Master Thesis
Artificial Intelligence: Exploring Why and How
Companies Decide to Adopt
Name: Malte Torliene Student number:11371013 Date: 23.06.2017 Version: Final Qualification: MSc. Business Administration- Digital Business Institution: ABS Supervisor: Prof. Dr. Hans P. Borgman
Statement of originality
This document is written by Malte Torliene who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Abstract
This research is designed to understand the impact of several factors on the adoption decision of artificial intelligence (AI) technologies. After AI established its significance across
organizational functions and industries, propositions derived from theory and practice will be tested during case studies. Important decision-making factors will be identified and analyzed in order to test the goodness of fit of two technology adoption frameworks for AI. Eventually, a new framework, according to AI requirements, has been crafted. Also, it has to be
understood how organizations decide to adopt AI, which factors play a role and which processes are affected. During interviews conducted in multiple case companies, findings show that companies adopt AI to augment humans in planning activities or fully automate certain tasks. The overall goal is to increase productivity or achieve a differentiation advantage. The newly crafted framework to explain how adoption decision are made describes AI adoption as a process rather than a product, based on factors from the TOE framework. Additionally, perception of the decision-making unit has been found to influence the outcome of the decision-making process and therefore been included in the framework.
Table of Contents
Introduction and Research Question ... 1
Literature Review ... 5 Artificial Intelligence ... 5 Technology Adoption ... 7 Innovation-Decision Process ... 7 TOE Framework ... 9 Conceptual Framework ... 15 Methodology... 15 Research Design ... 15 Case Description ... 19 ING Bank ... 19 Twentsche Kabelfabriek BV ... 19 MzB GmbH ... 20 Accenture Digital ... 20 Operationalization ... 20 Analysis Strategy ... 21 Results... 23 ING Bank ... 23 Twentsche Kabelfabriek BV... 26 MzB GmbH ... 29 Accenture Digital ... 31 Discussion ... 34 AI Adoption Decision ... 35
Goodness of Fit of Technology Adoption Models ... 40
Limitations ... 43
Future Research ... 44
Conclusion ... 45
References ... 47
Introduction and Research Question
Artificial Intelligence (AI), probably the most anticipated technology of the time, develops at a rate at which it becomes more and more exploitable for businesses. “Artificial intelligence is technology that appears to emulate human performance typically by learning, coming to its own conclusions, appearing to understand complex content, engaging in natural dialogs with people, enhancing human cognitive performance (also known as cognitive computing) or replacing people on execution of nonroutine tasks” (Gartner, 2017). Therefore, among
practitioners, AI is perceived as a new “virtual workforce”, that can completely automate and take over certain tasks. However, “a significant part of the economic growth from AI will come not from replacing existing labor and capital, but in enabling them to be used much more effectively” (Accenture, 2016, p.13). Nevertheless, AI has the ability to perform tasks previously dominated by humans, with a much lower rate of errors. Additionally, linking AI to huge knowledge databases, it is able to retrieve knowledge faster and more precise than any human brain could. These facts among other characteristics make AI as a technology
inherently different from any technology before (P1). Many industry reports talk about how AI may impact businesses and their productivity levels; however, it being so different requires a new methodology for adopting it, first (Digital/McKinsey, 2017). Not only changing
dimensions increase difficulties in using the technology, but also organizations are struggling to find the right business case for an investment or have difficulties hiring people with
required skills according to a study by Belatrix Software (2016). Also, an article about machine learning adoption claims, many organizations are not aware the technology is ready to be implemented in day to day operations and that information available is too difficult to understand (Kendall, 2017).
Resulting from the previously explained relevance, it is unknown which organizational functions and more detailed, which activities specifically benefit from AI. Classification have
to be made between routine- or expert-, client- or non-client-facing activities, and primary- or secondary functions to explore and understand respective decision-making priorities.
Investigating if customers will profit from more customized products or if AI will cut costs in the first place will determine sectors of application. Further, also the expected (financial) benefit per function and activity should be evaluated. It is suggested that if an organization expects high (financial) benefits from AI, it is more likely to decide on implementing it (Cao, Jones & Sheng, 2014) (P2). Along those lines also the perception of AI’s ability to increase productivity and therefore, decrease cost or required time/labor may play a role. Firms
assuming higher levels of productivity are expected implementing AI technologies faster than those who do not (P3). These expectations of relative advantage will only be measureable by obtaining the opinion of mangers involved in the process of those kinds of
technology/innovation implementation decisions. Although AI is not expected to replace human labor in large scales, the impact on internal and external labor of an organization utilizing AI technologies remains to be understood. However, supplier requirements may be re-defined through AI delivering functionality that has before been acquired through third parties but can now be kept in-house. Firms expecting to replace expensive human labor and third-party services through a one-time investment in AI are anticipated to adopt with a higher chance (P4). Asking involved stakeholders for reasons of the implementation, what
expectations they have or other factors according to theoretical models (presented later) could unveil important decision-making factors. Stakeholders may be shareholders who decide on making an investment, the IT department implementing the technology, employees carrying out impacted activities, or anybody else affected by the change. Summarizing, this study aims at investigating factors motivating companies to adopt AI. Based on practical implications, the first research question is: Why do companies decide to adopt AI?
Early research already has shown that AI increases productivity of organizations by reducing the required time of mainly planning activities (Eom & Karathanos, 1996). An article by
Hokey Min in 2010 evaluated pioneering efforts of AI in supply chain management, stating that the technology proved to be useful for improving supply chain decisions. Not only was AI perceived as being useful in supply chain management, also product output has shown to increase while costs were decreased in the study conducted by Eom and Karathanos in 1997. Summing up, AI already established its significance in two primary organizational functions. Besides the relevance to organizational functionality, studies have also shown AI’s influence across different industries. Especially industries where large amounts of data are analyzed, such as healthcare, agriculture or finance, have proven to be affected by AI technologies. In general, the field of AI research is wide spread across industries and different goals of application. Rarely it is being generalized, making it impossible to draw conclusions on the general AI population and assess implications across industries or fields of applications. Rather, individual studies investigate unique purposes within distinctive settings and measure a single effect out of a large variety of possible outcomes. Other studies qualitatively evaluate programming techniques or industry procedures (Kudo, Akitomi, & Moriwaki, 2015; Fethi & Pasiouras, 2010; Capone et al., 2015). Anyhow, most AI research misses a business
application context and rather focuses on technical and programming benefits and
components, creating the need for further exploration of business literature. More specifically, literature exploring the numerous effects on an organization’s decision-making process to adopt AI is lacking. “It should be noted that, although the adoption process consists of different stages, most studies focus on the dichotomous adoption/ non-adoption decision. Hence, we know little about the effect of different factors on various stages of the adoption process (Olshavsky & Spreng, 1996). Also, studies tend to focus on only one or a small number of factors” (Frambach & Schillewaert, 2002, p.165). Firms are struggling to make the decision to adopt AI due to missing or complex information on where and how to apply it (Kendall, 2017). Further, compared to previous technologies, AI may not necessarily be designed to serve a specific purpose. The most advanced form of AI can serve any task since
it is being built for “general purpose”. As a result, the technology lacks clarity as to where it should be applied. Hence, the process of finding a suitable business case for AI is much harder compared to previous technologies. Additionally, the decision to adopt might be related to laying off human labor. Depending on company culture or even personality traits of decision-makers, resistance may influence an adoption decision-making process towards rejection (P5). “It is important to examine the acceptance of innovations within organizations because, if there is no acceptance among the target group, the desired consequences cannot be realized and the organization may eventually discontinue the intended adoption” (Frambach &Schillewaert, 2002).
