Control of algorithms in industry 4.0 An explorative study on the application of control mechanisms on algorithms in industry 4.0. Master’s Thesis in the Master’s Programme Entrepreneurship and Business Design MARIA DEL CARMEN ESPEJEL CARRIÓN SIMON RISBERG Department of Technology Management and Economics Division of Entrepreneurship and Strategy CHALMERS UNIVERSITY OF TECHNOLOGY Gothenburg, Sweden 2018 Report No. E 2018: 096 MASTER’S THESIS E 2018: 096 Control of algorithms in industry 4.0 An explorative study on the application of control mechanisms on algorithms in industry 4.0. MARIA DEL CARMEN ESPEJEL CARRIÓN SIMON RISBERG Tutor, Chalmers: Magnus Eriksson Tutor, SKF Group: Karthik Venkitesh Department of Technology Management and Economics Division of Entrepreneurship and Strategy CHALMERS UNIVERSITY OF TECHNOLOGY Gothenburg, Sweden 2018 Control of algorithms in industry 4.0. An explorative study on the application of control mechanisms on algorithms in industry 4.0. M. ESPEJEL CARRIÓN, SIMON RISBERG © M. ESPEJEL CARRIÓN, SIMON RISBERG, 2018. Master’s Thesis E 2018: 096 Department of Technology Management and Economics Division of Entrepreneurship and Strategy Chalmers University of Technology SE-412 96 Gothenburg, Sweden Telephone: + 46 (0)31-772 1000 Chalmers Reproservice Gothenburg, Sweden 2018 During the spring of 2018, the authors conducted the master thesis in collaboration with SKF Group. This master thesis is the final step for finalizing the Master Programme in Entrepreneurship and Business Design at the Department of Technology Management and Economics at Chalmers University of Technology, Gothenburg. We especially would like to thank our supervisor at SKF Group, Karthik Venkitesh for the support and guidance given during the different stages of the master thesis. We would also want to thank Robin Ristander and Martin Jansson for their valuable feedback and thoughts given during formal and informal meetings. We would also like to give special thanks for our supervisor from academia, Magnus Eriksson who has motivated and guided us along the way. In addition, we would like to thank Bowman Heiden and Christoffer Hermansson for their additional support given. The authors would also like to thank Marcus Andersson and Johan Wiklund for being great opponents as they helped validate the quality of our thesis. Finally, we would like to thank our family and friends for the huge support given during our master thesis. Gothenburg, 25th of May 2018 Maria del Carmen Espejel Carrión Simon Risberg Abstract The world is on the brink of a fourth industrial revolution that is characterized by the interconnection and digitalization of machines and their environment which enables industrial companies in the manufacturing industry to create new value propositions that are based on smart products and services. At the core of this industrial revolution lies the key term called internet of things which includes a network of physical devices that are enhanced with sensors, connectivity and software. It turns regular physical devices into smart complex products that can capture and process data and turn it into valuable insights through data analytics. What drives the data analytics solution are the algorithms that are trained to solve a specific problem. As mechanical products are turned into smart products through embedded software, traditional industrial companies do not only have to think about protecting their hardware solutions but also their software solutions. As the main driving force for data analytics are the algorithms, industrial companies are faced with the challenge of controlling them in order to stay competitive in their respective industry. However, there is an uncertainty in regard to how to control algorithms which calls for an explorative approach. Therefore, the aim of the study is to explore how industrial and software companies are applying control mechanisms and provide recommendations on how industrial companies in the manufacturing industry should apply control mechanisms on algorithms to leverage a competitive advantage. As the phenomenon is rather nascent the study was conducted in an abductive approach and two methods of gathering information were done through a literature review and interviews. The research design used was a cross-sectional design in which two types of groups were interviewed based on qualitative research with semi-structured interviews in order to gather as much information as possible. The first group was IP professionals which helped to identify how well control mechanisms work in practice on algorithms while the second group which were software managers helped to identify the different trends in how companies that are working with algorithms in-house are applying control mechanisms on a daily basis for a competitive advantage. With the help of the two different groups interviewed the study established the fact that due to the very nature and intangibility of algorithms, the control mechanisms had to be applied both from an internal point of view in the company as well as from an external point of view towards the environment of the company. The study also found that the different control mechanisms applied on algorithms will differ externally depending on what level of the algorithms you want to control and in which interface that an industrial company is operating with whether it is towards a software supplier, a competitive industrial company or a customer. Keywords: algorithm, competitive advantage, control mechanisms, data analytics, industry 4.0, internet of things, knowledge assets. Abbreviations API Application programming interface BSD Berkeley Software Distribution DRM Digital Rights Management DTSA Defend Trade Secrets Act EPO European Patent Office EU Europe GPL General Public License IAM Intellectual asset management ICT Information and communication technology IoT Internet of things IP Intellectual Property IPRs Intellectual Property Rights ISO International Organization for Standardization IT Information Technology MRQ Main research question NDA Non-disclosure agreement OEM Original Equipment Manufacturer OT Operational Technology PRV The Swedish Patent and Registration Office PTAB Patent Trial and Appeal Board PTO Patent and Trademark Office R&D Research and development RQ Sub Research question TRIPS The Agreement on Trade-Related Aspects of Intellectual Property Rights UPC Unified Patent Court US United States Table of contents 1. INTRODUCTION 1 1.1. BACKGROUND 1 1.2. PRIOR RESEARCH 3 1.3. PROBLEM 3 1.4. PURPOSE 3 1.5. RESEARCH QUESTIONS 4 1.6. DELIMITATIONS 4 1.7. THESIS OUTLINE 5 2. INDUSTRY 4.0 6 2.1. THE FOUR REVOLUTIONS OF THE INDUSTRY 6 2.2. THE INTERSECTION OF OT AND IT 7 2.3. NEW BUSINESS MODELS IN THE FOURTH INDUSTRIAL REVOLUTION 7 3. METHODOLOGY 9 3.1. RESEARCH STRATEGY 9 3.2. RESEARCH DESIGN 10 3.2.1. IDENTIFYING THE THEORY AND REQUIRED DATA 10 3.2.2. THE EMPIRICAL APPROACH 12 3.2.3. DATA SAMPLE 13 3.2.4. DATA ANALYSIS 14 3.3. QUALITY OF RESEARCH 14 4. THEORETICAL FOUNDATION 16 4.1. INTERNET OF THINGS 16 4.1.1. INTRODUCTION 16 4.1.2. TECHNOLOGY STACK OF THE INTERNET OF THINGS 16 4.1.3. OPPORTUNITIES FOR INDUSTRIAL COMPANIES 18 4.1.4. CHALLENGES FOR INDUSTRIAL COMPANIES 18 4.2. BIG DATA ANALYTICS 19 4.2.1. A NEW AGE OF ANALYTICS 19 4.2.2. THE PROCESS OF BIG DATA ANALYTICS 19 4.2.3. THE METRICS OF DATA ANALYTICS 20 4.3. ALGORITHMS 20 4.3.1. THE IDEA OF ALGORITHMS 20 4.3.2. ALGORITHMS IN MACHINE LEARNING 21 4.3.3. ALGORITHMS IN OPEN SOURCE 22 4.3.4. ALGORITHMS IN AN INDUSTRIAL PERSPECTIVE 22 4.4. KNOWLEDGE ECONOMY 23 4.4.1. INTRODUCTION TO KNOWLEDGE 23 4.4.2. KNOWLEDGE TRANSFER 24 4.4.3. ORGANIZATIONS AND MARKETS BUILT ON KNOWLEDGE 25 4.5. CONTROLLING OF INNOVATIONS 26 4.5.1. IAM FRAMEWORK 26 4.5.2. INTELLECTUAL BUILDING BLOCKS 28 4.5.3. THE THREE ARENA MODEL 37 4.6. COMPETITIVE ADVANTAGE IN THE INDUSTRY 38 4.6.1. COMPETITIVE ADVANTAGE FROM AN EXTERNAL PERSPECTIVE 39 4.6.2. COMPETITIVE ADVANTAGE FROM AN INTERNAL PERSPECTIVE 40 4.7. SUMMARY OF THEORY 41 5. RESEARCH FINDINGS 42 5.1. PATTERNS FROM IP PROFESSIONALS 42 5.1.1. RIGHT BASED PROPERTY 42 5.1.2. SECRECY 48 5.1.3. CONTRACTUAL BASED PROPERTY 50 5.1.4. TECHNICAL CONTROL 52 5.2. PATTERNS FROM SOFTWARE MANAGERS 54 5.2.1. RIGHT BASED CONTROL 54 5.2.2. SECRECY 58 5.2.3. CONTRACTUAL BASED PROPERTY 60 5.2.4. TECHNICAL CONTROL 62 5.2.5. MARKET POWER 64 6. ANALYSIS AND DISCUSSION 66 6.1. INTRODUCTION 66 6.2. MANAGING COMPETITIVE INDUSTRIAL COMPANIES 68 6.3. MANAGING THE BUYERS 71 6.4. MANAGING THE SUPPLIERS 73 7. CONCLUSION 76 8. FURTHER RESEARCH 80 9. TABLE OF FIGURES 81 10. BIBLIOGRAPHY 82 11. APPENDIX 89 11.1. FIGURES 89 11.1.1. PTO GUIDANCE ON SUBJECT-MATTER ELIGIBILITY 89 11.1.2. EPO STEPS FOR DETERMINING PATENTABILITY 90 11.1.3. SUMMARY OF TRENDS IN CONTROLLING ALGORITHMS 90 11.2. IP PROFESSIONALS INTERVIEWS 91 11.3. SOFTWARE MANAGERS INTERVIEWS 91 11.4. IP PROFESSIONALS INTERVIEW TEMPLATE 92 11.5. SOFTWARE MANAGERS INTERVIEW TEMPLATE 94 1 1. Introduction The main focus of the master thesis is to understand how industrial companies can apply control mechanisms on algorithms in order to leverage a competitive advantage in industry 4.0. This chapter consists of different sections that outlines the background of the thesis, its problem, purpose and research questions. Furthermore, it explains in short what limitations the thesis has and it will end up with a thesis outline. 1.1. Background Digitalization has had a huge impact in the manufacturing industry and therefore, a new era has been established which is commonly referred to as industry 4.0 (European Patent Office, 2017). Industry 4.0 is something that is characterized by the increasing digitalization and interconnection of products, value chains and business models (PwC, 2016). In this future industry, machines are interconnected and are conceived as a collaborative community that can capture data and systematically process it into information with the main goal of explaining uncertainties such as when it is time for a machine to have maintenance which is also referred to as condition monitoring (Lee, Kao and Yang, 2014). The value of industrial companies was based on the exchange of goods which was seen as manufactured output (Vargo & Lusch, 2004). This type of value was limited due to the fact that products are made out of raw material which can be seen as a finite resource (Vargo & Lusch, 2004). However, with the emergence of the internet of things, industrial companies are able to create new complementary value propositions alongside their core hardware products that can be leveraged in the industry. These new type of value propositions are focused on induction and impetus of service (Lee, Kao and Yang, 2014). At the heart of this industrial revolution lies a specific phenomenon which is called the internet of things. It has the ability to transform products into complex systems that combine hardware sensors, data storage, microprocessors, software and connectivity in different ways (Porter and Heppelmann, 2015). These new types of complex products will serve to alter the industry in new ways. Thus, changing the nature of the competition with the emergence of new type of entrants such as software companies who can now become a part of different industries either as suppliers or competitors (Porter and Heppelmann, 2014). The interface between the suppliers and industrial companies will be reshaped as the suppliers can both be software and hardware companies due to hardware companies providing hardware while software companies can provide software for industrial companies or develop it together with them. The software companies as suppliers can leverage their knowledge in software as a strong bargaining power towards the industrial companies as their niche becomes rather wide due to the wide application of algorithms. The software companies can also be seen as competitors if the software isn’t tightly coupled with the smart hardware product which means that software 2 companies can offer their services directly towards the end