Value creation and value capture in AI offerings

(1)

Value creation and value capture in AI offerings

A process framework on business model development

Josef Åström

Industrial and Management Engineering, master's level 2020

Luleå University of Technology

Department of Business Administration, Technology and Social Sciences

(2)

Abstract

Purpose – The purpose of this study is to explore how AI providers ensure value creation and value capture dimensions when developing AI offerings. This by illustrating activities that builds the dimensions and to increase the understanding on how value creation and value capture interplay.

Method – To fulfill its purpose, this study adopted an inductive exploratory single case-study approach, conducted at a market leading telecom provider of AI related services. In total, 23 interviews were held with the case company, and the results were generated by applying a three step process of thematic analysis, where the framework’s phases and its underlying activities were identified.

Findings – This study’s findings are presented in a process framework, explicitly illustrating key activities for the value creation and value capture dimensions. The framework further suggests AI providers to design AI offerings by going through three phases, i.e. identifying prerequisites for value creation, connecting value creation with value capture opportunities and designing the value offering. It is also found that AI providers must develop multiple business models, and operate them simultaneously, to operationalize AI successfully.

Theoretical and practical implications – This study contributes to the theoretical understanding of AI by identifying activities building the value creation and value capture dimensions. In addition, the process framework can be used by practitioners when developing or refining business model architectures for AI offerings.

Limitations and future research – This study is limited by its scope, and future research is rec- ommended to perform extended studies where both providers and its customers are included. In addition, this study’s findings highlights the importance of developing and operating multiple business model to operationalize AI successfully. However, this also induces risks of business model cannibalization, which calls for more research.

Keywords: Artificial Intelligence; Business models; Value creation; Value capture; Telecom.

(3)

Acknowledgements

This master thesis project was my final course of the Industrial Engineering and Management program with specialization in Strategic Business Development at Luleå University of Technology (LTU). The project took place during the spring of 2020 in collaboration with a telecom provider in Sweden.

I want to thank all who contributed with valuable insights and support during my project, both from LTU and the case company. In particular, this project would not have been possible without all the informants at the case company, providing me with interesting information and giving me the opportunity to learn about AI. However, I want to direct a special thank you to a few people, extra important for me and my study. First, I want to express my gratitude to my supervisor at LTU, Wiebke Reim, for your valuable guidance from start to end and for challenging me to perform at my very best. Further, I want to direct special gratitude towards my supervisor at the case company, Mats Pettersson, for providing me with the best possible prerequisites to perform, and for your continuous support throughout the project. Lastly, I want to thank my fellow students in the opponent group for providing me with valuable feedback on how to improve my study.

2020-06-11

—————————- Josef Åström

(4)

1 INTRODUCTION

Artificial Intelligence (AI) is often referred to as a technology that facilitates creation of additional value for customers. This originates from advances in AI techniques that offer opportunities to mimic our cognitive behavior and automate processes of identifying and solving complex problems (Lee et al., 2019; Zhuang et al., 2017). AI’s characteristics as a capital–labor hybrid brings opportunities to aug- ment human labor at scale and speed, self-learn and continuously improve over time, according to Plastino and Purdy (2018), which brings seemingly enormous potential to create competitive solutions with support from AI. According to Lee et al. (2019), AI offer opportunities to improve operational efficiency and accelerate innovation by providing insights from large data sets and predicting unexpected events, which entails interest across many diverse industries.

The telecom industry is one of the most data intensive industries and telecom actors can be considered as enablers for global business ecosystems through provisioning of fast, secure and high quality communication services to billions of users, including enterprises and end-consumers (Lungu, 2018).

Proceedings by Elayoubi et al. (2017), highlights that the industry is characterized by the continuous increase of accessible data, more complex systems and demand of higher service performance levels, which requires integration of intelligent systems to manage complexity and significantly improve customer experiences. Telecom providers, focusing on network rollout services and implementation of technological innovation, are therefore driving the development of AI and how it can be utilizied to improve network performance by embracing the role as AI providers. Thus, AI providers in the telecom industry holds a substantial impact on all firms and end-users that are consuming communication services, which ultimately facilitates global expansion and substantial business opportunities for all telecom users (Lungu, 2018). The disruptive potential of AI, opportunities to improve operational efficiency and accelerate innovation, is therefore highly relevant for the telecom industry. However, Brock and Von Wangenheim (2019) highlights that successful integration of AI applications is difficult to achieve as it requires development of certain capabilities and resources, which requires heavy investments and long development cycles. This enhances opportunities for AI providers to specialize in the subject and offer AI solutions as a service to telecom operators.

The rise of AI brings opportunities for providers to create additional value by applying a proactive approach, manage uncertainty and thus also improve cost efficiency and increase revenue (Cockburn et al., 2018). However, to capitalize on the technology it is paramount to understand how the technology can be commercialized through a suitable business model. According to Valter et al. (2018), previous research fall short in the understanding on how firms successfully operationalize AI solutions through

(7)

their business models, especially with regard to technological advancements in the AI sphere. From a practical point view, this is supported in the Artificial Intelligence Global Executive Study and Research Report, by Ransbotham et al. (2019), as 40% of their respondents revealed that significant investments did not result in the sought-for business gains. To understand this, business models can be described as "a mediating construct between technology and economic value" (Chesbrough and Rosenbloom, 2002), and can be divided into two important functions: value creation mechanisms and value capture mechanisms (Chesbrough, 2007). In other words, the business model describes a set of activities that aims to satisfy the final customer, and how the operating firm captures economic return from the defined activities. Thus, when emerging technologies are introduced, business model concepts often need to be transformed in order to fully capitalize on disruptive technologies since it is equally important to innovate the business model as it is to build advanced, technical solutions (Chesbrough, 2007). With this said, recent technical progress within the AI field, and the lacking understanding on how the technical progress can be transformed into business gains, requires more research in order to enable successful application of the technology. This is supported by Lee et al. (2019), who mentions that further research is needed in order to understand how AI solutions can be commercialized through different business model archetypes. In addition, Nylund et al. (2020) declares the need for future research regarding how value can be captured when commercializing technological innovation in the context of industry disruption.

With regard to the identified literature gaps and practical challenges, the purpose of this study is to explore how AI providers ensure value creation and value capture when developing AI offerings.