As for academic implications, a comparison article will be built. In order to explain the effects of multiple factors when deciding to adopt AI within an organization, multiple technology adoption models can be used. In this study, a factor model will be compared to a sequential model. As a factor model, the TOE framework by Tornatzky and Fleischer (1990) will be applied. It gives a representation of which factors influence the decision to adopt a
technology. Additionally, the innovation decision process model by Rogers (1983) will be employed, describing a sequence of stages within the process of making an adoption decision. This study will compare and test these models on their goodness of fit, potentially resulting in the design of a new, hybrid framework describing which factors influence the decision to adopt AI at which stage of the decision process. Therefore, the second research question is as follows: How well do current theoretical frameworks predict the decision to adopt?
The remainder of this paper is structured in six sections. In the first section, relevant literature will be reviewed and propositions will be formulated. Derived from theoretical shortcoming, a new conceptual model will be proposed. Afterwards, the methodology section covers the research design, data collection methods, operationalization and the analysis strategy. In the fourth section the results will be described, followed by the discussion of findings. Finally, findings will be related to the overall research question.
Literature Review
During this research two streams of literature are particularly important in order to fully cover the research question. First, an overview of factors motivating companies to adopt AI has to be built. Based on innovative functions and newfangled areas of application AI features, a set of novel aspects emerged which will be discussed during the first part of this review.
Secondly, due to the distinctive nature of AI compared to previous technologies, different factors may influence the decision to adopt. Building upon this premise, two established technology adoption frameworks will be analyzed upon their fitness concerning AI. This analysis is focused on organizational processes rather than individual adoption. Due to the early stage of organizational AI adoption, the importance of investigating motives of adoption overweighs the need for understanding how individuals perceive and react to the technology. Without a doubt, as the adoption progresses, it becomes vital to also understand the impact on individuals.
Artificial Intelligence
Major books in the field of AI classify the technology according to its goals in application. Reoccurring categorizations are: automated reasoning and theorem proving, expert systems, natural language understanding and semantics, modeling human performance, planning and robotics and machine learning. Further categories have been formed around computational methods such as neural nets and genetic algorithms, languages and environment for AI and game playing (Luger & Stubblefield, 2004). Those classifications according to goals will be used to assess a company’s motivation to adopt AI and which purpose it is meant to serve. However, groupings according to computational methods will be disregarded due to business related nature of this research.
Industry reports further narrow down this goal oriented approach into groupings of AI’s impact and expected benefits. According to a report by Belatrix Software (2016), AI can improve operational efficiency, automate business processes, more quickly respond to market and customer changes and desires, improve sharing information of insights and information throughout an organization and lower security risks and improve responses to security threats. Evaluating these expected benefits will further specify the reasoning behind AI adoption. Taking a more traditional approach, one might consider Porter’s (1985) generic strategies distinguishing between achieving a cost leadership or differentiation advantage as well. Porter claims that a company’s competitive advantage is gained by one of two strategies. The first strategy is providing the lowest cost to the customer in a certain market to attract price-sensitive customers. This can be achieved by high asset utilization, low operating costs and control over the supply chain (Wright, 1987). The second strategy proposed is the
differentiation strategy. Here, a company’s competitive advantage is derived by providing the customer with a unique product. Prices may be high using this approach; however, the
company manufactures a product that is tailored to a customer’s need. Fundamentally
differentiating between aiming at decreasing costs or delivering a unique product may reveal the nature of AI adoption.
Continuing with Porter (1985), not only can his generic strategies be used to identify the motivation of AI adoption, using his model of the value chain may reveal in which organizational function the technology is primarily being utilized and how it impacts the overall business process. Porter’s value chain is a set of activities performed by a company. Activities are categorized in either primary activities or support activities. The objective of this process model is to show the added value to the product of each primary activity while considering supporting functions that enable primary activities to run successfully.
Identifying where AI is implemented allows to conclude about AI requirements. Depending on the organizational function it has been applied in, conclusions about high valued AI
capabilities and also strategic importance can be drawn. Adopting AI for instance within a primary activity with major impact on value addition may reveal changes in industry practices enabled by AI adoption. Application in a support activities however might explain a tendency to adopt AI in non-client facing activities (P23).
Technology Adoption
After reviewing literature aiming at explaining why companies adopt AI, literature on the adoption process must be investigated. Two major technology adoption frameworks intending to predict an organizations decision to adopt a new technology will be explained below. Special attention will be paid to the occurrence of shortcomings due to the pioneering nature of AI technologies.
Innovation-Decision Process
Understanding how adoption decisions are made is essential for the implementation of any new technology. However, new technologies may require new procedures to be adopted successfully. “The innovation-decision process is the process through which an individual (or other decision-making unit) passes from first knowledge of an innovation, to forming an attitude toward the innovation, to a decision to adopt or reject, to implementation of the new idea, and to conformation of this decision” (Rogers, 1983). The model consists of a time sequence of steps and stages involved when adopting a new idea and is widely applied by any organization. Rogers states, the decision to adopt an innovation, includes high levels of uncertainty, which distinguishes it from other types of decision-making. When talking about adopting new technologies, these are also considered and treated as innovations.