customer. At the same time the added value proposition of embedded software makes the bargaining power of customers rather low as the lock in effects become greater due to industrial companies being able to create closer relationships with the end-user through data collection and the training of algorithms thus increasing the switching costs for customers (Porter and Heppelmann, 2014). Figure 1 - Overview of industrial companies’ interfaces The true value enabler when it comes to internet of things is data analytics (European patent office, 2017). The reason for that is the fact that organizations in general want to know and need to know what is happening now, what is likely to happen next and what actions should be taken in order to get the optimal result (Levalle et al., 2011). According to a study conducted on several industrial companies by (Levalle et al., 2011) there was a mutual agreement that analytics is differentiating them from their competitors in their particular industry due to increased efficiency and a reduction of total costs. The driving force in data analytics are algorithms as they are aggregating raw data into valuable insights and therefore there is a need to control them. A rather simple explanation of what an algorithm is, is that it can be seen as a well-defined computational procedure that takes some sort of value or values as input and produces new value out of it. It can be seen as a sequence of computational steps that transform the input into output (Cormen et al., 2001). Algorithms have been recognized as being patentable but hard to defend from being invalidated in court due to increased scrutiny based on previous court cases as well as more aggressive challenges by patent infringement defendants (Prange, 2017). There has been very little research in regard to what type of control mechanisms can be used effectively on algorithms and more so if they can be seen as a sustainable solution (Prange, 2017). As previously stated we are on the brink of a new industrial revolution where data analytics will be seen as a differentiating factor and therefore there is a need to investigate how industrial companies can 3 build strategies around control mechanisms in order to control algorithms to keep a competitive advantage between the different interfaces in the future value chain (Levalle et al., 2011). 1.2. Prior research An in-depth literature review has been done in order to identify previous research. The areas of interest followed a certain structure with the main goal of narrowing down the scope. First of all, a literature review was depleted in the area of industry 4.0, followed by internet of things, data analytics and finally focusing in algorithms. A focus has been of understanding all of these four different components from a business perspective. The authors have also found information about applying control on software, however in regard to algorithms the research has been very limited which called for bundling the information in order to come up with the application of control mechanisms on algorithms. Lastly, the authors read literature about competitive advantage in the industrial sector in order to understand what gives an industrial company a competitive edge. 1.2. Problem Digitization is something that has been affecting a handful of industries and turning them into industry 4.0 (European patent office, 2017). It helps to create new value propositions for industrial companies as it gives them the opportunity to turn their mechanical products into smart software products through embedded software (Porter & Heppelmann, 2015). This means that they are not only focusing on manufacturing technique innovations anymore but have also started to move into and focus on induction and impetus of service (Lee, Kao and Yang, 2014). In the end what will differentiate one industrial company from another will be the capability of delivering highly qualitative data analytics as a service alongside their product offerings for their customers (Levalle et al., 2011). The problem arises because there is an uncertainty when it comes to the future application of control on software assets primarily algorithms in order to maintain a competitive advantage in the future industry 4.0 (Prange, 2017). So far, the authors have found very limited information about how one can practically implement control mechanisms in relation to algorithms in data analytics which calls for an explorative analysis. 1.3. Purpose The purpose of the study is to explore how industrial companies and software companies are applying control mechanisms on their algorithms. In addition to provide recommendations on how industrial companies in the manufacturing industry 4.0 can apply control mechanisms in relation to algorithms in order to leverage their competitive advantage in the future value chain. 4 1.5. Research questions MRQ: How can industrial companies in the manufacturing industry use control mechanisms on algorithms in order to leverage their competitive advantage in industry 4.0? The main research question has been divided into two sub research questions in order to identify the applicability of the different types of control mechanisms on algorithms in theory and practice but also to investigate what trends one is seeing in algorithms when it comes to control at other companies. RQ1: What type of control mechanisms are applicable on algorithms and how do they work in practice? This research question will provide an understanding of what type of control mechanisms that are applicable on algorithms from a theoretical and practical point of view. In order to answer the research question the authors will have a literature review as well as interviews with IP professionals. The results obtained from this research question will help to answer research question two. RQ2: What type of trends can be seen in the application of control mechanisms on algorithms? This research question will provide an understanding of how software and industrial companies are applying control mechanisms in practice when it comes to algorithms. In order to answer this research question the authors will have interviews with software managers from both industrial and software companies in order to understand how companies are applying control mechanisms on algorithms. 1.6. Delimitations The scope of the thesis will be focused on algorithms in regard to data analytics in industry 4.0. After conducting a thorough prior research, the authors noticed that there was a lack of research in the control of algorithms in data analytics. Therefore, hardware and networking related issues have been excluded. Due to time constraints, the study has been limited to a certain number of interviews with experts in intellectual property referred to as “IP professionals” as well as “software managers” which are experts in software architecture and development together with an in-depth literature review. Due to the business nature of the thesis, the authors have decided to have a general overview on the literature review when applying different control mechanisms with a particular focus in jurisdictions such as US, EU and Sweden. 5 1.7. Thesis outline Chapter 1, Introduction, which outlines a background, some prior research done in the area, a problem definition. Moreover, it defines what the purpose is, research questions and delimitations. Chapter 2, Industry 4.0, which will define and outline of the phenomena of the study. Chapter 3, Methodology, describes the research strategy chosen, the research design chosen and handles the quality of the research. Chapter 4, Theoretical foundation, describes the main theories and concepts guiding the research. Chapter 5, Research findings, comprises the empirical data of the study which is presented and summarized. The empirical data includes interviews with IP professionals as well as interviews with software managers in order to obtain empirical data for research question one and research question two. Chapter 6, Analysis and discussion, involves an analysis and discussion around the results of the thesis. Chapter 7, Conclusion, gives an explanation of where the different research questions are answered and compiles the most important results of the study. Chapter 8, Further research, gives an explanation for further research that should be done in this area. 6 2. Industry 4.0 This chapter introduces the fourth industrial revolution by going through the different revolutions that has led up to the fourth revolution, the intersection of two technologies that makes up for the new revolution and the new business models that are created in this new industry. 2.1. The four revolutions of the industry The word “industrial revolution” can according to (Schwab, 2015) be defined as a group of technological inventions that will fundamentally alter the way we live, work and relate to one another. Up to this date, the world has gone through four industrial revolutions that have helped society develop, become more effective in leveraging its resources and thus initiate economic growth. The first industrial revolution started in the beginning of the 1760s where society discovered that if you heat up water it will turn into steam and thus steam power was created which helped society move from an agrarian life based on farming into urbanization. The world started to rely on steam power in order to power everything from agriculture to manufacturing as well as helping people go from point A to point B with the help of railroads and steamships (Deane, 1965). In the beginning of the 19th century the second industrial revolution came to be. It was characterized by division of labor in the assembly line where workers became specialized in their specific area in order to enable mass production as well as the discovery of electricity (Kanji, 1990). What laid the basis for the third industrial revolution were the inventions that laid the ground for the capabilities of digitalization. Some of these inventions were mainframe computing, semiconductors, personal computing and the internet. These inventions made it possible to move from analog electronic devices to digital technology and devices thus disrupting global communications and energy (Salesforce, 2018). Through the recent development in cyber-physical systems which can be seen as the next step of computing and the internet of things we find ourselves at the fourth industrial revolution which entails connected appliances, machines, things, factories and industrial environments to the internet (Bloem et al., 2014). 7 2.2. The intersection of OT and IT At the core of the fourth industrial revolution lies the fusion between traditional operational technology (OT) that focuses on machines with the purpose of making different industries go around through engines rotating as well as transporting objects. On the other hand, there is information technology (IT) which has the purpose of supporting different business processes by combining different devices with each other thus providing different types of information through the internet (Bloem et al., 2014). These two different areas have for a long time been held apart as operational technology can be seen as rather industry specific based on hardware as well as software that is used to monitor different processes (Atos, 2012). Here, one needs to have specific knowledge about the different technologies while information technology is not bound to specific industries as it handles the general information flows and connectivity which is something that has been of growing importance for several industries in this new age (Atos, 2012). However, the information that is created by the different operational technologies in the different industries is seen as crucial as it serves as a basis for better decision making in order to increase efficiency as well as reduce a total cost expenditure in the manufacturing industry (Bloem et al., 2014). In order to gather the information from several operational technologies, information technology is necessary which means that a collaboration between the two technologies will generate create benefits for an industrial company. 