of Human Cognitive and Social Processes, and Management Thought. The Academy of Management Journal, 9(4), 294–307.

Melville, N., Kraemer, K., & Gurbaxani, V. (2004). Information technology and organizational performance: An integrative model of IT business value. MIS Quarterly, 28(2), 283–322.

Merhi, M. I. (2023). An evaluation of the critical success factors impacting artificial intelligence implementation. International Journal of Information Management, 69, 102545.

Pandey, A. & Mishra, S. (2021). Does the Executive Perception of the Value of Information Technology (IT) Influence the IT Strategy? A Case Study. Journal Of Information Systems Applied Research, 14(1), 24–35.

Papagiannidis, E., Enholm, I. M., Dremel, C., Mikalef, P., & Krogstie, J. (2023). Toward AI governance: Identifying best practices and potential barriers and outcomes. Information Systems Frontiers, 25(1), 123–141.

Porter, M. E. (2001). The value chain and competitive advantage. Understanding Business Processes, 2, 50–66.

Ragin, C. C. (2014). The comparative method: Moving beyond qualitative and quantitative strategies. Univ of California Press.

Rai, A. & Tang, X. (2014). Research commentary—information technology-enabled business models: A conceptual framework and a coevolution perspective for future research. Information Systems Research, 25(1), 1–14.

Rihoux, B. & Ragin, C. C. (2008). Configurational comparative methods: Qualitative comparative analysis (QCA) and related techniques. Sage Publications.

Rosenblat, A. (2018). Uberland: How algorithms are rewriting the rules of work. Univ of California Press.

Shaw, J., Rudzicz, F., Jamieson, T., & Goldfarb, A. (2019). Artificial intelligence and the implementation challenge. Journal of Medical Internet Research, 21(7), e13659.

Sjödin, D., Parida, V., Jovanovic, M., & Visnjic, I. (2020). Value creation and value capture alignment in business model innovation: A process view on outcome-based business models. Journal of Product Innovation Management, 37(2), 158–183.

Tallon, P. P., Kraemer, K. L., & Gurbaxani, V. (2000). Executives’ perceptions of the business value of information technology: a process-oriented approach. Journal of Management Information Systems, 16(4), 145–173.

Tavory, I. & Timmermans, S. (2014). Abductive analysis: Theorizing qualitative research. University of Chicago Press.

Vargo, S. L. & Lusch, R. F. (2008). Service-dominant logic: continuing the evolution. Journal of the Academy of Marketing Science, 36(1), 1–10.

Venkatesh, V., Morris, M. G., Davis, G. B., & Davis, F. D. (2003). User Acceptance of Information Technology: Toward a Unified View. MIS Quarterly, 27(3), 425–478.

Windsor, D. (2017). Corporate citizenship: Evolution and interpretation. In J. Andriof & M. McIntosh (eds.), Perspectives on corporate citizenship (pp. 39–52). Routledge.

Yang, A. & Ji, Y. G. (2019). The quest for legitimacy and the communication of strategic cross-sectoral partnership on Facebook: A big data study. Public Relations Review, 45(5), 101839.

Zott, C. & Amit, R. (2008). The fit between product market strategy and business model: Implications for firm performance. Strategic Management Journal, 29(1), 1–26.


1 Doctoral Researcher at Uppsala University, Box 513, SE-75120, Uppsala, Sweden and Doctoral Candidate at the MIT Research School