The generic innovation-decision-process (IDP) carried out by organizations looks as follows (Figure 1): The process is initiated by setting an agenda. Here, general organizational
by matching an innovation to solve a current organizational problem, including methods of scenario analysis to test the feasibility of the innovation. Firms that have identified AI as a suitable solution to a problem that its facing, are assumed to be more likely to consider adopting AI (P6). Rogers also describes potential characteristics of an innovation such as: relative advantage, compatibility, complexity, trialability or observability. “The perceptions of an innovation by members of an organization’s decision-making unit (DMU) affect their evaluation of and intention to adopt a new product (e.g., Ostlund, 1974; Tornatzky & Klein, 1982; Holak et al., 1987; Rogers, 1995). The perceived benefits, including economic
incentives, of adopting the innovation should exceed those of alternatives, if organizations are to consider adopting (Anderson & Narus, 1999). Indeed, the perceived net benefit the
innovation offers has an important effect on the organizational adoption (Robinson, 1990; Mansfield, 1993)” (Frambach & Schillewaert, 2002, p.164) (P12). It has to be noticed that factors such as trialability may be moderated by suppliers of AI solutions. By offering lower introduction prices or trial periods the risk connected to adopting a new technology may be reduced and hence, the likelihood of adoption increases (Fisher & Price, 1992; Ram & Jung, 1994; Kotler, 1998) (P11).
It is assumed that the higher the compatibility appears to be and the less complex the technology is, the more likely organizations are to adopt (P8). Further, it is suggested, the easier it becomes to observe AI in action (as more companies adopt) or even having the opportunity to try it, the higher chances become that firms who manage similar processes will follow (P9, P10)). However, as Mark Purdy (2016) states: “Contrary to traditional assets that depreciate with time, AI assets gain value as time passes, given the self-learning nature of these technologies. These compounding asset appreciation effects create a big first-mover advantage, so this is an important reason for CIOs to make early investments.” Investigating the technology aspect may involve interviewing those in charge of finding suitable
technology, since this is where the perceived complexity plays a role.
Moreover, after an idea has been initiated, the implementation phase begins. Frist, the
innovation is altered to fit the organizational need and organizational structures are adapted to support and catalyze the innovation. Then, the relationship between organization and
innovation can be defined more distinctly and it can be put into actual use. Finally, the innovation loses its separate identity, however during this phase, it can also be discontinued. The innovation-decision process model provides an outline for implementing an innovation. Even though a sequence of steps is important for every organization to organize their
decision-making processes, it misses out a number of factors to consider when deciding upon adopting an innovation/technology. Therefore, this model provides a valid structure for deciding upon adoption but cannot be used without being complemented by a model that covers additional factors influencing a decision.
During this study, the implementation phase of this model is not of importance, due to a focus on the adoption decision rather than its implementation.
Figure 1. Innovation-decision process, adapted from Rogers (1983)
TOE Framework
A model complementing the innovation-decision process is the technology, organization, environment (TOE) framework by Tornatzky and Fleischer (1990). The TOE framework consists of multiple aspects that influence the technology adoption decision of organizations (Figure 2). Contrarily to the IDP, the TOE framework suggests the decision to adopt a technology being a product of three factors. Simultaneously acting in combination instead of
passing through stages of a decision-making process leads to a final decision. One factor is “Technology”, consisting of availability and characteristics of the new technology. Here, all technologies relevant to the firm should be considered. In this context, again, adoption is influenced by relative advantage the new technology provides, the complexity of it and the compatibility with experience and existing value (Rogers, 1995). Hence, in the context of this study, the bigger the perceived relative advantage and its compatibility with the current organizational infrastructure, the higher the likelihood of a firm to decide to adopt AI technologies. At the same time, it is expected that if the technology is perceived as less complex, chances of deciding to adopt are higher.
The next factor is “Organization”, which refers to characteristics of an organization. Not only descriptive aspects such as size, slack, structure or communication are considered, but also top management support or IT expertise of business users (Borgman, Bahli, Heier & Schewski, 2013). As organizational behavioral theory suggests, flat firms with low
formalization are assumed to adopt innovations and thus, AI faster due to their high flexibility (P13). At the same time, firms possessing slack are suggested to have higher chances deciding to implement AI technologies due to high costs of data analysts required to implement an AI system (Zaltman et al., 1973) (P14). Generally, small firms possess characteristics such as low formalization and a flat structure but do not develop as much slack. Contrarily, large firms are often highly formalized with a strong hierarchy but own the required resources. Therefore, the impact of an organization’s size on the decision to adopt may vary with structure and degrees of formalization. Further, “considering that increasingly non-IT employees - or at least their management - are involved in strategic IT decisions, their perception and understanding of the targeted technologies is important” (Borgman, et al., 2013). Hence, it is suggested that non-IT employee’s perception of the technology impacts the decision to adopt AI accordingly (P16). Also, top management support is expected to
organizations productivity top management might indirectly support the adoption by
allocating more resources, facilitating a supportive vision or introducing a reward structure. (Venkatesh & Bala, 2008). Depending on the degree of trust top management has in AI, adoption might be accelerated by direct support as well. Educating, communicating,
participating or manipulating among other instruments of direct top management involvement may determine adoption speed (Jasperson, Carter & Zmud, 2005; Morgan & Zeffane, 2003; “Why Senior Managenemt”,2017). On the other hand, distrust may significantly slow down or limit adoption.
Lastly, environmental factors refer to any interaction an organization has with its external environment such as competition intensity or the regulatory environment (Thong, 1999). Competition intensity is “the degree that the company is affected by competitors in the market" (Zhu, Xu, & Dedrick, 2003). In this case, expenses are high and return on
investments are unpredictable or even perceived to be low due to the productivity paradox. Therefore, competition intensity is expected to be negatively related to AI adoption, even though previous literature has found positive effects of low and high competition intensity (Baldwin & Scott, 1987; Kamien & Schwartz, 1982; Makadia, 2017) (P18). Moreover, non-adoption may result in competitive disadvantage if adopted by others. Regarding the
regulatory environment, organizations are suggested to have a lower likelihood of adopting when strict regulations are applied. Since AI requires huge amounts of data, government regulations are assumed to be stringent in order to guarantee privacy and also safety when working with AI (European Union, 2016) (P19).
The TOE framework well considers all technological, organizational and environmental factors impacting the decision to adopt a technology. However, the model does not take into account when which of these factors affect the decision (P22). As describes before, the decision-making process can be divided in multiple steps, of which each has different characteristics and requirements. Hence, neither the innovation-decision process model, nor
the TOE framework can be used without being complemented when understanding
technology adoption decisions of organizations. Therefore, a third model, combining factors and in which sequence to consider them, will be crafted. Additionally, to each model’s incompleteness, the TOE framework especially lacks considerable factors affecting the adoption decision of AI.