2.3. New business models in the fourth industrial revolution Due to the emergence of embedded software in products which are evolving the manufacturing industry, new business models are being created. Industrial companies are not only looking to improve the manufacturing technique anymore, but they are moving towards a focus on service as well (Lee, Kao and Yang, 2014). The specific term is called servitization and it emphasizes a stronger customer focus where combining products with services, support and knowledge are the most important elements for a competitive advantage in the future industry (Vandermerwe & Rada, 1989). Because of the service content view that the fourth industrial revolution brings to the table the business models naturally change as well towards a more network-centric view where the focus is on creating an ecosystem based on a network across the horizontal value chain (European Commission, 2016). In this view, new type of elements are added or updated such as the value proposition, the revenue mechanism, the target market, the value network and the competitive strategy (European Commission, 2016). With these new value networks, one can no longer just examine the major competitor in a specific domain anymore, but one must also examine the network of firms that relate to that specific competitor (Kothandaraman and Wilson, 2001). The three components that will intertwine in order to create an impact in the total value chain 8 will be value, core capabilities and relationships. In order for value to be created for the customer the firm has to fully integrate the resources to use the core capabilities of the firm. In regard to core capabilities a firm can be lucky if it has at least three or four major core capabilities. In today’s environment the core capabilities of different companies are becoming narrower and sharper (Kothandaraman and Wilson, 2001). This leads us in to the last component which is relationships. In order to create customer value in today’s environment one has to gather as many core capabilities as possible which one gains through relationships with companies that have different core capabilities (Kothandaraman and Wilson, 2001). 9 3. Methodology This chapter consists of three main components where the first one is research strategy (3.1) that explains how the overall research strategy for the thesis has been. The second component which is the research design (3.2) focuses on what type of design that was chosen for the study and what it means, the quality of research (3.3) is the final component and assesses that the quality of the research remains stable. 3.1. Research strategy Looking at the overall research strategy and what it means for research (Bryman and Bell, 2015) mentions that it helps to explain and tighten the link between theory and research and stresses that this is by no means a straightforward matter. The study has been conducted from an abductive approach which means that one starts with a puzzle through an observation which cannot be fully explained with current theories. This means that the authors will try to find the best explanation for the puzzle (Bryman and Bell, 2015). The reason for that is because the market in terms of algorithms is seen as rather new for the manufacturing industry compared to the software industry as they are slowly moving into new territory in which they will add on value propositions based on software which means that not much has been written in regard to control of algorithms in software in the industrial sector. Epistemological considerations are important for the study due to it concerning the question of what is or should be regarded as acceptable knowledge in a certain discipline (Bryman and Bell, 2015). In short words it asks the question “How can we justify that this is acceptable knowledge?”. It has the focus on the difference between the art of positivism which means that true knowledge is primarily founded on input from the senses and constructionism which means that true knowledge does not come from the senses but instead comes from the reasoning of the human mind. Ontological considerations are also something that are important and according to (Bryman and Bell, 2015) it questions whether the social world can be viewed as something external that can exist without social actors which is a positivist approach or if it simply can’t exist without social actors which can be seen as a constructionist approach. In other words, it asks the question of “whether a phenomena or object can exist without the human mind being there to think of it”. From an epistemological point of view the study can be seen from a constructionist point of view due to it being mainly qualitative interviews and the questions are formed in a way that the interviewees are answering the questions based on personal experiences as well as best practices which is then translated into an industrial perspective. 10 The study at hand is looking into how one can implement different control mechanisms on algorithms. Looking at it from an ontological perspective the study can be seen from a constructionist approach because first and foremost the object that the study is focusing on are algorithms which cannot exist by themselves without social actors being involved. Secondly, the frameworks that are being used in order to understand how to control algorithms and what it means have also been constructed by social actors which further emphasizes the constructionist approach. 3.2. Research design When talking about the design of the study that is made for the host company it is based on a cross-sectional design. According to (Bryman & Bell, 2015) a cross-sectional design is defined as “The collection of data on more than one case and at a single point in time in order to collect a body of qualitative or quantitative data which are then examined to detect patterns of association”. Going back to the study at hand it focuses on how industrial companies in the manufacturing industry should implement control mechanisms on algorithms to leverage their competitive advantage in industry 4.0. The goal of the study is to understand what control mechanisms are applicable on algorithms from a theoretical point of view and how well they work in practice. In addition, the authors want to understand how companies that are working with algorithms are dealing with the issue of managing and controlling algorithms in order to understand where the trend is heading to later provide recommendations for industrial companies in the manufacturing industry. The journey of this study has been divided into four parts where the first one is focusing on how the authors identified the theory and what type of data had to be collected. The second part focuses on the empirical approach and describes how the authors conducted the study and what techniques were used. The third part is focusing on the data samples where the different groups who were targeted by the study are explained. The last part, which is the data analysis, goes through how the data was compared and why it was compared in the way it was. 3.2.1. Identifying the theory and required data During the phase of identifying the theory as well as what data would be required there was a question of iteration and brainstorming due to the abductive nature of the research strategy on the basis that the control perspective on algorithms was fairly new in an industrial setting. One of the most critical parts of carrying out a research project is reviewing the literature in the chosen subject area because it provides the basis that one can use in order to justify the research questions and thus being able to build a research design (Bryman & Bell, 2015). The authors have had a supervisor both from the institution that the thesis is being written at as well as the host company that the current project is being held at. By having both supervisors from academia and the host company, the authors got help from academia which was good because 11 the supervisor who is handed to a specific group is usually someone who is well versed in the research process (Bryman & Bell, 2015) while the supervisor from the host company helped the authors to identify the important aspects of the industry. Due to the first sub-question being “What control mechanisms are applicable on algorithms and how do they work in practice?” and the second sub-question being “What type of trends can be seen in controlling algorithms?” the authors needed to identify relevant theory in several fields such as information technology, law and industry in order to first explain how algorithms bring value to the industry as well as explain how to control them. 3.2.1.1. Internal meetings The authors had regular internal meetings that were both formal and informal with employees and managers at the host company in order to understand the value proposition of the host company but also to understand the overall value chain in the manufacturing industry. The meetings also served as a way of defining the technological components and gathered as much information as possible in regard to them, but they also helped the authors to investigate internal documents in order to gain knowledge in the area as well as understanding their pain points of the industry in which the authors could later build upon. 3.2.1.2. Identifying literature In order to identify literature and frameworks for the study the authors used the information about the technology gathered from the host company as well as information obtained during the education, through the authors’ own research and from the supervisor at the institution. Three general parts were identified where the first part was the technological concepts behind the industry 4.0 in which the authors pictured a funnel where the authors started wide and went down to the core of what brings value in a service-based internet of things solution for industrial manufacturing companies. The first subject introduced was about the industry 4.0 and what it means for the future of the industry in regard to the change in business models. The second subject focused on Internet of Things by looking at its components, the opportunities and challenges it brings. The third subject is focused on one of the technologies that bring value in an internet of things solution which is data analytics. The last part of the funnel are the algorithms which is the fuel for data analytics in order for it to bring value in an industrial context. The second part was in regard to how to control the algorithms which was found through court cases, regulations and articles which focused on different control mechanisms, categorization of knowledge assets as well as knowledge transfer and the role of knowledge due to the very nature of algorithms. Lastly, in order to understand what defines a competitive advantage in the industrial sector the authors also defined Porter’s five forces model. 12 3.2.2. The empirical approach When identifying the practice, the authors used interviews in order to gather the required empirical data. The reason for this was due to the exploratory nature of the phenomena which called for combining different fields in order to answer the main research question. The empirical approach focused on two different interview subjects where the first type was “IP professionals” who had knowledge around how control mechanisms worked in a practical sense in regard to algorithms. The second type of interview focused on software managers at software companies and industrial companies as the authors’ most important criteria was that the companies were developing algorithms in-house. The software managers were interviewed in order to help answer the second research question on where the trends are currently at in terms of controlling and managing algorithms inside software and industrial companies due to them having a more technical approach as well as more hands-on knowledge around algorithms than IP professionals. 