More specifically, the framework will illustrate activities that builds the dimensions and increase the understanding on how value creation and value capture interplay. The purpose will be answered through a single-case study conducted at a market leading AI provider within the telecom industry. From a theoretical point of view, this study will extend the understanding of AI and how value creation and value capture mechanisms relate to each other. Additionally, it will have practical implications as the framework will provide further insights for AI providers that are aiming to commercialize AI techniques through suitable business models. The presented framework, and its activities, will further increase the understanding on how AI providers realize AI related opportunities.

(8)

2 THEORETICAL BACKGROUND

The theoretical background will elaborate on the concepts of Artificial Intelligence and Business Models, and describe the progress within each research field respectively. First, the theoretical background reviews the research field of AI, defines the concept of AI and its current potential of impact.

Second, it reviews the research field of business models through its two elementary functions: value creation mechanisms and value capture mechanisms (Chesbrough, 2007). The theoretical review on value creation mechanisms mainly center on how AI facilitates value creation and, further, the theoretical review on value capture mechanisms center on contracts and pricing models.

2.1 Artificial Intelligence

This section discuss different concepts of AI, where technical progress within the field is described in order to define AI and its potential impact. Further, it center on AI as a General-Purpose Technology and its implications on industry structure. Thereby, this section will lay the foundation for identifying value creation and value capture activities needed to successfully implement AI.

2.1.1 Definition and AI maturity levels

Fascination and interest in AI, both publicly and in academia, has caused much discussion and thus also confusion regarding its precise definition. In research, AI is often refereed to as machines ability to mimic human-like behavior through processes such as learning, reasoning and self-correction (Kok et al., 2009; Lee et al., 2019; Zhuang et al., 2017). Further, AI can be explained through four requirements – i.e. natural language processing, knowledge representation, automated reasoning and machine learning (Kok et al., 2009). Respectively, this implies requirements of abilities to communicate in natural language; abilities to have and store knowledge; abilities to perform reasoning; and abilities to learn from its environment. However, discussing AI in terms of current abilities to fundamentally transform businesses, this definition should be carefully used as AI can not be considered as a technology that is mature enough to replace all aspects of human cognition (Agrawal et al., 2019;

Kakatkar et al., 2019). Rather, rapid progress within the area of machine learning algorithms enables substitutions of some aspects of the human cognition, such as prediction (Agrawal et al., 2019; Mishra and Silakari, 2012). Thereby, machine learning constitutes the sub-field within AI that is making the most impact across industries (Shaw et al., 2019). This enables firms to utilize the massive increase of data, abilities to store large data sets and improved computer speed, in order to build AI techniques to foresee unexpected events or probabilities (Agrawal et al., 2019). Machine learning algorithms constitutes valuable tools for optimization of sub-tasks, which enables optimization and automation of

(9)

decentralized processes (Shaw et al., 2019; Arel et al., 2010). Hence, machine learning techniques are valuable tools in complex natures as such algorithms can be fed with data and perform predictions with an accuracy that out-competes human cognition. Machine learning can thus be understood as a tool used to fill in missing information in order to improve prediction outcomes. Regardless of activity and its context, if enough data is accessible, machine learning will probably improve the operation’s output in comparison to the existing techniques (Agrawal et al., 2019).

However, machine learning should not be expressed synonymously with AI, even though the rapid progress within the research field of machine learning techniques has brought most attention with regard to its benefits. Nevertheless, machine learning is more focused on improvement of certain sub- tasks, where AI has a far more extensive domain of application. As mentioned by LeCun et al. (2015), further progress within the field of AI will emerge from systems that combines learning algorithms with complex reasoning, meaning that the system’s collected knowledge can be applied in different settings. This implies a more universal area of application, rather than just sub-task optimization.

Simple reasoning and machine learning has been combined for a long period of time (Bottou, 2014), however more research is needed in order to reach a more extensive application level of AI and thus also substitute more aspects of human cognition (LeCun et al., 2015). This means that the practical meaning of AI, and its impact, can be understood to which level reasoning and machine learning techniques are combined. Kaplan and Haenlein (2019) communicated this through three maturity levels of AI, where each maturity level’s potential business impact is explained. The first maturity level solely includes machine learning and its operational impact is therefore limited to specific application areas.

Learning, gained from a specific setting, can not be applied in new areas and dramatically changes of input parameters will decrease the algorithms functionality. The second maturity level extends the operational impact to include several areas of application, meaning that learning can be applied to new, unexplored settings without human intervention. This is therefore where simple reasoning comes in, which allows an expanded usage area of powerful machine learning algorithms. Lastly, the third level of maturity includes a fully self-conscious system with the ability to interconnect its creativity and general wisdom with operational problems in any area. In addition, the third level maturity level of AI entails abilities to outperform humans in any area, i.e. out-compete all aspects of human cognition, which will consequently make humans redundant (Kaplan and Haenlein, 2019).

This understanding of AI will be applied to evaluate how far telecom providers have come in the development of AI, which will further lay the foundation for the understanding on how value is created when AI is introduced to the offerings.

(10)

2.1.2 Industry structure and AI as a General Purpose-Technology

Artificial Intelligence is, due to its technical dynamism, by many considered to be a general- purpose technology (GPT) in the digital era (Bresnahan and Trajtenberg, 1995; Brynjolfsson et al., 2017; Cockburn et al., 2018). GPTs are further constructed in a way that makes them applicable to a large number of different customers, and brings productivity gains throughout the economy, according to Gambardella and McGahan (2010). An example can be found in electricity as a GPT, where all industries and firms are dependent on the electricity infrastructure in order to perform internal operations.

Electricity providers are thus able to capture a small portion of value out of a large customer base by standardization of the new technology. Further, Gambardella and McGahan (2010) argues that intro- ductions of GPTs creates intermediate technology markets where downstream enterprises are provided with required resources and capabilities that are needed to capitalize on the technology. The industry structure and how it is affected by AI, as a GPT, should therefore be considered when aiming to capture value by providing AI related solutions. The view on AI as a GPT (Bresnahan and Trajtenberg, 1995;

Brynjolfsson et al., 2017; Cockburn et al., 2018) implies that intermediate technology firms will position themselves as AI-providers, whose purpose is to develop the resources and capabilities required to form AI-related solutions (Gambardella and McGahan, 2010).