Novel abilities of AI require new factors to be considered affecting the decision. AI does not only have the ability to augment humans, by taking over routine, repetitive tasks, it can for many tasks function as a human itself. Through capabilities such as natural language-, text- or image recognition it possesses sensing abilities comparable to humans’. Although, these technology characteristics can be interpreted as part of the framework, they should find special consideration. Replacing human labor might eventually change a multitude of organizational factors such as size, structure or communication and flexibility which then again affects future adoption decisions. Technological side effects of this kind have never been described before but may play a significant role when adopting AI.
AI’s ability to learn, results in an intangibility of return on investment because, unlike different technologies, its value increases over time. The reason being, once invested, it enhances its knowledge and capabilities using machine learning without any further intervention. Thus, instead of depreciating, it will gain value, making it impossible to even predict its impact on a company’s productivity (P15). This new aspect of technology has not found consideration in theory yet, however it is of importance due to its potential effect on decision making. Having continuous return on a single investment might very well increase chances of adoption.
An additional, novel technological aspect that must be considered deciding upon
implementing AI is its ability to be combined with a variety of machines. The process of making a machine smarter by adding AI to it is called cognification. Machines that so far only run repetitive tasks will be able to learn, sense and/or adapt to new situations by themselves.
This area of application makes AI very flexible and has not been discussed within this framework before. While usually new technologies serve a certain task, AI technologies can be adopted in a variety of processes, making it very attractive to invest in. Not only could it possibly cut operational costs and increase productivity, it may also increase customer satisfaction or product quality by delivering better, more customized services.
Further, in order to employ advanced AI technologies, large amounts of data and computing power may be required. The technology is often “fueled” with data of which it then derives conclusions and finally learns and unfolds its potential. This is specifically relevant when deciding upon AI adoption because firms who cannot provide the data necessary might very well decide not to adopt. This aspect of data availability is not covered by this framework and should therefore be added in order to be applicable in an AI context. While the availability of big data might restrict AI adoption, access to required computing power is not expected to hinder the adoption due to the emergence of cloud computing technology (Purdy & Daugherty, 2016) (P20, P21).
Furthermore, perceived advantages of an adoption are left behind. Since actual benefits cannot be calculated before a decision has been made, perception of those may play significant role when adoption AI. Along with this argument, concepts like hype or first mover advantages and experience come into play. It can be assumed that if benefits are expected to be large by decision-makers, the tendency to adopt a technology is higher. “The perceived innovation characteristics can be considered as cognitive indices (or beliefs) reflected in an attitude towards the innovation (Rosenberg & Hovland, 1960; Le Bon & Merunka, 1998). This attitude can be manipulated, whereas actual technology characteristics cannot. Also, competitive pressure may be perceived differently across different individuals. A somewhat sensitive person may perceive little competition already as a challenge while others do not, affecting the evaluation of risk making the decision to adopt. As mentioned before, a technology’s complexity, trialability or observability may be perceived as low
because of the emergence of a supplier between the company and the technology who moderates the relation. There is conceptual and empirical evidence that, in organizational settings, attitudinal components mediate the influence of external variables, such as
motivation, on behavioral intentions (Le Bon & Merunka, 1998). Similarly, attitude theory (e.g., Triandis, 1971; Fishbein & Ajzen, 1975) hypothesizes that beliefs mediate the impact of external influences, such as persuasive communication and/or active participation on
decisions” (Frambach & Schillewaert, 2002, p.164). Adding such a perceptional level of analysis may help to understand why technologies have been adopted by first movers, but as expectations were not met, might eventually have not been established as an industry wide standard.
Figure 2. Technology, organization and environment framework, adapted from Tornatzky and Fleischer (1990)
Conceptual Framework
From the described limitations of the TOE framework and the IDP a new model has been crafted (Figure 3). Instead of technological and environmental factors directly influencing the decision to adopt a technology, a perceptional layer has been included. The perception of the decision-making unit then influences the decision to adopt, along with organizational factors. Since the adoption decision is a sequential process, not all factors are considered at the same point in time (Rogers, 1995). While some may be considered setting an agenda within an organization, others may be considered matching a technology to organizational requirements.
Figure 3. Conceptual framework adapted from Frambach &Schillewaert (2002)
Methodology
Research Design
This research is designed to understand the impact of several factors on the decision-making process of adopting AI technologies. After having established its significance across
organizational functions and industries, the question if AI is just another technology needs to be answered. More specifically, does it possess characteristics technologies have never proven to possess before? Hence, can it be adopted like any IT technology before or does it create the need for a novel approach? Due to the early life cycle stage of AI adoption, the research utilizes a qualitative, explorative approach to gain an understanding of the nature of this emerging technology (Strauss & Corbin, 1990). In order to guide the process of
answering the research question, propositions derived from current theoretical frameworks and practice will be tested on its fit with AI during case studies. Important decision-making factors will be identified and analyzed in order to test the goodness of fit of two technology adoption frameworks for AI. Eventually, the proposed conceptual framework, according to AI requirements, can be confirmed or current frameworks will be revised. Also, it has to be understood when organizations decide to implement AI, which factors play a role and which processes are affected. Hence, the research questions this study tackles is as follows: Why do companies decide to adopt AI? And, how well do current theoretical frameworks predict the decision to adopt?
To understand the decision-making process of adopting AI in companies, numerous factors had to be considered. Not only were there many factors, but also several different
stakeholders involved. Partly, observations from outside a company could deliver insights on organizational, environmental and technological characteristics, however, more insightful answers have been originated by being inside the system. Diving into an organizational environment allowed to explore a variety of new factors that may influence such an adoption decision from multiple angles. Facing the following questions could only be answered by interviewing involved stakeholders (unit of analysis) as psychological insights must be gained: Who brought the idea to the company and what problem was it supposed to solve? What was expected of it and has there been hype or doubt about implementing it? Who triggered the decision-making process and who ultimately decided? Coming to more
organization related concerns such as: Were employees scared of losing their jobs and showed resistance before the decision has been made? Finally, also which factors did the decision-maker consider when evaluating the fit between the organization and AI? Such factors are described by the previously mentioned theoretical frameworks, but to what extent do they help understanding the decision to adopt AI? Keeping interview questions open and allowing room for discussion enabled the researcher to gain as much insight in technological
characteristics as possible. This semi-structured method of collecting data also allowed the researcher to explore new insights gained during the interview and adjust the order questions are asked to create a natural conversation. Where possible, interviews have been conducted face to face to also capture subtle information communicated through gestures and mimics of the respondent (King, 2004).