3.2.2.1. Interviews Frameworks that were used for the basis of the interviews can be divided into two different ones: the three-arena model and the intellectual building blocks. For the IP professionals, the interview template was based on the two frameworks which was the three-arena model created by (Petrusson, 2015) in order to understand how IPRs on algorithms would handle the different arenas as well as the framework for intellectual building blocks in order to capture the different control mechanisms. For the template for the software managers, the controlling innovations framework by (Petrusson, 2015) was used once again but from a more technological perspective due to the limited knowledge from the software managers about IP. In addition, the authors used theory about knowledge transfer which is being made through implicit and explicit knowledge in organizations in order to grasp the situation between developers and managers as well as gaining deeper knowledge in specific control mechanisms such as secrecy and technological control. The authors decided to conduct the qualitative interviews in a semi-structured manner due to the explorative approach of the study and the limited research in the field. Through the semi- structured interviews, the authors could focus on a rather specific topic but at the same time leave room for the interviewees to develop their replies (Bryman & Bell, 2015). When constructing the interview templates, the authors had high emphasis on balancing the standardization and structure of the questions in order to keep the interviewees to the subject at hand but at the same time give them leeway (Patel & Davidsson, 2011). In regard to standardization one considers how much responsibility should be left to the interviewer in leading the interview and in regard to structure one has to consider to what extent the interviewee is free to interpret the questions based on their previous experiences (Patel & Davidsson, 2011). Something that was regarded as very important for the authors was the integrity and confidentiality of the interviewees which meant that all the interviewees were 13 asked if the authors were eligible to record them as well as if they would remain anonymous in order for them to be able to answer as freely as possible (Patel & Davidsson, 2011). 3.2.3. Data sample The interviewees were divided into two different groups where group one as previously described consisted of IP Professionals and the second group consisted of software managers at different companies that worked with algorithms in-house. Interviews took place with both of the groups during the same time period as well as during the same amount of time, 30 minutes to an hour face-to-face interview. Overall there were 18 individuals that were interviewed where half of the interviewees were from group one and the other half of them from group two. The number of individuals interviewed were 9 individuals per group based on the reasoning that the time for this part of the research was rather limited. According to (Bryman & Bell, 2015) the part of interviewing, doing the transcription as well as analyzing the transcripts are very time consuming. These two groups served the purpose of bridging the gap between an understanding of how control mechanisms worked in practice in algorithms as well as an understanding of the trends around knowledge transfer and what control mechanisms were used on algorithms. Another purpose that the groups fulfilled was the fact of being able to answer the second part of research question one with the help of insights from IP professionals and being able to answer research question two with the help of software managers. The IP professionals ranged from individuals who were working at senior positions at companies in regard to technology and software to individuals who were working at IP consultancy firms (see appendix 10.2 & 10.3). One mutual factor was that both types of IP professionals had a vast knowledge in dealing with software in regard to intellectual property. Looking at the second group interviewed the software managers were managers in both industrial companies that were working with embedded software as well as managers in pure software companies who were working with digital products in order to understand if there were any differences and remove any type of bias. Two important criteria were established for the software managers to make the quality of the study higher. The first criterion was that the companies that the software managers worked at had to develop their algorithms in-house as IT consultancy companies tend to develop algorithms for others which do not give them the same incentive of having control of the algorithms. The second criterion was that the companies had to be mid-sized to large-sized companies as the authors had the hypothesis that companies of that specific caliber had stronger management techniques of algorithms in place than a small sized company. 14 3.2.4. Data analysis In order to analyze the data gathered from the qualitative interviews from the two interview groups the authors used thematic analysis which is seen as one of the most common strategies when it comes to analyzing qualitative data (Bryman & Bell, 2015). Themes were defined based on the frameworks as well as the literature around knowledge transfer that had been used in order to set the scene and to not get lost in the data collected. The analysis was done separately by the two authors in order to not let personal emotions and values get in the way. When the separate analysis had been made, the authors met and discussed their individual results and created a summary for each of the different interview groups in order to get rid of unnecessary data that did not fit into the research purpose. According to Bryman & Bell (2015) one of the biggest difficulties with qualitative interviews is the fact that it can in a very fast pace generate a large amount of data due to its reliance on prose. It can easily be described as an attractive nuisance due to the attractiveness of its richness but there is a difficulty in finding a good analytical path through the richness. 3.3. Quality of research During a business research the quality is of utmost importance. In quantitative research the most important criteria to look at is the reliability as well as the validity. However, due to qualitative research not being measurable in the same way that quantitative research is, both the reliability and the validity definitions become a little vague (Bryman & Bell, 2015). According to the definition of validity and reliability, there are absolute truths in the social world but according to (Guba & Lincoln, 1994) the social world can exist of more than one possible truth. Therefore, this study will use the framework of credibility, transferability, dependability and confirmability in order to assess the quality of the research. From a credibility point of view (Bryman & Bell, 2015) are talking about whether a result can be seen as believable or not. In order to verify that the results from the explorative study were believable, the authors used the method of triangulation which entails that one has more than one method or source of data when studying and explaining a social phenomenon (Bryman & Bell, 2015). In the case of this particular study, in order to verify the data gathered for answering the first research question, the authors used both literature studies such as court cases and regulations together with interviews with IP professionals. In order to verify the data of the second research question the authors had internal meetings with the IT department at the host company as well as interviews with software managers that were working both in embedded software as well as with digital software. Transferability is a way of checking whether or not the findings that one does in the research hold up in some other context or even the same context at some other time (Bryman & Bell, 2015). The object that is being investigated are algorithms and how to control them in regard to an industrial setting. Here, the authors noticed that there was a difference in how software and industrial companies were distributing and controlling their algorithms which means that the 15 result is transferable to a certain extent to other domains. Due to the explorative approach of the study there is a possibility that the context will look different at a later stage as industrial companies are starting to adopt the same kind of strategies in controlling algorithms as software companies are doing. In regard to dependability one is looking at the trustworthiness of the research (Bryman & Bell, 2015). To ensure that the study remains reliable and trustworthy, the authors have taken measures to protect but also keep both all the theories, court cases, transcriptions and recordings on an encrypted space that is easily accessible with a key. However, due to the confidentiality of the interviews they will not be replicable as the likelihood of the same individuals being interviewed in the same type of study is low. Finally, confirmability is a criterion that recognizes that complete objectivity never can occur in a business research but still encompasses that it is important for researchers to show that they haven’t let personal values gotten in the way to sway the conduct of the research and therefore the findings (Bryman & Bell, 2015). In order to live up to this criterion, the authors made sure to create open questions for both groups that were interviewed in order for the interviewees to give as open answers as possible. At the same time the authors also decided it was important for both of them to be at every interview, so the situation didn’t become colored based on their respective backgrounds. When doing the analysis of the data that was gathered the authors did this separately at first and after that they met in order to reduce the objectivity as much as possible. 16 4. Theoretical Foundation In this chapter, the theoretical concepts that are used to analyze and discuss the research findings are presented. The chapter is built up like a funnel where the underlying concepts that make up for industry 4.0 such as internet of things (4.1), data analytics (4.2), algorithms (4.3) are introduced. After the funnel is presented, the authors mention the theory of the knowledge economy (4.4), control of innovations (4.5) and competitive advantage (4.6) which helps to strengthen up the research in order to answer the main research questions and sub research questions. 4.1. Internet of things 4.1.1. Introduction An important part of industry 4.0 is a technology phenomenon called internet of things which is something that have gathered a lot of traction in several industries. It has been next to impossible not to come across this term (Wortmann and Flüchter, 2015). The core concept of it all is that everyday objects can be equipped with different capabilities such as identifying, sensing, networking and processing (Whitmore, Agarwal and Da Xu, 2014). It proposes to attach technology embedded with software to everyday devices such as different type of sensors in order to gather information from the environment and therefore gain additional functionality such as being integrated to a larger network and being able to communicate with other devices (Whitmore, Agarwal and Da Xu, 2014). 4.1.2. Technology stack of the internet of things The inventions that make up for the internet of things can be divided into core technologies which are the technologies that keeps the solution together as well as the enabling technologies which are used on top of the core technologies in order to deliver value (European patent office, 2017). In regard to the core technologies they can be defined in three different layers where the first layer is called the device layer which focuses on the specific hardware such as sensors, actuators or processors that gets added to the existing hardware components along with embedded software that manages and operates the functionality of the physical thing (Wortmann and Flüchter, 2015). The second layer is the connectivity layer which handles the communication between the device(s) and the cloud with the help of communication protocols (Wortmann and Flüchter, 2015). Lastly there is the cloud layer that serves the purpose of communicating with, provisioning and managing the connected devices through a platform that develops and executes internet of things applications (Wortmann and Flüchter, 2015). 