2.2 Value creation opportunities through AI integration

Value creation is often referred to as an essential function of the business model as it forms the basis for satisfying the customer’s interests (Chesbrough and Rosenbloom, 2002; Chesbrough, 2007; Zhuang et al., 2017). It is defined by a series of activities needed to create a product or service that aims to solve certain problems and meet customer expectations (Chesbrough, 2007). Therefore, it originates from the customers’ needs and how technology can be applied in order to solve identified customer issues through a set of activities. Further, the value creation function describes how resources and capabilities should be allocated in order to create the desirable value within the value creation network, where the value creation network consists of firms cooperating to enhance the value creation function (Barney, 1999), e.g. providers, customers and partners. A firm’s ability to create value is therefore bounded to the capabilities obtained through its value creation network. In this section, activities reflecting these capabilities are discussed in order to understand how AI creates value.

Further, value-creation processes derived from AI can be divided into different categories, i.e. (1) activities related to increased efficiency and cost reduction (Agrawal et al., 2019), and (2) activities related to revenue and growth (Cockburn et al., 2018). Activities related to increased efficiency and cost reduction refers to improvement and refinement of already existing operations, such as maintenance or

(11)

daily production processes (Agrawal et al., 2019). Hence, the operations’ output will be the same and the customer will not perceive any revolutionary changes in the perceived value, however cost efficiency or time spent on certain activities, related to the operation, will significantly improve. For instance, Agrawal et al. (2019) directs focus towards benefits with predictability and its impact on labor. They argue that AI can substitute parts if decision processes through digitized solutions, which subsequently will reduce the need for repetitive work tasks and thus induce cost savings. In other words, AI can perform the same task in a more cost effective manner in comparison human labor. However, Cockburn et al. (2018) argues that the potential of AI goes further than cost savings and increased efficiency, as it also moves towards the domain of revenue and growth. Activities related to revenue and growth means that, due to recent advancements of the AI technology, value can be created through its ability to support decisions and improve its outcomes (Cockburn et al., 2018). This basically implies that AI algorithms are fed with data and subsequently generate information that was not previously available, as AI is capable of solving complex problems and provide insights with higher accuracy. Information gaps are thereby filled with information that reflects reality to a greater extent. This further enables humans and technology to take actions with support from more accurate information. From a business perspective, support from AI enables managers to make more accurate decisions under conditions of uncertainty, and will thereby complement decision processes and improve its outcomes. Plastino and Purdy (2018) further argues that AI facilitates revenue and growth by acceleration of innovation and development of new solutions, resulting in new revenue streams. For instance, in the domain of drug development, Berg Health began to monitor trillions of data points derived from cancerous and non- cancerous cells, which brought insights that led the development of new cancer-fighting drugs (Thomas et al., 2016). This entails lower costs in the development of new drugs, but it also brings a new level of innovation that can not be achieved by human intervention on a standalone-basis.

However, AI should not be considered as an innovation that creates value on a stand-alone basis. Rather, Pisano and Teece (2007) argues that every innovation requires complementary products, technologies or services to fulfill its purpose. For instance; every mobile phones needs connectivity networks; airlines needs airports; and hardware needs software. Thus, the innovation must be embed- ded with the complementary product, technology or service in order to create value for the customer and if the firm is lacking capabilities and/or resources to deliver the complementary offerings, it might limit their ability for value creation. It is therefore important to understand how AI creates value, and subsequently ensure that complementary products, technologies or services are available within the value creation network. Due to the limited understanding of AI applications, how value is created and

(12)

how it can be captured, this highlights the importance of more research.

2.3 Value capture opportunities through AI integration

In this study, value capture is defined as mechanisms ensuring economic return from value creation and that profits are shared in the value creation network (Sjödin et al., 2019). Hence, value capture mechanisms builds on the provider’s role in the value creation network and ensures that profits are fairly distributed in value creation network. More specifically, Sjödin et al. (2019) mentions that governance mechanisms and legal agreements (e.g. contracts) plays a substantial role in a firm’s ability to capture value. Legal agreements increases transparency as it holds both customers and providers responsible for certain activities during the contract period, which can be understood as important as it has direct impact on each party’s performance, and subsequently also ability to capture value. Further, Gassmann et al. (2013) mention that pricing models constitutes important mechanisms in the value capture sphere as it communicates the relationship between cost structures and the applied revenue mechanisms, and thus also how the provider aims to sustain a profitable relationship with the customer. This paper will therefore look into pricing models as the core of value capture, and how contractual agreements facilitate or hinder value capture when AI is introduced. Contractual agreements can thus be understood as an extension of value propositions and pricing models, which articulates on what basis value can be captured. However, it is not enough to only understand single interrelationships between providers and customers when considering value capture opportunities with disruptive innovations, according to Gambardella and McGahan (2010). They argue that, to understand disruptive innovation and its impact on value capture, businesses must apply a broader perspective of AI and its overarching impact on industries. Therefore, to build a better understanding of how value can be captured, this section will review the literature on industry infrastructure and its impact on value capture mechanisms when offering AI. Further discussion will center on different contracts, different pricing models and their impact on value capture opportunities.

2.3.1 Value capture through pricing models

According to (Liozu et al., 2012), pricing models can be divided into two main categories, which can be framed as cost-based pricing models and value-based pricing models. The cost-based pricing model is referred to as the dominant model within the pricing domain, and it builds on the costs required to obtain the created value and subsequently a margin is added in order to ensure profitability for the provider. This margin is often determined with regard to the solution’s uniqueness in relation to competitors’ available solutions in the marketplace (Liozu et al., 2012). Hence, cost efficiency becomes a central aspect when building upon the cost-based pricing model, as lower costs equals

(13)

more attractive prices on the marketplace. Cost-based pricing models are preferable with regard to its implicitly understandable value propositions, both towards customers and providers, as costs are relatively easy to measure, meaning that providers can easily communicate how value is created through their offerings (Hinterhuber, 2008). However, despite its benefits, pure cost-based pricing models is recognized as problematic since it fundamentally ignore the customers’ perceived value and decreases the providers’ incentives for improvement and innovation (Simon et al., 2003). Due to the limitations of cost-based pricing models, both researchers’ and practitioners’ focus is shifting towards value-based pricing models (Simon et al., 2003; Hinterhuber, 2008). This is particularly apparent when developing pricing models for software applications or information goods, e.g. AI offerings, as the marginal costs are near zero (Choi et al., 2010). Unlike cost-based pricing models, the value-based approach builds on the value perceived by the customer and thus it also pay less attention to the costs required when delivering the value. According to Hinterhuber (2008), the two most profound challenges with value- based pricing models are linked to value assessment and value communication. This means that the success rate of value-based pricing models are more dependent on the providers ability understand and quantify the value created for the customer, and subsequently communicate the value created to the customer through a strong value proposition. With this said, both the cost-based approach and the value-based approach are linked to different benefits and risks, however it is unknown how AI providers relate to the two models, and to which degree value-based pricing models are suitable when AI is introduced, which highlights the importance of further investigation.