By recording and transcribing the interviews, gathered data can be analyzed systematically. Using codes and grouping codes into broader categories, common themes, patterns or relationships with previously defined propositions were identified (Strauss & Corbin, 1990). A set of “starter codes” has been designed before the analysis based on important concepts in literature. Other codes emerged during the analysis due to the explorative nature of the data collection. A transcription of the interview enabled the researcher to add rigor to the study by using qualitative data analysis tools such as Nvivo.
In order to identify factors influencing the process of deciding to adopt AI technologies, a multiple case study was conducted. To ensure internal validity and structure the data analysis, propositions have been designed from established technology adoption frameworks
(Appendix A). Those frameworks aim at predicting an organizations tendency to adopt a technology based on different factors. Interview questions then have been formulated to test these propositions. During this study, by according or contradicting with suggested factors, the likelihood of adopting was determined. Also, internal validity has been improved by running a pilot interview. This initial feedback then was used to reword or rephrase critical
questions or to assess if the intended concept was measured. When conducting such a
research, a central question to answer is how many cases should be chosen? Aiming at testing and building theory, multiple cases are of advantage due to its robustness to observer bias and its augmentation of external validity (Yin, 2013). Four cases have been selected upon
identification in trade press, increasing the reliability of findings compared to single case studies (Herriott & Firestone, 1983). Using such purposive sampling methods ensured the case’s observance with unconditional requirements of this investigation (Stone, 1978). Organizations which (recently) decided upon implementing AI technologies qualified as a potential case regardless of their industry. The rather low number of cases (4) was justified by the explorative nature of this research. In order to gain initial insights in the subject matter, it was not necessary to achieve many replications between cases to assign more certainty to the results. Moreover, there were no rival explanations to the topic as it is still emergent and very little research has been conducted (Yin, 2013). The degree of explanatory power in this qualitative study, has been estimated based on observations in the case companies. High accordance of observations with factors/steps described in both theoretical models therefore resulted in high explanatory power. While some cases were embodied by manufacturing companies, others were by service providers. It has not only been differentiated between the tangibility of their outputs but also if AI has been implemented in their primary or secondary functions. This distinction was necessary to explain which factors affect AI in different areas of application. Was it for example primarily used for planning and analytical activities in manufacturing firms or was it also applied in a final product or service and where was the difference in application. This difference could be interpreted in terms of AI’s focus on improving internal processes of a company or firms targeting a whole new range of
innovative end-products. Therefore, the subject of this study were organizational functions or activities where AI has been implemented.
Case Description
In order to explore the adoption of AI to the largest extent possible, case companies of contrasting sizes and industries have been selected. In the following, cases will be briefly described to give an initial insight as to who the companies using AI are and how the technology has been applied.
ING Bank
ING is a Dutch, international bank operating in retail and wholesale banking services. By striving to be “Clear and Easy, available Anytime and Anywhere, to Empower and to Keep Getting Better”, (ING, 2016) ING is a leader at digital banking. The bank employs almost 52,000 people of which 13,660 are in the Netherlands. The interview has been conducted with Bolke de Bruin, head of advanced analytics technology, in Amsterdam. ING applies AI throughout the organization, solving organizational issues at hand, decreasing time to market and also to enable personalized customer service. The technology has mostly been
implemented in support functions improving the bank’s primary activities.
Twentsche Kabelfabriek BV
The Twentsche Kabelfabriek BV (TKF) is “a leading supplier of connectivity solutions built on advanced technologies” (TKH, 2015). While being part of the TKH Group, the telecom department employs more than 600 people. The interview partner was André Goorden, operations manager at TKF. The company acquired an innovative camera system powered by AI to detect failures in production. In this case, the technology has been implemented into the company’s operations, directly influencing core value adding processes.
MzB GmbH
MzB is a family-led SME with a pickup and delivery service model connecting farmers with the food industry. The interview has been conducted with the managing director, Dirk Krümpelmann. A planning tool based on AI technology has been adopted to enhance capabilities of secondary activities.
Accenture Digital
Accenture Digital employs approximately 384,000 people across the globe and intends to “help clients in nearly every industry to use digital technologies to deliver more meaningful and relevant customer experiences across all channels and customer segments, as well as to create new products and business models and to optimize the efficiency and effectiveness of their internal operations” (Accenture, 2015). An interview has been conducted with Jens Frühling, the principal director. The technology has been applied throughout the division to improve internal processes, gain experience using it and to extent the product portfolio. Notably, also within primary functions AI is of great value.
Operationalization
Appendix A shows which interview question was designed to test which of the previously stated proposition. By asking how a certain concept (covered in a proposition) affected the decision to adopt AI within each case allowed to investigate the practical implications of theoretical literature within an AI context. This way, also the derived limitations of current frameworks could be analyzed. Multiple questions have been used to test a single proposition to cover it to the full extent. Not only were questions based on literature, also theoretical concepts were used to classify answer, enabling a more rigor analysis (Appendix B). Different authors making the same propositions in their works have been identified and referred to, to increase construct validity. If authors used a similar conceptual framework, their research
instruments have been used to support interview questions developed for the purpose of this study.
Analysis Strategy
In order to investigate how and why organizations decide to adopt artificial intelligence (AI), interviews have been conducted (Appendix A). The nature of this research is inherently explorative as the technology is still in its early stages of adoption. Even though AI is around for decades already, only today prices for the necessary computing power are low enough and sufficient data is available to effectively utilize the technology. Interview questions were based on propositions derived from established theoretical frameworks and often also include categories that possible answers can be classified in, to guide discussion.
As a very first step, the interview recordings had to be transcribed. Such a transcription allowed to easily share the data but also to improve the workflow. When reading, instead of listening to an audio file over and over again, finding the right paragraphs and sentences that one might be looking for was done much quicker. Also, when archiving the data, a text format is much more suitable due to the ability to query documents for search terms. Further,
working with a text file enabled the use of qualitative data analysis tools such as Nvivo. Using these tools improves the rigor of any qualitative study since numbers can be assigned to frequencies of a words occurrence or graphs can be created, a quantification qualitative research often lacks.
In order to actually analyze the gathered qualitative data, it first needed to be structured. Unstructured data is hard to work with, as regularities and patterns are almost impossible to detect. Organizing the data gathered within each individual started with assigning descriptive codes to sentences and paragraphs of the gathered data. Using this first set of codes created a summary of the transcription, establishing an overview of topics within the data set. Next, analytical codes were added. This second set of codes tackled a deeper level of analysis.
While descriptive codes only describe the data, analytical codes go beyond describing and aim at a more detailed evaluation of the data (Saldaña, 2015).