17 The enabling technologies which are previously described is what helps IoT solutions bring value to the customer and differentiate one solution from the other, but they are not necessary for the solution to work. They can be divided into five parts and are located in the IoT cloud (European patent office, 2017). The first part is analytics and data management which is software that stores, processes and analyzes the data generated by the devices. There is the IoT application which is software that coordinates the interaction of people, systems and devices in a specific purpose. Process management focuses on defining, executing and monitoring process around the devices. The application platform provides developers with the tools necessary to further develop the particular IoT solution and add specific functions that ties to their needs (Porter and Heppelmann, 2015). Figure 2 - IoT technology stack (Porter & Heppelmann, 2015) 18 4.1.3. Opportunities for industrial companies With the help of internet of things firms will be able to create new value propositions that they have not been able to do before. It will first and foremost lead to an expansion when it comes to product differentiation as software will allow companies to customize their new smart products to fit their customers’ needs which was not a possibility before when they were merely mechanical (Schmidt et al., 2015). This will serve as a way for industrial companies to provide tailor made solutions which further leads to the enhancement of differentiation and price realization (Porter & Heppelmann, 2015). Firms will be able to form stronger relationships with their customers through the capturing of their rich historical data and product usage data (Porter & Heppelmann, 2014). This data can later be used for gaining competitive advantage in the specific industry domain as it can serve as a training set for training smart algorithms in order for them to learn a certain pattern (Das, 2016). The data can also be used in order to improve the firm’s own physical product when it comes to R&D (Raconteur, 2016). Due to the notion of data being there to train algorithms in order to perform pattern recognition this means that the future solutions become tailormade for a specific business need. This in time reduces the bargaining power of the buyers significantly as the switching costs then become too high (Porter & Heppelmann, 2014). It will also open up new avenues for differentiation and value-added services as well as offering superior performance, customization and customer value relative to traditional substitute products (Porter & Heppelmann, 2015). 4.1.4. Challenges for industrial companies In terms of challenges there is a high probability of new industry entrants. As products are getting more complex by including software they also require a new type of expertise which opens doors for software companies to make a business out of it as well (Porter & Heppelmann, 2014). An example of this is the increasing need for complex algorithms in order to monitor and control different types of smart devices which is a core capability that many traditional industrial companies do not have (Spencer et al., 2004). There will also be a challenge when it comes to the standardization of software in order to reach full compatibility between the technologies in the internet of things stack (Eclipse Foundation, 2016). Due to the new opportunity of being able to turn mechanical devices into smart devices industrial companies might go overboard by adding too many functions and features thus driving up costs as well as reducing profitability (Porter & Heppelmann, 2015). Another challenge is that traditional suppliers that provided physical components will decline as software will play a bigger role than hardware due to the fact that there is a different need as the products are smartly connected. Therefore, strong new suppliers will be part of this new industry by being providers of new value that was not needed in the past such as software and analytics. A key 19 aspect to consider here is that new suppliers can leverage the relationships with the end customer and gain access to product usage data which can provide new services to the suppliers (Porter & Heppelmann, 2015). 4.2. Big data analytics 4.2.1. A new age of analytics One of the biggest enabling technologies when it comes to the internet of things regardless of a particular industry can be regarded as big data analytics which focuses on analyzing all of the data that is picked up from the environment by sensors and delivering some type of output (European patent office, 2017). When defining big data analytics, it can be divided into two main sub processes where the first one is data management which involves the processes and supporting technologies of acquiring and storing data that is ready to be analyzed. The second one is data analytics which refers to techniques used to analyze and acquire certain intelligence from big data (Gandomi and Haider, 2014). During the last five years, sensors have led to an unprecedented rate of data creation and therefore drives the need for data analytics in order to create value (Gandomi and Haider, 2014). Many organizations are now searching and planning for the best way to extract value from their data in order to be able to compete in their particular industry (Levalle et al., 2011). 4.2.2. The process of big data analytics The process of extracting insights that are valuable for the company in big data analytics can be defined as five different components where the first three components are located in the data management process while the last two are located in the analytics process (Gandomi and Haider, 2014). The first one is called acquisition which is focusing on the acquisition of data through different sources such as sensors, text documents, sound or video (Cavanillas, Curry and Wahlster, 2016). In the second component which is the extraction, cleaning and annotation of the data a portion of the collected data is being chosen based on algorithms that have been created to carry out specific instructions (Gandomi and Haider, 2014). The basis of the third component is focused on the integration, aggregation and representation of data which refers to the combination of data from different sources in order to provide the user with a unified view (Lenzarini, 2002). When it comes to the analytics part it firstly consists of modelling & analysis which is focusing on the actual modelling of the infrastructure which will be used to analyze the selected and aggregated data (Buckingham Shum and Deakin Crick, 2012). Lastly, the final component of the process is called interpretation and is focused on how the organization itself is interpreting the data presented by the analytical process (Han et al., 2014). 20 4.2.3. The metrics of data analytics When it comes to data analytics there is a question of what defines a competitive data analytics solution and how to characterize it. Laney (2001) developed a framework that is called the three Vs in order to explain what big data analytics is as well as it’s challenges. The first one is called data velocity which refers to the speed of data processing (Deroos et al., 2011). Data variety is the second measurement in the framework and it focuses on the need to have more compatible data formats in order to create much more effective data management (Laney, 2011). Due to the recent explosion of sensors and smart devices the data in an enterprise have started to become more complex because it doesn’t just include traditional relational data, but also raw semi structured and unstructured data from different sources (Deroos et al., 2011). The final characteristic of data analytics is the volume which refers to a huge volume of data because organizations record anything that they possibly can (Deroos et al., 2011). This means that if organizations don’t manage to handle the massive amount of data coming in, there is a chance of them being overwhelmed by it. However, with the right type of technology and algorithms organizations have the possibility of identifying and extracting the data that is useful for the specific area of application and leave other unnecessary data behind (Deroos et al., 2011). One could say that these three measurements present an opportunity but also an increasing challenge for organizations when they try to become as competitive as possible in their respective markets. Depending on how well the algorithms are constructed they can help with speeding up the data processing which is the velocity, they can help transform data from different sources into compatible and understandable formats which is the variety and lastly, they can find the value in large data volumes which is the volume. 4.3. Algorithms 4.3.1. The idea of algorithms Looking at what an algorithm represents it can be seen as a procedure that takes some type of value or set of values as input. The procedure then uses the input in order to create new value in terms of an output (Cormen et al., 2001). An algorithm usually helps to solve a well- specified computational problem. At the same time the statement of the problem specifies a certain relationship between the input and the output which is the specific computational procedure for achieving that input/output relationship (Cormen et al., 2011). In order to be able to define what counts as a “correct algorithm” the algorithm has to for every input instance produce a correct output. We can then say that it solves the given computational problem (Cormen et al., 2001). As an algorithm is only seen as an instruction it means that in order for an algorithm to be useful it has to be implemented into a certain context. This context is source code which is written in a specific programming language which translates the algorithms into source code. In order for the computer to understand and execute the algorithms into action, the 21 source code gets translated into machine code through a program called a compiler. Together, these three different components make up for the software. Figure 3 - Execution of an algorithm 4.3.2. Algorithms in machine learning It has been previously stated that an algorithm can be characterized as a procedure that takes one or several input values and use these values to create new value in terms of output (Das, 2016). However, one can also train algorithms to learn from their mistakes which is commonly referred to as “machine learning” (Das, 2016). Through the help of machine learning algorithms become better in terms of accuracy and speed which in business terms translates to more accurate and efficient analytics on data. In machine learning one often talks about two separate ways of learning where the first one is called supervised learning and the second one is called unsupervised learning (Michalski, Carbonell and Mitchell, 2013). In supervised learning a system produces outputs which in the case of big data analytics can be viewed as decisions. The input is regarded as historical labeled data based on the specific application area and the system then “learns” the relationship between the input and the output based on machine learning techniques that are made up out of algorithms (Das, 2016). On the other hand, in unsupervised learning one is reorganizing and enhancing the inputs in order to place a certain structure on unlabeled data (Michalski et al., 2013). These two types of learnings are just two different approaches when it comes to machine learning but what is seen as important value when creating a mathematical model when training the algorithms that make up for these two types of learning are the data sets used for the appropriate application area (Quiñonero-Candela, 2009). The reason for this is because in the real world the conditions in which we use the algorithms we train will differ from the conditions in which they were developed (Quiñonero-Candela, 2009). 