2.3.2 Value capture through contracts

Beside selection of suitable pricing models, providers can commit to different commercial arrangements affecting how value can be captured by the provider. These arrangements are further communicated to the customer through contracts, that acts as a mediator between providers and customers by stating how, and under which conditions, created value can be captured (Sjödin et al., 2019). In this section, two different contracts will be discussed. The first contract category, referred to as outcome-based contracts, is the most advanced offering as it includes a more provision of a comprehensive solution.

Rolls-Royce and the "Power-by-the-Hour" concept is a classic example of providing performance, or outcome, generated by engines rather than provision of the actual engine or complementary add-on services. This illustrates that, in outcome-based contracts, providers are compensated for the outcome, or performance, rather than single add-on services or service agreements. Many firms are shifting towards outcome-based business models in order to create and capture more value from the technology (Visnjic et al., 2017) as potential to create additional values through AI applications enhances providers oppor-

(14)

tunities to offer process outcomes, rather than single products or services. Such offerings, however, implies requirements of close collaboration between the provider and customer (Sjödin et al., 2019) and outcome-based contracts incorporates both gain and pain share between involved parties (Hou and Neely, 2018). To ensure that a fair value of portion is captured, Sjödin et al. (2019) argues that the providers that are offering outcome-based contracts must evaluate profit potential, define performance indicators, maintain incentive structures, design a profit formula. Evaluating profit potential means that the provider must identify how performance outcomes can be translated into business gains for the customer, and thereby ensure that the offered outcome aligns with customer needs. This evaluation activity must involve a diverse set of stakeholders from the customer side, including R&D and top management, in order to ensure that value can be captured. When this is understood, the provider must further recognize what profits can be shared with regards to the customers needs. Defining performance indicators can be considered as a key value-capture activity that ensures financial alignment in the outcome-based contracts. Performance indicators, in terms of KPIs, acts as a translator for the operational outcomes as it measures the operations’ financial impact. Thus, it is paramount that both parties agree upon the selected measurements in order to ensure that the profit share is based on fair conditions.

The second contract category, framed as the licensing model, is considered to be a default mech- anism for value capture when offering GPTs, e.g. AI, according to Teece (2018). This is further supported by Moeen and Agarwal (2017) who mention that licensing models enable value capture from a greater number of sources with little to none effort from the providing firm. Therefore, in contrast to the outcome-based contracts, the licensing model offers additional revenue streams while the providing firm’s workload remains the same. In addition, Moeen and Agarwal (2017) argue that the licensing model offers the ability to reach customers outside the industry boundaries, meaning that the customer base must not be limited to the industry in which the providing firm is active. Thus, the licensing model enables value capture from innovations that are already created by targeting a large set of customers, and by transferring the responsibility on how the innovation is used to the licensee.

However, Teece (2018) also mention major challenges when profiting from innovations through licensing models. It is mentioned that GPTs tend to generate large spillover effects which implies extensive economic contribution to customers, while the providing firm tend to absorb only a small portion of the value created. This points out the importance of capabilities to protect innovations with support from legal mechanisms (e.g. non-dislosure agreements, patents, trade secrecy) or by ensuring ownership of complementary assets (e.g. brand, distribution channels) to secure profitability (Pisano and Teece,

(15)

2007; Teece, 2018).

However, when discussing AI as a GPT, it is unknown how the created values should be provided in order to fully capitalize on the technology. With basis on the reviewed literature, providers may choose to offer AI applications through outcome-based contracts, licensing or a combination of the presented contracts when aiming to capture created value. This highlights the literature gap.

(16)

3 METHOD

3.1 Research approach and strategy

In this study, a qualitative approach was applied in order to answer the purpose to explore how AI providers ensure alignment between value creation and value capture when developing AI offerings.

The qualitative study was further characterized as an inductive exploratory case study, which was found suitable due to the limited theoretical understanding of activities needed to successfully implement AI, especially with regard to recent progress in the AI field. Hence, the inductive exploratory approach enabled that the purpose and the theoretical background was gradually formed in parallel with the researcher’s emerging understanding of the case (Thornhill et al., 2009). At an early stage of the study, exploratory interviews where held in order to increase the understanding of the case and thus also direct the study towards unexplored research gaps.

This study further emphasized the approach of a single case-study in order to fulfill its purpose, which was found suitable due to the limited number of articles published within the subject, especially towards the telecom industry. Therefore, a single case-study enabled a more in-depth understanding on AI’s impact on businesses. Further, the case company selection was based upon three evaluated criterions. First, the selected case-company provided AI solutions to their customers, which induces that they possessed a proper understanding of AI and potential application areas of the technology.

Second, the selected case company possessed a large and global customer base, where their offerings cover telecom provision for 1.5 billion end-users, meaning that their AI offerings covered numerous customer segments. This induces that they possessed a diverse portfolio of AI applications, and thus also possessed understanding of AI and how it can be utilized in a wide range of different settings.

Third, they had experiences from building the design of AI related offerings. Thus, their market understanding, experiences of being an AI provider and knowledge in AI constituted the basis for extending the current literature through in-depth understanding of business opportunities that follows with AI.

3.2 Data collection

In this study, primary data was acquired through interviews in three waves, where 23 interviews were held throughout the three waves. The data collection took place during the spring of 2020, where the first interview wave was initiated in February 2020 and the third interview wave ended in May 2020. More information about the interviews is visualized in Table 1. Beside data collection through interviews, the interviewees shared internal documents that were reviewed in order to acquire a better understanding regarding the case company’s provision of AI-related solutions. However, secondary

(17)

Table 1: Summary of conducted interviews.