Many codes were created while working through the data set, others, the “starter codes”, were predetermined by the literature providing the foundation of the interview questionnaire. For example, the TOE framework predicts the dichotomous decision of an organization to adopt or not to adopt a technology by considering technological, organizational and environmental factors (Tornatzky & Fleischer, 1990). As part of the technological factors a technologies complexity, observability and trialability influences the decision to adopt a technology. In order to obtain an exact measurement of those distinct factors, they have been included in the interview and also been chosen as starter codes due to their importance to academia. Since this research is of explorative nature, it was impossible to predict answers of respondents and create a complete set of codes consisting only of “starter codes”. Especially out of the box answers therefore often required new codes. After organizing the data, individual codes could be categorized to a broader context (Strauss & Corbin, 1990). Common themes were:
motives-, adoption process-, capabilities-, impact of AI and technological, organizational-, environmental factors. The resulting coding structure allowed to draw conclusions and identify relationships to answer the research question. The process of drawing conclusions was supported by displaying grouped codes in a diagram. This diagram then visualized potential relationships between codes (Appendix C). Results gained by drawing conclusions and identifying relationships then were used to refine fundamental theory.
Across all cases conducting during this study, findings have been aggregated. By displaying discoveries in tables and graphs, significant cross-case conclusions were highlighted.
Results
In order to investigate the research question, why companies decide to adopt to AI,
propositions will be accepted or rejected based on data gathered during interviews (Table 1). In the following part findings within each individual case will be described. After drawing conclusions from single cases, findings can be aggregated, compared and discussed across multiple cases.
ING Bank
Comparing the coding structure with propositions derived from theoretical frameworks allows to achieve initial results of this investigation. The first proposition stresses the inherent
difference of AI compared to previous technologies. This assumption was confirmed by the head of advanced analytics technology of ING (P1). “First of all, I don’t consider AI a technology”. The assumption of an increased tendency to adopt when large benefits are expected could not be confirmed (P2). A team always works on solving current organizational problems. They identify hypotheses including AI applications that may solve their issue. After researching and testing, the hypothesis may be accepted or rejected, resulting in AI
implementation or further research. Hence, the hypothesis that firms who identified AI as a solution to an organizational issue are more likely to adopt can be confirmed (P6). Along those lines, the assumption of aiming at increasing productivity must be rejected as organizational issues are not always related to productivity deficiencies (P3). The same reasoning applies to the proposition that high perceived relative advantage increases
likelihood of adoption. Solving an organizational issue does not necessarily result in relative advantage over legacy methods of dealing with an issue (P12). Facing new issues may require innovative solutions which cannot be compared to the past. Furthermore, avoiding large human capital expenses is not a motive according to ING (P4). Even though the head of
advanced analytics technology stated that the banking sector should not take away much capital from the economy and drastically reduce human capital in the financial sector, his task was not to outright eliminate workplaces. Again, AI is used to solve current organizational issues and aims at creating new customer insights to provide better service instead of consciously reducing employee counts. Resistance of employees is not considered being a factor influencing the decision to adopt AI but top management support is valued as very important (P5, P17). Based on the interview, resistance should be reduced by assigning change agents who belief in and support the new technology. Hence, an adoption decision is made and afterwards it is dealt with resistance by tackling any concern employees may have in a changing environment. While the proposition of resistance constraining adopting can be rejected, top management support is crucial to a successful adoption of AI according to ING. Within the company, top management drove the change where lower level employees could not push through their opinion. Therefore, top management support can be the deciding factor for implementing change or in this case, implementing an emerging technology.
According to the IDP framework, the lower perceived complexity and the higher the
trialability of a technology, the more likely it is the it is going to be adopted. This proposition can be confirmed based on the interview with the head of advanced analytics technology at ING (PP8, P10). Complexity was rated low and trialability extremely high. At the same time, observability and compatibility did not play a role (P7, P9). AI was applied for
organizational-specific tasks which cannot not be observed anywhere else. Hence, it was not considered when designing a solution. Also, compatibility with current organizational IT infrastructure did not influence the adoption process. “No, we are not going to be hold back if the foundation is missing. We just build our own infrastructure and make it work for the cases we work on”. Clouds to easily enhance computing power or big data that AI can be applied on have been described as drivers of AI. However, during the process of making an adoption decision, they have not been considered (P20, P21). Next to overcoming initial difficulties of
assembling a skilled team of data analysts, the deciding factor of adopting AI is the availability of monetary resources. The interview revealed large costs associated with employment of AI, more specifically the employment of qualified data analysts. The right place of application has to be found and then the actual work of designing a solution may start. This process can take very long, making the project very capital intensive. Small firms with undiversified revenue streams and little slack will probably have difficulties spending much time on finding an AI solution. Whereas large organizations possessing slack may be able to afford spending time on finding the right case for AI. Hence, the proposition linking larger organizational size and more slack to increased likelihood of AI adoption can be confirmed (P14). However, the assumption anticipating lower importance assigned to costs because of AI’s increasing value over time through its ability to learn must be rejected. As another organizational factor, IT expertise did not play a role when considering an adoption because of its difference compared to traditional information technologies (P16). Only the AI team is concerned working with AI, the IT experience of other employees does therefore not matter.
Considering the external environment revealed that competition for ING is not intense because of competitors but rather through changing customer demands and needs. By implementing AI applications more value can be delivered to the customer helping to incorporate banking activities in their daily life without much ease. At this moment, there is no other opportunity to deliver the extra service and therefore the adoption decision has not been influenced (P18). Also, the regulatory state of affairs did not impact the decision to adopt AI (P19). As the head of advanced analytics states: “Laws do not affect AI itself. They influence what we create with AI. But they do not influence the decision of using or not using AI”.
The IDP framework describes the process of adopting an innovation, where first an agenda is set and then an innovation is matched with the company’s needs (Rogers, 1983). Asking the
interviewee about the order of consideration of each factor influencing the adoption decision was answered by claiming there is no distinct order (P22). However, the proposition derived from the IDP framework stating technologies will be matched to the organization
requirements can be confirmed. According to ING, every AI application is tailored to a special organizational need. Therefore, every case of adoption is fully matched to the company. Due to the problem solving focused approach, single projects were initiated. Whenever AI was the solution to an organizational issue, it was implemented without any difference to other problem-solving initiatives.