22 4.3.3. Algorithms in open source In the software community there is something called open source software which involves software developers at many different locations as well as organizations sharing code to develop and refine software on different kind of terms (Lerner and Tirole, 2003). From a business model perspective there is not so much written about open source but at the same time there are rather strong social norms and legal protections that have been crafted to discourage people from profiteering on the work of their peers (Chesbrough, 2006). With algorithms there is no difference, here there are several basic open source toolkits that can be downloaded in order to use, train and build upon already existing algorithms. However, in order for them to work properly for a specific use case it involves a lot of knowledge first of all from the industry specific domain they are to be implemented in but also knowledge about how to tweak them in order to get the right result for the right customer which is where knowledge about a specific domain comes into play. 4.3.4. Algorithms in an industrial perspective There has been a change when it comes to the domain of algorithms. Before, algorithms were used only for pure software products. However, due to growth of embedded software i.e. software that is embedded into hardware products the field of algorithms are becoming more interdisciplinary based (Das, 2016). This means that different types of knowledge such as computer science, business, economics, statistics and most of all it is important to as a company be armed with the necessary quantity of the specific domain knowledge in order to be able to ask the right questions to get the most value out of the constructed algorithms (Das, 2016). In the end, it is important to understand that the data in itself cannot bring value. Without algorithms that are created with the help of a knowledge specific domain to analyze the data it will remain unstructured and therefore hard to understand (Das, 2016). 23 4.4. Knowledge economy 4.4.1. Introduction to knowledge The term of knowledge-based economy gives economic growth when technology and knowledge are identified (OECD, 1997). The knowledge-based economy means an increment of knowledge intensive jobs and a decisive factor is an economic growth of information in information sectors. In addition, intangible capital assets are becoming more valuable than tangible capital assets. These improvements in the economy are expressed as a fast expansion of job creation when accessing knowledge through training, education and transfer of information and knowledge. This new economy generated is expanded to the whole economy and not only to communication services, information and technology sectors (Foray, 2006). This means that specific economic characteristics of knowledge are created. Knowledge in itself is difficult to control due to it being like fluid which entails that it can easily leak out in multiple ways. It can become non-rival which means that the creator of knowledge does not have to produce more units for reaping the benefits which makes it infinite. Knowledge in itself is also cumulative which means that it can be built upon in order to spawn new goods and services (Foray, 2006). With the emergence of new technologies based in information and communication technology (ICT), many businesses are moving towards a knowledge-based economy. This new type of economy is based on reshaping businesses to generate financial value by utilizing intellectual property rights (IPRs). Therefore, new business models are created, and industrial companies are reshaping their businesses (Petrusson and Heiden, 2009). In an industrial economy, generating and extracting has been done through production, distribution and sales of physical goods (Petrusson and Heiden, 2009). To create financial value to the firm, a material value chain has captured raw material, production, distribution, retailing and aftermarket services by a hierarchical structure. Material value chains can be based on vertical integration or horizontal integration. On the one hand, vertical integration happens when the firm controls several steps on the material value chain. On the other hand, horizontal integration occurs when the firm has control on one of the steps of the material value chain in relation to several products (Petrusson, 2004). Figure 4 - The material value chain (Petrusson, 2004) 24 In an intellectualized economy, the value is created through the management of assets, capital and property. Assets are seen as valuable objects, property is seen as objects that can be commercially transacted and capital is seen as an object that creates financial value (Petrusson and Heiden, 2009). Therefore, an intellectual value chain is characterized for intellectual structures by the usage of intellectual building blocks based not only on offering physical products and services but also by offering virtual products and licenses. The most important aspect in an intellectual value chain is the transaction of the object that has been claimed intellectually (Petrusson, 2004). In this new economy it is crucial for a firm to be able to create structural capital based on human capital that can be transformed into different value propositions such as physical products, virtual products, license offers or services. If a firm cannot do this the risk exists that it will become a knowledge provider to other firms (Petrusson, 2004). Figure 5 - Intellectual value chain (Petrusson, 2004) 4.4.2. Knowledge transfer With the reproduction of knowledge, a problem arises in relation to technology transfer and transmission of educational and occupational skills. One type of knowledge can be tacit which refers to knowledge that is not possible to be expressed externally from the person who possess it. Therefore, this type of knowledge is difficult to transfer and reproduce. An example of this is when an employee is leaving a company due to retirement and the employee transfer his/her know-how in order to ensure that this knowledge will be transmitted to the following generations. However, due to the increment of external factors, turnover and mobility; the methods of knowledge management are affecting internal labor markets. Therefore, new ways of knowledge retention and transmission are used, and knowledge codification plays a major role here as it can be recorded and expressed in a specific language (Foray, 2006). Another type of knowledge is called explicit knowledge which is characterized by its ease to communicate and transfer (Grant, 1996). 25 This type of knowledge can be seen as simple instructions which is easily transferable from one person to another through the help of a document. Knowledge is difficult to control by firms and therefore, leaks and spillovers can occur. Involuntary spillovers lead to market competition and therefore, knowledge can be used as way of creating incentives and increment performance through imitation, use and absorption of new knowledge. An example of this is when knowledge is a codified instruction, such as software, and therefore it is almost impossible to control it as practitioners and specialists are able to understand the instructions. Nonetheless, knowledge is not only based on codification, it is also based on tacit knowledge in which practical experience is required. This practical experience can give a level of control if the know-how is kept as a secret to avoid spillovers. Additionally, controlling knowledge can be done through having complementary assets that can be specific capacities to utilize this knowledge such as technological capacities (Foray, 2006). 4.4.3. Organizations and markets built on knowledge Knowledge is also changing the way that both firms and markets are operating. Before, companies and markets were only operating on a resource-based level. Now however, they also have the opportunity to operate on a knowledge-based level. When discussing the difference between a resource-based focus of a firm and the knowledge-based focus of a firm one can see that the primary objective for a resource-based firm is to maximize the value through optimal deployment of existing resources and capabilities while developing the firm’s resource base for the future (Grant, 1996). An example of these kind of resources could be regarded as manufacturing in order to be able to transfer goods to the customer in order to gain revenue (Grant, 1996). From a market perspective this meant that the main exchange medium was tangible goods in terms of products (Arora, Fosfuri & Gambardella, 2000). This meant that what gave a competitive advantage was the fact that a company had great financial capital which was gained through efficient productivity which was measured per employee in the company (Bell, 1976). A typical model that was used in this kind of thinking was for example the well-known value chain created by Michael Porter in which one focused on primary and secondary activities within the firm where both were focused towards productivity in itself and producing physical resources (Porter, 1991). The problem with this kind of thinking today is the fact that these kinds of resources that can be exemplified into raw materials are finite which means that they are not unlimited (Vargo & Lusch, 2004). Turning to the knowledge-based focus of a firm it is also based on transferability. However, not only from an external aspect but it is also concerned with the transferability of knowledge internally. The main aim of the firm is to master the codification of knowledge in order to get a wide diffusion of knowledge through the firm (Bell, 1976). To be able to set up a wide diffusion of knowledge inside of a firm one needs proper coordination within the firm, a clear organizational structure, a spread out allocation of decision making rights, clear definition of firm’s different boundaries and a good innovation structure in order to turn the knowledge into value (Grant, 1996). From a market perspective, the knowledge-based focus opens up the door 26 for smaller actors as well because they don’t need to have substantial resources in order to make a profit because one can license their services and technology to companies that are more capable of manufacturing and commercializing the final product (Arora, Fosfuri & Gambardella, 2000). This affects the competitive advantage because companies can now solely rely on immaterial products and services that can be quite hard for competitors to imitate due to different reasons such as the fact that this kind of market is relying very strongly in the knowledge that resides inside of human capital which can be seen as operant resource in order to produce value (Constantin & Lusch, 1994). 