Interview Informant Wave Position Country

Duration

(min) Type

1 I15 1 Program Manager Data and Analytics SWEDEN 30 F2F

2 I17 1 Principal Researcher SWEDEN 120 F2F

3 I7 1 Commercial Manager US 60 VIDEO

4 I8 1 Head of Capabilities Development UK 30 VIDEO

5 I9 1 Service Portfolio Manager SWEDEN 120 F2F

6 I1 2 General Manager INDIA 56 VIDEO

7 I10 2 Head of Business Development UAE 32 VIDEO

8 I11 2 Senior Technical Consultant SWEDEN 48 F2F

9 I12 2 Head of Regional Sales Engagement SWEDEN 44 F2F

10 I13 2 Global Head of Portfolio UK 35 VIDEO

11 I14 2 Key Account Manager SINGAPORE 54 VIDEO

12 I2 2 Service Portfolio Manager UK 71 VIDEO

13 I3 2 Business Development Director US 55 VIDEO

15 I6 2 Service Portfolio Director SWEDEN 60 F2F

16 I7 2 Commercial Manager US 48 VIDEO

17 I8 2 Head of Capabilities Development UK 42 VIDEO

19 I14 3 Key Account Manager SINGAPORE 46 VIDEO

20 I16 3 Principal Researcher SWEDEN 60 VIDEO

21 I17 3 Principal Researcher SWEDEN 120 VIDEO

22 I5 3 AI and Automation Development Lead SWEDEN 30 VIDEO

23 I9 3 Service Portfolio Manager SWEDEN 120 VIDEO

data, in terms of internal documents, was only used in an initial phase of the project and its primarily purpose was to build a better understanding of the case company, their AI-related offerings and characteristics as a telecom provider.

3.2.1 The first interview wave

In the first wave, meetings were scheduled with business developers that were working on AI related business models. This mainly included open dialogues through unstructured interviews, where the chosen case company explained their current business models, their progress within the AI field and their view on how AI influenced their business. With this information as a basis, the purpose of the first wave was to gain insight in the company’s situation and their practical challenges. Further, interviewees were selected based on suggestions from the supervisor and from the interviewees themselves. In total, six interviews were conducted in the first wave and they were not recorded, however notes were taken in order to acquire the most relevant information.

(18)

In parallel with the data collection of the first wave, literature within the subject of business models and business related AI research, were studied. This to evaluate whether the discussions in the first wave interviews were covered by literature, if the discussions contradicted theoretical evidence or if unexplored research gaps could be identified. Thereafter, the purpose was formulated with the aim to cover both practical- and theoretical knowledge gaps.

3.2.2 The second interview wave

An interview guide was formulated as a basis for the conducted semi-structured interviews, which can be found in Appendix A. This was found suitable as it enabled an in-depth understanding of AI’s impact on the business model, where the interviewees were able explain their view in detail, without leading questions. During the development of the interview guide, representatives from the case company gave continuous feedback in order to ensure the relevance of the questions with regard to the case study’s context and the company’s internal knowledge. This open dialogue with representatives from the case company also gave them insight in the study, which allowed them to support the project by identifying appropriate informants that were able to answer the questions with regard to the interviewees’ knowledge.

Further, interviewees were selected based on corporate position and their knowledge within the subject. Portfolio managers were interviewed in order to perceive the broad perspective of their provision of AI solutions, and 12 informants were interviewed in order to increase the in-depth understanding of particular application areas of AI. A diverse set of interviewees were intentionally selected in order to acquire a nuanced view on AI operators’ business opportunities. In addition, to enable that all interviewees were interpreted correctly, all interviews in the second wave were recorded and transcribed.

3.2.3 The third interview wave

The third interview wave was performed in order to confirm the acquired data, which enabled correction of misinterpretations during the second wave interviews. This ensured that the acquired data was accurate with regard to reality and the current situation within the case company. Additionally, it gave the interviewees the opportunity for self-correction, which can be considered important since the interview guide was not sent out to the interviewees in advance of the interviews. Thus, this limited the interviewees’ opportunities for preparation, which could have affected the quality of the interviews negatively and the purpose of the third interview wave was to mitigate this risk as the interviewees got the opportunity to add more information to the previously answered questions. Finally, the interviewer was given the opportunity to ask supplementary questions in order to ensure that all aspects of the case study was covered. To fulfil its purpose, informants in the third wave were selected based on

(19)

the information they provided in the first or second interview wave. For instance, if the informant contradicted his/her colleagues, or if the researcher found that questions were left unanswered, another interview was scheduled. Additionally, when selecting interviewees in the third wave, diversity was considered in order to ensure that different perspectives and contradicting ideas was considered.

3.3 Data analysis

This study emphasized a three step process for analyzing the collected data, including (1) identifying first order activities, (2) identifying second order activities, and (3) identifying aggregated phases.

The three step process is, according to Gioia et al. (2013), a preferable approach when analyzing exploratory data due to its systematic features that enables generation of new concepts and ideas. This was therefore found suitable as the aim of this study is to develop a framework in an unexplored research area. Further, Lincoln (2007) argues that activities related to data analysis and data collection are interconnected and dependent on each other, and that they tend to proceed together. To consider this and enable the exploratory approach, the data analysis took place simultaneously with the data collection. Thus, this made it possible to perceive the data in relation to the study’s purpose, and thus also interpret when saturation has occurred. In addition, this methodology enabled continuous iteration between empirical findings and the theoretical background, which made it possible to seek for obser- vations that extends the current understanding of the investigated phenomena. The three steps of the data analysis are further explained in this section, and an overview of the data structure can be found in Figure 1.

1. Identifying first order activities

The first step was to identify key activities mentioned by the informants, and this with little attempt to categorize the terms (Gioia et al., 2013). Thus, all activities that were covering the interconnections of value creation, value capture and AI were taken into consideration. The identification of activities, without categorization, facilitated the traceability of the study’s findings and made it possible to go back to the originating terms and understand them in their context. In order to successfully identify the first-order activities, all interviews were transcribed and read carefully. The identified activities were highlighted in the transcribed documents, and were subsequently compiled in an external document to enable an overview of the first-order activities.

2. Identifying second order activities

Secondly, the collected data was processed by seeking for similarities and differences among the identified first-order activities, and thereafter cluster interlinked activities into overarching activities (Gioia

(20)

et al., 2013). Each identified second-order activity was labeled to describe its underlying first-order activities in an theoretical language, which made the data comparable with the reviewed literature.