Twentsche Kabelfabriek BV
Within TKF, AI was not perceived as being different from previous technologies (P1). AI simply replaces an old technology which cannot deliver the performance AI can. Simply put, expecting an improved performance on multiple measures resulted in adoption. Hence, large expected benefits and a relative advantage compared to the legacy technology contributed decisively to the final decision (P2, P12). Furthermore, the operations manager of TKF also anticipated to gain a competitive advantage. By adopting a new camera system powered by AI, not only can it operate faster, but also in much greater detail. Detecting a failure is not based on a dichotomous decision any longer, but the system analyzes the failure and decides if it is critical to product quality. This ability eliminates waste and reduces rework, resulting in lower operational costs and therefore, higher productivity (P3). Even though lowered costs have been considered as one decision factor, the competitive advantage stems from being able to deliver a wider product portfolio including very small cables, nobody else will be able to produce. Thus, a differentiation advantage will be gained.
While saving material costs, the proposition stating a tendency to adopt to save human resource expenses must be rejected (P4). The task AI is performing has already before been performed by a less advanced technology. By adopting AI in this case, employees will be
augmented and continue operating the system thorough mainly monitoring tasks. Thus, there has been no hype, nor resistance to adopt the new technology (P5). Although productivity has been increased and the product portfolio extended, AI was not adopted to solve an
organizational issue (P6). The implementation solely incarnates the replacement of an old system with a new one.
As for technological factors, AI has been perceived as somewhat complex but also as very simple to observe, try and implement reducing the risk of adoption (P8, P9, P10). The complexity derives from the magnitude of data provided through the new system. As the operations manager specified: “Employees were not able to understand all the parameters provided by the technology and got confused.” The new camera system was not developed in-house but acquired through a supplier, accelerating the speed of adoption. Instead of
employing data analysts to continuously improve operational software, TKF representatives visit an exhibition once a year, presenting major improvements in hard and software. The assumption expressing a higher likelihood of adoption through reduced risk by the supplier can be confirmed (P11). Observability as well as trialability have been diminished by offering the customer to send in samples and test if failures can be detected according to
specifications. By attaining the certainty of a fully functioning system, the risk of making the purchase and thus adopting AI, is minimized. Contrary to the proposition articulating a tendency to adopt AI when big data and cloud computing are available, this did not affect the decision-making process to adopt within TKF (P20, P21). During this process, the system has also been matched to the company’s requirements. The operations manager stated that the technology was about 70% ready to be implemented when reviewing the supplier. One component that has already been compatible was the infrastructure, which again benefitted the decision to adopt (P7). The remaining 30% had to be matched to fit the organization. Part of the adjustments for example was the user interface enabling employees to monitor the production.
While the adoption was favored by the afore described technological factors of AI,
organizational characteristics somewhat restricted the adoption in this case. TKF being part of the TKH group, the board wants all members to buy machinery among member companies. However, cameras provided internally cannot reach the performance of those available outside the group and building up the capabilities to write the AI software powering those cameras would take too long. This internal conflict of interest restricts the adoption and until this time, limits adoption to a few production lines. Therefore, company structure negatively influences the adoption (P13). Also, the operations manager contemplates: “Small,
independent departments with a focus on innovation will always adopt technologies faster than large, bureaucratic firms focusing on cost leadership.” Furthermore, employing skilled labor within Europe instead of outsourcing production increases likelihood of early adoption. By keeping the production in Europe, TKF consciously accepts a cost disadvantage and tries to gain a differentiation advantage through an extended product portfolio powered by
innovations. Another organizational factor is the support of the management educating employees and helping AI being adopted without resistance (P17). Even though replacement of human labor was not at stake, people were still afraid of change and needed guidance throughout the process.
Within their industry, no company makes use of this AI powered camera system for fault detection. Further, the case company competes based on their unique product portfolio which low cost manufacturers cannot produce. Although competition is fierce, the company’s advantage is centered around taking risks to steadily improve their quality. Hence, in this case, a high competition intensity does not lower the likelihood of adoption AI (P18). No second environmental factor influencing the decision to adopt has been identified as no laws or regulations hinder the use application of AI technology (P19). Throughout the decision-making process of adopting the technology, technological factors have been considered first, followed by organizational and finally environmental factors (P22).
MzB GmbH
Comparable to TKF, MzB also acquired a whole AI solution from a supplier instead of developing capabilities in-house. Hence, it is not perceived as inherently different than other technologies (P1). Even though large benefits were expected, including improved bottom and top line performance, those were not the distinctive decision factors. The main goal of
applying the AI technology was to augment employees and free them up for more value adding activities. Hence, expected benefits and increased productivity have been considered making a final decision, however those were pleasant byproducts trying to solve an
organizational issue in form of support their employees (P2, P3, P6). Indirectly, freeing up current employees for more important tasks results in employing less people in the future. Even though future expenses on human capital will be reduced, this has not been considered making the decision to adopt (P4). The new, AI-powered planning tool, is presented as just another tool to help employees do their job, without calling it artificial intelligence. Hereby, resistance to adopt should be avoided (P5). Besides education on how to use new tools, preventing resistance is the only way of management support. Considering prevention as a support strategy implicates a positive relationship between management support and tendency to adopt AI (P17).
With MzB’s operations consisting to a large extent of logistics and distribution, planning will always be an organizational concern. Many different parameters have to be considered and achieving the most efficient outcome is a never ending goal the company will pursue. Thus, AI has been identified to solve this organizational issue which increased the likelihood to adopt (P6). The managing director stated: “Using AI, not only can it take up the planning, it is able to consider a lot more factors in a much shorter time, achieving a performance a human could never compete with.” Also compared to previously used planning software, AI delivers more value by continuously learning and improving. This relative advantage increased the
likelihood of adoption, even though it did not reduce the importance of costs connected to the adoption. (P12, P15).
Considering technological factors, understanding the problem AI was supposed to solve was simple. Solving the problem however, is a highly complex procedure. Since the case company understands the problem and is able to exactly describe requirements on the software,
perceived complexity is low. Again, in this case, the supplier reduced the risk of adoption by influencing observability and trialability. Beforehand testing the added value by “feeding” it with old data and create a planning which then has been compared to results achieved with manual planning showed an improvement by 14%. Also, a trial version has been offered to make sure the technology can keep up with day to day operations. It has been identified that the software already matched the organization’s demands, only a few industry specific rules had to be added. The most important technological factor in this case is the compatibility with current infrastructure. The AI tool requires large amounts of data input. If the necessary infrastructure to gather all the data was not available, the implementation would not be possible without further ado (P7). This goes along with the availability of big data. Making the decision to adopt is enabled by big data being available to feed the AI software (P20). Required computing power however, was available at the company’s servers without considering cloud computing when deciding upon adopting AI (P21).