4.5. Controlling of innovations 4.5.1. IAM framework When working in collaborations in a knowledge-based intensive setting, such as an IT environment which is characterized by rapid growth and change in regard to algorithms in software it is important to understand what type of assets are created both internally and in collaborations. In order to handle this type of situation, in where one has the need to define what can be claimed and how to leverage it, (Petrusson, 2015) created the IAM framework which is divided into four steps. The first step - concerned with how to construct a proper process for claiming the intellectual assets. The second step - concerned with the process of evaluating and positioning oneself towards the outside world. The third step - concerned with making decisions in regard to the efforts of utilization and how to implement them. The fourth step - concerned with how to manage the defined intellectual assets, intellectual property rights as well as contracts within the organization as well as with the environment. Figure 6 - IAM framework (Petrusson, 2015) 27 4.5.1.1. Categorization of intellectual assets When it comes to the first step of the IAM framework, related to claiming knowledge assets, it follows a process where the knowledge assets need to be identified, claimed and analyzed first. Once this process is done then it is possible to claim them as property. This process starts with “the cloud” which is based on diffuse and undefined knowledge that is later turned into a clear and manageable intellectual asset list. The following figure 7 shows an example of how it is possible to turn diffuse knowledge into knowledge assets (Petrusson, 2015). Figure 7 - Categories of knowledge assets (Petrusson, 2015) Knowledge assets create value for society and therefore, a tool of categorizing knowledge assets have been created to capture the value of these assets. The tool captures the following knowledge assets (Petrusson, 2015): Data - consists of unstructured information or raw data and it can be gathered through surveys, statistical measurements and interviews. The data as a knowledge asset is possible to control as the data can be documented and collected. Databases - are structured and searchable data that is organized in a systematic manner. Observations - are conclusions or correlations based on empirical data collected. In order to be controlled and objectified, the observations should be concrete and clear. Theoretical framework - are based on theories or models that are based on results and relationships. These types of frameworks can be utilized to create new assets an example of this is the creation of software and procedures as implementation of theory. Solution - solves problems that derives from scientific research results. Technical solutions and inventions are categorized as solutions. Visualization - comprises design elements such as drawings, sketches and prototypes which add value. 28 Instructions - provides advice on how certain activities need to be performed. Instructions can be carry out by people and machines such as algorithms which are implemented in software. Software - is a set of systematized and automated data that executes certain activities such as being a software for implementing algorithms. This category can be divided into different subcategories where layers and code modules play an important role. Narrative - is a particular story that can be materialized in interviews, studies, literature. Creation - is artistic creativity connected to arts such as music, design, painting and so on. Algorithms as knowledge assets can be categorized as a solution, instruction and software. By applying these categories on the new assets that are created they can more easily be claimed. These knowledge assets can now be turned into intellectual property assets. It is important to understand the interplay between knowledge assets as value creation objects and intellectual property assets as legal objects. Examples of these intellectual property assets can be inventions, designs, trademark, trade secret and artistic and literary work (Petrusson, 2015). 4.5.2. Intellectual building blocks Intellectual building blocks also referred to as intellectual building bricks are intellectual constructions that can be used as value propositions, as personification of innovation and identity in order to get financial value. The means of property claims as structural building blocks are: technical control, market power, secrecy, right based property and contracts-based property (Petrusson, 2004). Figure 8 - Intellectual building blocks (Petrusson, 2004) 29 4.5.2.1. Rights based property Intellectual property rights are the rights given to the creator or inventor of an intellectual property asset. The intellectual property rights are patents, copyright, trademarks and design rights (WTO, 2018). In an intellectualized economy, the core of the businesses is based on assets, capital and property. Therefore, businesses are utilizing intellectual property rights (IPRs) to control the innovations and extract financial value. (Petrusson and Heiden, 2009). Patents A patent is a document that describes an invention that is issued by a government office after filing an application and it gives an exclusive right to the inventor to use the invention for a limited amount of time, normally 20 years. When an individual has obtained a patent, it grants the individual with the right to exclude others from making, selling and distributing the invention as long as the right holder pays their yearly maintenance fee (WIPO, 2004). In addition, a patent may relate to a product, process (method) or use (WIPO, 2004). In order to be eligible for patent protection the invention needs to be patentable subject matter. In general terms, patent protection is available for inventions that are related to all fields of technology. In order to be entitled for patent protection, the invention needs to fulfil three separate criteria (TRIPS, 1995): 1. Novelty (new) - An invention has to be seen as new in relation to what was known before the filing date 2. Industrial applicable (utility) - The invention must be able to be susceptible of industrial application 3. Inventive step (non-obviousness) - The invention must have a certain leap of knowledge in terms of what is known In numerous countries software-related inventions can be patentable subject matter if these inventions have a technical effect (TRIPS, 1995). The software-related invention can reside as an algorithm (WIPO.int). When drafting a patent application in relation to computer software that contains specialized algorithms, its claims normally comprises apparatus and method claims which in itself is tied to a technical solution (WIPO, 2007). Moreover, the description of the patent application related to an invention in the field of computer could be potentially accompanied by flow diagrams to be understood by a person skilled in the art (PCT, 2018). In US, algorithms are mathematical concepts and therefore, they are considered as abstract ideas by the courts which are a judicial exception for the protection for software-related inventions (USPTO, 2012). Currently, the US patent system has strengthened up after the Alice Corp v CLS Bank Itnl decision in relation to algorithms (USPTO, 2012). Before, the Alice decision, two Supreme Court cases (Bilski v. Kappos and Diamond v. Diehr) were declared invalid due to its subject matter being considered abstract. In the Alice case, there was a company called Alice Corporation that owned four patents related to computer programs and 30 electronic methods for financial trading. They noted that a company called CLS was using similar technology. Therefore, Alice decided to accuse CLS for patent infringement. Nonetheless, CLS filed a suit against Alice stating that the claims were invalid. In 2014, the US Supreme Court declared the patents of the Alice corporation invalid due to the fact that the claims were related to an abstract idea and in order to be considered as a patent eligible subject matter, the claims implemented in the computer needed to ensure that amounted significantly more than the actual abstract idea. At the moment, the Patent Trial and Appeal Board (PTAB) in US is invalidating software patents that have been granted in the past after Alice decision because as the claims of these patents do not comprise significantly more that the abstract idea as an exception (USPTO, 2012). After the Alice case, the Patent and Trademark Office (PTO) established some guidelines for applicants in order to apply for a patent in relation to a computer related invention (See appendix 11.1.1). In Europe, a computer-implemented invention refers to the ambiguous term of software. The board of Appeal of European Patent Office (EPO) stated that a computer algorithm can be seen as a purely mathematical exercise or as a procedure that explains how a machine is executing some tasks. An algorithm that defines a procedure always has a technical effect. Therefore, in order for an algorithm to be considered for patent protection in Europe, the board stated that it needs to have a “further technical effect” in order to have a technical character (Enlarged Board of Appeal, 2010). In addition, a computer programme as such is not a patentable invention (EPC, 2016). For a patent to be granted for a computer-implemented invention, it needs to solve a technical problem. Moreover, an algorithm is defined as a “procedure or sequence of actions” that can be considered as a patentable invention when it is executed by a computer and provide a technical contribution (Directive 2009/24/EC). The EPO also provides guidance of patentability in relation to computer implemented inventions (See appendix 11.1.2). One practical problem in relation to patentability of computer-implemented inventions is that there is no harmonized patent system in Europe. Nonetheless, Europe is striving for having a Unified Patent Court (UPC) that is supposed to come into force this year in which European patents will have a unitary effect. However, this court will not have any jurisdiction in relation to national patents. The national patents will be enforced through the patent office of each EU member state (UPC, 2018). In comparison, US has only one patent system that is applied by the Patent and Trademark Office (PTO) and US courts. In Sweden, a merely program for a computer is not an invention (Swedish Patent Act, 1967). The Swedish Patent and Registration Office (PRV), mentions that a computer program can have different meanings and one of them is to be considered as a “method for solving a mathematical or logical problem, an algorithm”. An algorithm can be patentable if it is used in a technical context and has a technical solution (PRV, 2018). 31 Copyright Copyright protects the rights of the creators when it comes to the expression of ideas but not the ideas as such. In addition, copyright relates to literary and artistic creations and gives the owner exclusive property rights which prevents unauthorized use. There is no need for registration in order to obtain copyright protection, when the work is created it is automatically protected. The duration of copyright protection depends on national law, but it cannot be less than 50 years after the author’s death. Moreover, copyright protects two types of rights; moral rights which protect and conserve authors’ and creators’ connection with the work and economic rights that permit the authors and creators to obtain financial reward by others for using their work. Authors can transfer their economical rights to third parties and in return receive some payment. One can transfer the economic rights in two ways: either by assignment in which the copyright property rights are transferred and the person assigned becomes the right holder or by having a license as in some countries assignment is not allowed. By licensing copyright, the owner will still have ownership but will authorize a third party to execute some activities for a period of time (WIPO, 2016). Copyright does not protect methods, mathematical concepts or procedures and therefore it does not protect an algorithm per se. Nonetheless, copyright protects computer programs as literary work in any mode of expression (WCT, 1996). Computer programs, either in source or object code, are protected as literary works under the Berne Convention (TRIPS, 1995). In Europe, the legal protection of a computer program states that only the expression of a computer program is protected, and the ideas and principles are not protected by copyright. According to this principle, algorithms per se comprises ideas and therefore they are not protected under the European Directive (Directive 2009/24/EC). However, the copyright holder can prevent an unauthorized person from copying the code that implements an algorithm. In addition, copyright might not prevent the protection of others that implements the algorithm by using different code (Digital Single Market, 2016). In Sweden, computer programs are protected by copyright. Moreover, the copyright that is created by an employee related to a computer program as his working tasks or by following instructions by the employer is considering to be transferred to the employer. Otherwise, it will depend on what the parties agreed on in the contract (Act on Copyright in Literary and Artistic Work, 1960). Open source software is software that it is made available to users and it is free due to the fact that it gives to the user freedoms. Open source software is protected through copyright and the user is subject to an open source license in which the person is able to copy, modify and redistribute the code under certain conditions. The most common open source licenses are General Public License (GPL) and Berkeley Software Distribution (BSD) license (WIPO, 2018). 