Hence, in this step, the iteration between the data analysis and the literature background was initiated.

3. Identifying third order phases

In the third step, the identified second-order activities were interlinked by the creation of aggregated phases. These phases constituted the third and most abstract level in the analysis and to ensure its trustworthiness, iteration between the three steps (i.e. first-order activities, second-order activities and third-order phases) were made to confirm the logic behind their connections. Simultaneous iteration with the reviewed literature also enabled building of a framework that represents how value creation and value capture mechanisms interact when offering AI related solutions. The three steps, and how they are interlinked, composes the data structure which is visualized in Figure 1.

(21)

- Under stand AI appli cabi li ty and i ts benefi ts

- Evaluate li m i tati ons w i th AI

Assessi ng AI m atur iy level

- Connect oper ati onal acti onm odels - Establi sh and over ar chi ng cloud

i nfr astr uctur e

Evaluati ng needed i nfr astr uctur e elem ents

- Assess potenti al for cost effi ci ency - Assess potenti al for r evenue

i ncr ease

- Assess busi ness gai ns

Assessi ng value cr eati on oppor tuni ti es - Evaluate local custom er

r equi r em ents

- Identi fy r equi r ed i nvestm ents

Assessi ng i ndustr y r eadi ness

Identi fyi ng pr er equi si tes for

value cr eati on

Fi r st-or der Activi ti es Second-or der Activi ti es Thi r d-Or der Phases

- Acqui r e and develop AI exper ti se - Develop under standi ng of i ndustr y

needs

- Nar r ow dow n AI appli cati on ar eas - Developi ng a data str ategy

Developi ng a com peti tive edge

- Enabli ng conti nuous developm ent - Faci li tati ng scalabi li ty

Developi ng value deliver y m odels - Evaluate r i sks w i th m ulti ple

busi ness m odels

- Evaluati ng r i sks w i th i nfr astr uctur e elem ents

Per for m i ng r i sk assessm ent

Connecti ng value cr eati on w i th value

captur e oppor tuni ti es

- Ensur e sales team com petence - Ti e AI oper ati ons to busi ness

outcom es

Cr eati ng a value str uctur e

- Cr eate a com m on gr ound for data appr eci ati on

- r esolvi ng ow ner shi p i ssues

Refi ni ng contr actual agr eem ents

- Developi ng value i ndi cator s - Ali gni ng pr i ci ng m odels w i th

contr actual agr eem ents

Deci di ng pr i ci ng m odels

Desi gni ng the value offer i ng

Figure 1: Visualization of the data structure.

3.4 Quality improvement measures

Qualitative studies can be evaluated through four key quality measurements – credibility, confirmability, transferability and dependability (Lincoln, 2007). Thus, this section will further explain how this study relate to the four quality measurements.

3.4.1 Credibility

Credibility refers to the internal validity of the qualitative study, which means that credibility evaluates the extent to which the study is comparable to reality, according to Shenton (2004). The author

(22)

further states that familiarization with the case company’s organizational structure and culture, improves the credibility measure. In this study, the first interview wave was conducted in order to acquire this understanding as the informants gave an introduction about the company and the telecom industry in general. Further, to ensure that credibility is taken into account, the data was acquired by interview- ing a wide range of informants from different levels of the organization, including top management, business developers working with specific offerings or use-cases, and technical engineers. Different expertise areas, experiences and backgrounds gave the study different perspectives on the investigated topics. In addition, the same interview guide was used during the interviews, which made the answers comparable. Thus, the wide range of informants induced triangulation via data sources, which is one way to improve the credibility measure (Shenton, 2004).

3.4.2 Confirmability

Confirmability refers to the researcher’s objectivity during the qualitative study, where high confirmability measures induces that the study’s findings represents the information given by the interviewees, rather than the researcher’s subjective ideas (Shenton, 2004). To ensure this, the confirmability measure was considered when developing the semi-structured interview guide. The interviewer asked open-ended questions, which allowed discussion rather than straightforward questions and answers (Thornhill et al., 2009). Thus, the interviewers role was to ensure that the discussions remained within the research area and that all the questions were covered, where the interviewee was permitted to speak freely. In addition, before the second interview wave was initiated, the interview guide was shared with the supervisor at the case company as well as the supervisor from the university. This allowed the researcher to adjust the interview guide and include others’ thoughts and ideas.

The third interview wave intended to secure that the collected data was correctly interpreted, as some initial findings where discussed with the interviewees. This enabled the informants to correct eventual misinterpretations made by the researcher, and thus also mitigate the risk that the interviewer’s subjective ideas influenced the findings.

3.4.3 Transferability

Transferability refers to external validity of the qualitative study, which means that transferability evaluates the extent to which the study is applicable in other contexts or situations, according to Shen- ton (2004). The author further mentions that the results of a qualitative study, due to its limitations of generic data collection, is only applicable on the investigated context and strong transferability measures are thus nearly impossible to reach. This is also the case of this case study, however this study aims to increase transparency by presenting methodological boundaries so that the transferability mea-

(23)

sure can be evaluated by the reader. These boundaries include background information about the case study, information about the interviewees and data collection methods.

3.4.4 Dependability

Dependability refers to a quality measure that consider that the study can be repeated, meaning that the same result is evident if the study is repeated with the same informants, methods and context (Shenton, 2004). However, as the circumstances in the investigating contexts are continually chang- ing, it becomes difficult to reach a high dependability measure when performing qualitative studies, according to Fidel (1993). Therefore, this study intended to provide a detailed description of the applied methodology, including research approach and strategy, operational details of the data collection method and how the data was analyzed. This information was included in order to provide a transparent description of the methodology and to enable understanding of the underlying reasons for the findings presented in this paper, which can be further used to repeat the undertaken approach in the same context. Eventual deviations in future research can thus also be explained as this study’s methodology is thoroughly described.