As for organizational factors, the managing director of MzB stated that for a SME acquiring an AI solution is only possible through suppliers. Employing data analysts to develop an AI solution or write any other large software for an organization specific problem is too
expensive. Consequently, SMEs will always have to wait for a supplier to appear and provide the solution for them at an affordable price. Therefore, the company’s size influences the outcome of deciding to adopt AI. The larger, the higher the likelihood of adoption due to the availability of slack (P13, P14). While size delays the adoption, the company’s IT expertise favors the decision to adopt AI. Most employees will not need to have extensive IT
knowledge. Especially in this case, the top management did not have any experience with IT, however, when communicating requirements and specifications to the supplier, IT expertise of employees will help to successfully share information and achieve the desired outcome (P16).
Even though international competition is tough, the decision to adopt has been made. By adopting AI which increases productivity and enables to achieve higher quotas, a competitive advantage should be gained. Thus, competitive pressure led to a higher likelihood of adoption to remain competitive (P18). Additionally, environmental factors were a considered when evaluating costs of the investment in AI. A large part of law is dominated by licensing regulations. Acquiring software always goes hand in hand with consecutive costs which are considerably high for SMEs. Hence, strict software regulations decrease likelihood of AI adoption (P19). Again, throughout the decision-making process of adopting the technology, technological factors have been considered first, followed by organizational and finally environmental factors (P22).
Accenture Digital
Findings at Accenture Digital do not confirm AI being inherently different from previous technologies (P1). As technology focused consulting company every new, “big” technology will be adopted and incorporated into the existing product portfolio. This process is natural to keep up with technological developments and cannot be considered an organizational issue (P6). The adoption in this case serves two reasons. The primary reason being to provide customers with fresh, emerging innovations. Of course, this is not completely altruistic. Facing fierce competition, the product portfolio has to be state of the art to provide the customers with a comprehensive service experience and to stay competitive. This finding contradicts with the proposition derived from literature stating a decreasing tendency to adopt AI when competitive pressure is intense (P18). The second reason for the adoption is the
improvement of internal processes. Not only should operational efficiency be improved but also experience gained by utilizing the technology on a daily basis (P3). Therefore, aiming at achieving a differentiation advantage through a unique consulting experience distinctly influenced the decision to adopt (P2, P12). However, relative advantage over previous
technologies used for similar task did not impact the decision-making process. Further, laying off employees or avoiding human labor expenses has not been described as an influencing factor (P4). Also, resistance of employees to accept AI does not impact the adoption (P5). Implementing new technologies is part of the company’s culture and will therefore not trigger a conflict.
As for technological factors, levels of complexity were perceived from low to high depending on the exact application. However, through partnerships with large technology developers the impact of complexity is decreased. Due to this support and cooperation the perceived
complexity any AI application does not influence the likelihood of adoption (P8). Also, trialability and observability were significantly increased. The joined research and
development enabled Accenture Digital to observe and also try out the technology before deciding to adopt, increasing the likelihood of adoption due to reduced risk when
implementing (P9, P10). Hence, the risk of adoption is reduced by the supplier (P11). However, this is not the main reason for the collaboration. Rather the team up happened to deliver faster and more customized access to technology when consulting a customer. Partnerships also reduce the importance of compatibility with current organizational infrastructure. The close relationship to suppliers simplifies the process of making
adjustments to fit the innovation to be adopted, which excludes compatibility as a high-impact decision-making factor when adopting AI (P7). However, Accenture Digital still tends to adopt applications which require only minor adjustments to the organization, while for customers often unique solutions are crafted. Those applications then are matched to fit the company’s needs or the company is matched to fit the application. Furthermore, the
availability of cloud computing enabled AI to be implemented (P21). According to the managing director, 80% of all AI application within the company are cloud-based. Only 20% run on local servers, expressing the importance of cloud computing when deciding to adopt AI. Also, without the availability big data, AI would not be implemented as data would be missing that the software could be applied to (P20).
Organizational size did not affect adoption. Contrary to the proposition stating an increased tendency to adopt within smaller companies due to their ability to react more flexible to environmental influences, the case company adopted AI at the earliest possible stage. The large size did not restrict adoption, rejecting the assumption (P13). Since the company has extensive IT expertise, the overall likelihood to adopt was drastically increased and the decision to adopt was catalyzed. In fact, IT expertise is the cornerstone of the business model and enabled Accenture Digital to make the decision to adopt early. By obtaining first mover advantages the aforementioned competitive differentiation advantage can be fueled. Since the adoption is part of the company’s strategy as decided by the top management, support is strong, favoring technology adoption decisions (P17).
Considering environmental factors, privacy laws somewhat influenced the decision to adopt. While AI can be applied without any regulatory issues to a large extent, application to sensitive data will not always be possible. Therefore, when working with private data, regulations may decrease the likelihood of adopting AI (P19).
Table 1. Proposition findings
Discussion
Multiple case studies have been conducted to explore the reasons behind a company’s
decision to adopt AI technologies and to find out, if current technology adoption frameworks are suitable for an AI context. In the following section, findings across cases, as well as the new conceptual framework crafted for the purpose of this study, will be discussed.
AI Adoption Decision
During interviews, it has been found that the goal of AI application often was planning and/or automation of processes. The benefits case companies expected were to increase operational speed or accuracy, improve operational efficiency or being able to quickly respond to changes in the market or customer needs. Also, traditional software was replaced by smarter software to emulate human-like decision-making capabilities. The aforementioned benefits can be summarized in two categories: increasing productivity or speed. Therefore, any company considering improving either one of those two factors is expected to have a high likelihood of adoption AI in the future. However, by supporting the development of the technology in other areas of applications as well, it might provide even more benefits.
Not a single case suggested that the adoption aimed at replacing human labor but rather to augment employees or give them more pleasant, value adding tasks. It is implied by the findings that jobs for human labor will focus on monitoring and increasing customer service levels through direct interaction from human to human. Thus, in practice, a reduction of workplaces in the near future is highly unlikely. Employees will be freed up for more value adding tasks, possibly reducing the urge for new employments. Since no case company considered replacing human labor in the first place, there was no exceptional resistance to change. Also, decision-making units seem to have understood the sensitive perception of the topic by the general population and often do not confront employees with the term “AI” itself. Education and communication of purpose and goals of AI applications among the general public should reduce fear of adoption.
Evaluating technological factors considered by theoretical frameworks, it has been noted that some factors, depending on the supply of AI, apply differently compared to previous
technologies. During multiple case studies, two different ways of AI supply have been
identified. Either, companies develop the capability to build intelligent software themselves or they buy it from a supplier. There are distinct differences to each of the methods. First of all,