32 Trademarks A trademark is a sign that characterizes the goods of a company to differentiate from competitors. There are different types and categories of signs such as: word marks, device marks, colored marks, audible marks, etc. The general criterion of protection is that the function of the trademark needs to be distinguished from other products or services and therefore it needs to be distinctive. The other requirement is the potential harmful effects when the trademark is misleading and infringes the public authority or morality. A trademark can be protected by use or by registration and the protection can be unlimited in time if the renewal fees are paid to the administration offices (WIPO, 2004). Algorithms cannot be protected through trademarks directly. Nonetheless, it is possible to obtain trademark protection for software or computer programs that contain the algorithms. The international trademark classification states that all computer programs and software can be trademark protected (WIPO.int, 2018). Therefore, a trademark can protect the name of a company that develops algorithms or a product name where the algorithms form part of it. By having a trademark, the owner gets an exclusive right and by this it can prevent competitors from using it (WIPO, 2004). Design rights Design rights protects the appearance and non-functional features of a product. To be granted a design protection, it needs to be novel which is sometimes referred to as original (WIPO, 2004). The Council Regulation on Community designs protects the design rights if the product fulfil the criteria of novelty and individual character. Designs can have registered or unregistered protection through the Community designs (WIPO, 2012). In relation to computer programs, in Europe, design rights are protected when the program includes preparatory design work (Directive 2009/24/EC). 4.5.2.2. Secrecy The first international agreement that protected trade secrets was the TRIPS Agreement in which the protection for undisclosed information was established. Trade secrets have a dual nature, they are confidential as they remain non-public and only known by a restricted number of people. In addition, they are commercial as they are shared with a limited number of people to have a practical value. The requirements for trade secrets are secrecy, commercial value and reasonable efforts to maintain them secret (OECD, 2015). Trade secrets give a strong competitive advantage in the marketplace by keeping information confidential and in most countries, a third party is prohibited by law to use or copy confidential or secret information in the absence of the owner's consent (Sheikh, 2018). In Europe, a harmonized Trade Secret Directive will take place in 2018 which will improve trade secret protection (Directive 2016/943). 33 To keep the confidentiality of a trade secret, employees in a company should sign non- disclosure agreements and non-compete clauses on their employment agreements. However, this is not enough to prevent the disclosure of confidential information once the employees have left the company. Promoting employee loyalty is a key element to protect trade secrets as a business strategy. In addition, once the employment is terminated, an exit interview is indispensable in which the employer will mention the obligations to maintain confidentiality after the employment and the consequences of breaching the obligations (Sheikh, 2018). A complementary way for obtaining protection of algorithms is through trade secrets. Trade secrets is an effective way of controlling algorithms, same as having patents or even more important than having a patent but this will depend on the circumstances and the type of strategy that you want to apply on your business. Normally companies tend to use trade secrets in order to hide relevant information. In this case, with algorithms, it can be seen as a way to leverage their competitive advantage. An example of this is Google search engine algorithm which is kept as a trade secret (Trade secrets: the hidden IP right, 2017). The scope of the trade secrets is unlimited compared to patents. Nonetheless, legal challenges arise in different jurisdictions as the regulation of trade secrets differ from country to country. In jurisdictions where there is no legislation about trade secrets, contractual control is a way to regulate misappropriation of trade secrets. In US, the policy of trade secrets may include a formula, program, device, method or process and it needs to enable an economic advantage to be considered as a trade secret (Trade secrets: the hidden IP right, 2017). The Defend Trade Secrets Act (DTSA) states that source code, algorithms, data sets and programs can be considered as trade secrets. In regard to algorithms, companies should make sure that they use complementary measures in order to keep the secrets safe such as physical protection that may include security networks, having documented information and by having contractual provisions (Prange, 2017). One measure to keep algorithms as a trade secret in a safe way is through an information security management standard, known as ISO 27001, with the main characteristic of managing sensitive information from a company (ISO, 2018). 4.5.2.3. Contracts based property A contract is a tool used in the construction of a structural order with the main goal to create and extract financial value (Petrusson, 2004). The way in which firms utilize contracts is to legitimize their relationships with different type of actors in the social world. With the help of contracts, one can set up legal obligations between two or several parties to make sure that a breach by any of the parties will result in some type of remuneration for the affected parties. Traditionally, the legal approach used was a reactive approach meaning that when a person was encountering a legal dispute, the way to move forward was to turn to a lawyer. Nonetheless due to digitalization, new types of contracts have arisen and as of the fast IT development, contracts need to be interpreted not only by lawyers but also by engineers and managers. Currently, the 34 legal approach used is proactive law which consists of a combination of legal thinking, skills, practices and procedures that enable the recognition of different opportunities in good time in order to take an advantage from them. Furthermore, proactive law can act as a preventive action by identifying potential problems and it is also seen as a way of generating value, risk management and build up relationships by avoiding disputes and litigation (Haapio, 2006). Proactive law has changed the way of looking at contracts, rather than being just a simple legal tool, contracts are seen as a management tool that enable value creation and collaboration to create strategies for competitive advantage (Siedel and Haapio, 2010). A lawyer tends to take into account the different elements of a contract, regulations and contractual clauses. Nonetheless, in business practices contracts are seen as a tool to achieve a successful deal and relationship (Haapio, 2006). The figures below show a clear distinction between a lawyer’s view of a contract and from a business view. Figure 9 describes the elements of a contract that should be taken into account by a lawyer, these ones are divided between visible (what is agreed) and invisible terms (what is not agreed). Figure 10 states the functions of a corporate contract from a business perspective (Haapio, 2006). Figure 9 - Elements of contract from a lawyer’s view Figure 10 - Functions of a contract from a (Siedel and Haapio, 2010). business perspective (Siedel and Haapio, 2010). To achieve a successful business deal, business managers and lawyers should work together in order to establish different tools to help with the contracts such as checklists, templates and processes that will be helpful for the employees in the company to use them systematically and revised contracts continuously. The challenges of the managers should be addressed when drafting a contract and to help with that one can see the contract as an analogy of a puzzle that can reflect the goals of each party. Figure 11 shows the contractual terms of the puzzle analogy. The contract in this puzzle is a combination of technical, implementation, business /financial and legal parts that needs to be coordinated together. To create a successful business deal in which all the parties are synchronized, the puzzle should be assembled together. If a firm manages to understand how the pieces fit together the firm will be able to achieve contract literacy which means that it will understand what is agreed on in a contract and therefore avoids legal gaps (Haapio, 2006). 35 Figure 11 - Contractual terms- puzzle analogy (Siedel and Haapio, 2010). In order to protect the algorithms through contractual means it is key to understand what is going to be the subject matter that will be licensed. Will the algorithm be regarded as an instruction, will it be regarded as software or will it be regarded as a technical solution? It is crucial that the subject matter is precisely defined. In addition, it is important to understand the nature of the IPRs before entering into a contract and this can be done through patent protection, copyright, trademarks and trade secrets (Cameron and Borenstein, 2003). If the algorithm is regarded as an instruction it can be protected as a trade secret and therefore from a contractual point of view it can be protected through non-disclosure agreements (NDAs) which keep confidential information secret to third parties (European IPR desk, 2015). If the algorithm is implemented into software, the owner will get copyright protection and the software can be licensed to others (Bond, 2004). It will be protected through copyright which makes it eligible for software license. The three main license agreements in software relevant for the study are user license agreements in which the user of the software has a right to use the software, joint venture license agreement is done when two or more parties want to develop new technology together and the last one is open source software which is a royalty-free license based on certain terms and conditions (Bond, 2004). When the algorithm is defined as a solution it is eligible for patent protection as long as it is tied to a technical function which means that one can grant an exclusive or non-exclusive license to another party for using the same solution (Bond, 2004). Internally, a company can regulate contracts with their employees by having employment agreements that contains confidentiality clauses in which the employees cannot disclose any confidential information which can be regarded as algorithms during and after his/her employment and if the confidential information is disclosed it will comprise a breach of contract (WIPO, 2004). 36 4.5.2.4. Technical control Technical control can be regarded as a way of restricting access to different assets through technological means. In an intellectua