(24)

4 RESULTS AND ANALYSIS

This section presents the results, visualized in a process framework that builds upon three phases, i.e. identifying prerequisites for value creation with AI, connecting value creation with value capture opportunities and designing the value offering. The first phase center on activities needed to ensure value creation value with AI. Ultimately, it aims to identify value creation opportunities while also ensuring that the providing firm is able to realize the value creation activities. Further, the second phase intend to build a bridge between the value creation and value capture dimensions. This by connecting value creation with value capture opportunities, where the aim is to interconnect value creation opportunities with the targeted customer segment. Lastly, the third phase intend to design the value offering and how value is captured, where the previous phases must be considered. This as different value creation opportunities mediates different approaches in the design of pricing models and contractual agreements. Hence, this calls for iteration between the three phases in order to ensure that the created value can be captured. The process framework is visualized in Figure 2, and an overarching visualization of the data structure can be found in Figure 1. Additional quotations from the informants are found in Appendix 2. Further, this section will present a detailed explanation of the three phases and their underlying activities.

Figure 2: Framework of business model development for AI solutions.

4.1 Identifying prerequisites for value creation with AI

The first Third-Order phase center on value creation and activities needed to establish a value creation network possessing the capabilities needed to create value with AI. It is further built upon

(25)

four Second-Order Activities, framed as assessing AI maturity level, evaluating needed infrastructure elements, assessing industry readiness, and assessing value creation opportunities. In this section, the four Second-Order Activities will be further explained in consecutive order.

Assessing AI maturity level. When discussing value creation, the informants highlighted the importance of understanding the technical aspects of AI. This implies that the provider obtains an understanding of the technical aspects, where it is applicable and how it generates value. Preferably, this evaluation process should result in an detailed framework explicitly describing the technical progress of AI in accordance to Kaplan and Haenlein (2019), who describes different maturity levels of AI. The technical understanding will further lay the foundation for building an environment where its benefits can be realized. However, it is also fundamental to understand the limitations of the current AI technology, as this will have a direct impact on how value is created, as stated by the AI and Automation Development Lead:

"... when something new comes in [to the system] and the ML is not trained to handled it, it will become a problem. Reasoning handles that because now we are combining information that you may have seen in the past from other domains, or external data sources, or ML."- I5

Evaluating needed infrastructure elements. When the technical understanding of AI is possessed by the provider, it is further important to understand what surrounding infrastructure elements that are needed in order to create the desired value. It is mentioned by the informants that the main benefits with AI are derived from its ability to predict unexpected happenings and events that will cause faults in the network. Output from the AI model can further provide information on when, where and how faults can be handled before they occur, which offers the ability to apply a proactive approach of maintaining the network. However, the perceived customer value is therefore also dependent on how, when and where actions are taken, meaning that the information provided by the AI model does not create value by itself. Rather, operational actions taken by the receiver of the information will decide whether, and to what extent, the fault can be handled which further decides how the customer perceives value. Thus, a model for when, where and how operational actions are taken, based on the provided information from the AI model, is required to create any value from the information provided by AI. This operational action model can further involve humans or complementary techniques enabling automation, according to the Service Portfolio Manager in UK:

"We say ’yeah, we predicted it’, but if we don’t take action we will not create value, we

(26)

now need to take it to the next step with automation, so using our interconnected platforms to enable our automation platform and having activities that we can do automatically of the back of these predictions is really the end game for us"– I2

Further, it is also important that the infrastructure is built upon an overarching cloud solution that enables learning in an environment that includes as much data as possible. According to the informants, the potential of AI increases if the algorithms are trained in a centralized environment where they can access larger amounts of data, rather than decentralized customer locations with limited data access.

Therefore, this demands an infrastructure that supports learning activities and thus also value creation.

"We are now gong to be running our tools on a private cloud, we got the data on a private cloud, and effectively, we have to move the data from the customers tools to our tools, to be able to leverage the benefits with automation and AI etcetera."– I2

Assessing industry readiness.Throughout the interviews, the informants pointed out the importance of industry readiness. This includes governmental regulations in different geographical areas, which affects how data can be utilized when creating value through AI solutions. Governmental regulations mainly encounter end-subscriber information, e.g. personal information about individuals or consumer patterns. Evaluation of these regulations is important in order to realize limitations regarding value creation activities, however regulations are varying between different geographical areas which requires analysis on a decentralized basis. Additionally, it is of importance to evaluate whether the customers obtains the needed data and their approach to sharing data. The customer’s approach to data, on a general basis, will have direct impact on how and when it can be accessed, and ultimately how the AI provider will be able to create value, which highlights the importance of evaluating the customer’s interests in this domain. Lastly, industry readiness involves customer’s approach to act on insights provided by AI models. This requires that customers trust the provided information and subsequently realizes changes in accordance with the provided insights. However, this will likely require heavy investments or re-prioritized action plans, which can lead to customer resistance. As previously stated, the ultimate value create by AI is directly dependent on how the provided information is utilized, which highlights the importance of assessing industry readiness.

Assessing value creation opportunities. Thirdly, the technical AI understanding should be translated into potential value creation opportunities, with respect to the identified prerequisites in the surrounding environment. This study proposes assessment in three areas, i.e. cost efficiency improvement, revenue increase and business gains. Thereby, this study’s results confirms that AI operations supports

(27)

cost efficiency through its ability to provide decision support by handling large and complex amounts of data. Insights, provided by AI, can thus be used to make informed decisions and subsequently reduce costs, while the process output remains the same. Additionally, this study proves that information provided by AI can facilitate revenue increase by providing qualitative information that improves process output, according to the Head of Capabilities Development at Alpha:

"People at the lower job stage are doing quite repetitive work and shift work so, at the first step we view automation within that area where we would use AI to automate that work and clear away all that noise"- I8

However, while insights provided by AI can have a direct impact on decisions on an operational level and thus also directly impact cost efficiency and revenue increase, this study extends this view as AI also provides higher level insights resulting in overarching business gains. This means that AI is capable of delivering insights at both operational- and business levels, which demands different application areas and thus also results in different values. With this said, this study proposes to make a distinction between operational revenue increase and overarching business gains. The informants ex- emplified with numerous different business gains created by AI, however, increased awareness among actors in the value creation network was mentioned as a fundamental business gain. This is particularly important due to the increased volume of devices to handle, more diverse service requirements and increased complexity of network operations. Despite this change, AI provides the ability to correlate different data sources to identify and measure created value, and thereby increase awareness on how the network should be operated.

"For the existing customers, we have dashboards so we can monetize what the SLA is, so we won’t end up signing anything we can’t, or need to keep, there will be many assessments to see, and it is going to help both operators and us. So, when it comes to SLAs, we will sign and being more aware than the earlier SLAs, aware of the value we are creating because we can see and define that. We are aware of the cost which we will enter because it will be our tools that we need to reinforce for a valid reason"– I5

To sum up, evaluation of opportunities within the three value categories, i.e. cost efficiency, revenue increase and overarching business gains, is required in order to understand the value creation dimension when AI is introduced. In addition, to confirm that the identified values are possible to realize, it is required to iterate between the four Second-Order Activities.

(28)

4.2 Connecting value creation with value capture opportunities

The second phase aims to build a bridge between the value creation- and value capture dimensions, with basis of the gained knowledge from first phase. Unlike the first phase, this phase is more customer oriented as it aims to specify how and where values will be created, and subsequently tie the identified opportunities to the customers’ needs. The three second order activities, i.e. developing a competitive edge, developing value delivery models, and performing risk assessment, will be described in consecutive order.

Developing a competitive edge. Informants stated that the value creation network must not only possess in-depth AI expertise and proper understanding of industry needs, but also abilities to combine AI expertise and industry understanding when developing the AI offering. Given the identified prerequisites surrounding the AI solution, the provider must therefore define and develop the scope and the business area and to which customer segment the AI solutions will be applied. Hence, in this stage, the provider must utilize their industry understanding in order to develop a conceptual framework for potential AI application areas. In other words, this means combining AI expertise with industry understanding to further exploit the potential of AI. The purpose of this phase, i.e. developing a competitive edge, is therefore to tie value creation opportunities to actual customer needs, and to refine value creation activities with regard to different customer segments. As the Business Development Director in the US expressed:

"Operators around the globe have most data available, and they are the least innovative in terms of AI and ML worldwide. So, to understanding those structured data components, which includes understanding the performance counters, which is a decision tree mecha- nism. So as we have the technical operations expertise and the telecom experience, we see that for each scenario, which model works best"- I3

However, when developing the competitive edge by blending AI expertise and industry understanding, it is further important to fully exploit the providing organization’s capacity in terms of industry understanding. It is mentioned that AI offerings are limited by the providing organization’s imagina- tion, and that the number of application areas will increase exponentially over time. Thus, the concep- tualization of potential AI usage areas should not be limited to the traditional AI delivery organization, but rather spread across the organization in order to enable AI application towards any customer. The Commercial Manager in the Us expressed it accordingly:

"One unique there is, on our radio side, we will have more knowledge radios than we do of

(29)

our competitors. And, on the managed services side, we do operate everybody’s network, so how do we blend those two and not cross any inappropriate boarders, but how do we drive our advantage in operating networks?"– I7

All informants mentioned that data constitutes a central aspect when discussing AI solutions, as data has direct impact on the algorithm’s output. Therefore, it is important to access large amounts of data in real time in order to enable value creation through AI. However, as previously stated, data usage is often limited by data sensitivity and the customers’ general resistance to sharing data. There- fore, it is important that the AI provider extends their AI focus area with a suitable data strategy that communicates why, when and how data must be accessed in order to enable value creation through AI. Essentially, the provider must develop a clear strategy regarding how the data is used and where it is stored. This as resistance to sharing data often occur when the customer is unaware of how data will be used and who has access to the data. Additionally, the data strategy must communicate in what format the data is accessed, and how it is transferred into readable format which can be handled by the algorithms.

"80% of the activity with machine learning is getting the data in the format that we can then learn upon, and I think the more customers we do this on, the more maturity we have and the quicker we can do new implementations."– I2

Developing value delivery models.To realize the conceptual AI framework it is further important to develop value delivery models, stating how the AI solution is brought to the the customer. In this area, the informants mentioned the importance of enabling continuous improvement of data sources and the AI solutions in different contexts, where the AI provider’s competence is required to ensure that value is perceived by the customer. The informants mentioned that this requires close collaboration between the AI provider and its customers as this induces better preconditions when AI solutions are applied.

This means that outcome-based business models increases opportunities to improve value creation and value capture, where the provider-customer relationship is characterized by high customer engagement.

This as R&D investments and development of AI solutions, performed by the AI provider internally to acquire the technical capability, can be customized and integrated to the customer’s environment. This is mentioned as inside-out development, supporting continuous development and industrialization of AI solutions through intense customer engagement. When discussing outcome-based delivery models, the Head of Regional Sales Engagement in Sweden expressed the key benefits accordingly:

(30)

"We had a holistic view [when developing AI solutions], from our perspective and the delivery perspective, meaning that it is more inside-out. But we didn’t always connect it to the customer’s needs, outside-in, what do they need? How can it be bought? How is the value assessed? We mostly looked at it from our perspective, but when this is done, we will have an extremely powerful offering"– I12

It is further mentioned that tight customer engagement creates lock-in effects, where the customer becomes dependent on the provider’s abilities to leverage the potential values offered by AI solutions.

Thus, outcome-based delivery models are beneficial in order to improve customer retention measures and thus also secure long-term profitability. However, intense customer collaboration has its limitations when it comes to scalability, since the AI provider will only engage with the most profitable customers and thus also exclude customers where the value capture opportunities are considered to be lower. In this study, it is found that licensing strategies supports scalability to a larger extent due to the limited commitment level required by the provider. This licensing delivery model can therefore be beneficial when considering AI as a GPT with the potential to capture value from a large number of different customers. This is further supported by the General Manager in India:

"You can take it off the shelf and buy it and use it. So, this is a new situation for us and for the customer. For us, we are getting additional revenue from the same efforts, from developing those algorithms, and we are also getting an entry into the customers domain by even offering very limited solutions. So, it helps us expanding our offerings."– I1

With this said, multiple delivery models must co-exist in order to fully capitalize on AI scalability benefits while simultaneously enabling continuous development. Importantly, when developing value delivery models the provider must consider both inside-out development opportunities and, at the same time, recognize scalability.

Performing risk assessment. The informants mentioned several risks associated with outcome based- and licensing models, and highlighted the importance of thoroughgoing assessment of risks.

When discussing outcome-based risks, the informants primarily focused on customers’ resistance to sharing data and development of value indicators. Insufficient or misdirected value indicators have a direct and negative impact on the provider-customer relationships which entails limited data access.

Also, when discussing the licensing model, it was mentioned that these risks are amplified. This as the provider has less control of the processes where the AI-model is applied, since the provider solely relies on the performance of the AI model and thus also disclaims responsibility for the interconnected

Value creation and value capture in AI offerings