Implementing robotic process automations based on artificial : the role of the buyer-supplier relationship

(1)

Implementing Robotic Process Automations based on Artificial

Intelligence: the role of the buyer-supplier relationship.

Master Thesis - Final version

Supervisor: Dr. Andreas Alexiou

Date of submission: 21th of June 2018

Study and track: Master of Business Administration – Strategy Track

Institution: University of Amsterdam – Amsterdam Business school

Author: Friso Broertjes

(2)

2

Statement of originality

This document is written by Student Friso Broertjes 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.

(3)

3

List of tables and figures

p.

- Framework 1 – Conceptual framework ... 13

- Table 1 – Descriptive statistics, correlation coefficients and scale reliabilities... 26

- Table 2 – Convergent and discriminant validity assessment ... 28

- Table 3 – Goodness of fit statistics ... 30

- Framework 2 – Results SEM ... 31

- Table 4 – Direct effects ... 32

- Table 5 – Indirect effects of trust ... 34

- Table 6 – Total, direct and indirect effects of trust ... 34

- Table 7 – Results post hoc analysis ... 37

- Appendix - Table 8 - Questions used for the constructs of the survey ... 66

- Appendix - Graph 1 - Frequency graph with sizes of companies in survey ... 68

(5)

5

Abstract

The buyer-supplier relationship is deemed to have a substantial influence on the success of technological implementations provided by third parties. This study investigates to what extent this relationship also has an effect regarding the implementation of Robotic Process Automations (RPA) that are based on Artificial Intelligence (AI). It uses the concept of the suppliers’ 'desorptive capacity' to uncover the importance of transferring knowledge to the buyers’ side. A mixed method approach is used in order to form generalizable conclusions while also providing explanations for the proposed effects. First, data was analysed that was obtained via a survey with the responses of executives describing their implementation process and relation with the supplier. Then, semi-structured interviews were conducted to give more insights into the effects. The results show the importance of trust and the desorptive capacity of the supplier for the implementation to succeed. The findings deepen the understanding of implementing AI-based RPA technologies and further showcase the usage of the concept of desorptive capacity for explaining successful technology transfers.

(6)

6

1. Introduction

Oxford University predicted that 35% of all jobs will be automated in the year 2035 (Frey and Osborne, 2017), and while this will not to the full extent be attributable to Robotic Process Automation (RPA) or Artificial Intelligence (AI), they are expected to have a substantial role. RPA and similar technological trends are predicted to cause a steady increase in productivity and efficiency among a wide range of industries in the coming years as more and more sectors are starting to notice its benefits (Hirsch, 2017). The increase in demand for these technologies of course will stir up the suppliers market. It is anticipated that before 2020 the domestic and offshore outsourcing of AI-based RPA’s will globally grow between 5% and 12% a year depending on the function and process. Then before 2030, changes in types of supplier-buyer relationships are predicted and rivalry will extensively increase among suppliers (Willcocks and Lacity, 2015).

Due to the rise in interest in RPA and AI, academic research about the implementation and buyer-supplier relationships are also increasing (Lacity and Willcocks, 2015). As RPA and AI will most likely have a more prominent role in business, it is important to understand the drivers of a successful implementation of these technologies. Research in general technology adoption is widespread and is of course applicable when studying the adoption of RPA and AI. However, the implementation of AI-based RPA’s in particular might present some difficulties due to its supposed complexity. Multiple non-academically tested statements have been made, claiming that the addition of AI to an RPA increases the intricacies and raises the need for high quality collaboration between the client and the third party. More time is needed for implementation as they require constant adaptations and are no simple plug-and-play technologies (Schaefer, 2016; Ramaswamy, 2017). This means that general theories of technology adoption perhaps fall short when explaining the implementation of individual

(7)

7 technologies as they fail to take into account its specific characteristics. This shortfall of consideration for the intricacies of individual technologies forms a limitation since it can lead to inaccurate predictions and assumptions. So the implementation of AI-based RPA’s needs further attention, which formed the motivation for this research. First, a general introduction into the subject will be given.

RPA most commonly refers to configuring software to do the work previously done by people (Willcocks, Lacity and Craig, 2015). AI is harder to define as its interpretation is relatively dependent on its context and it encompasses a large variety of subfields. The general consensus is that AI relates to a system using cognitive abilities and processing information with rationality and reason as to adapt behavior to meet its goals in differing environments (Fogel, 2006; Laird, Wray, Marinier and Langley, 2009; Nilsson, 2014). So while no generally accepted definition exists of AI-based RPA, it can be referred to as software doing work previously done by people using a sense of reasoning similar to that of a person. Since most firms do not possess the necessary capabilities to design an AI-based RPA, suppliers are mostly contracted to make it happen. As AI-based RPA has its complexities, a knowledge gap might exist between buyer and supplier and in this situation, the social aspects have proven to influence as they can help with the knowledge transfer (Kim, Kim and Cho, 2014). Relational embeddedness like informal interactions, norms, trust and identification have shown its importance for the successful implementation on an organizational level (Nahapiet and Ghoshal, 2000). A concept that has been used more and more to explain the success of technology transfer is the ‘desorptive capacity’ of a firm. It is described as “a firm’s outward technology transfer capability” by Lichtenthaler and Lichtenthaler (2010).

Relating to this, this research will fill multiple gaps in the literature. While absorptive capacity is a concept that has been thoroughly studied and applied (Cohen and Levinthal, 2000; Zahra and George, 2002), the concept of desorptive capacity is less explored. The founders of

(8)

8 the concept have tried to instigate further researches into its appliance (Lichtenthaler and Lichtenthaler, 2009, 2011) and this research will fill this gap by using it to explain the success of AI-based RPA implementations provided by third parties. Both the effect it has on the implementation performance as the factors leading up to desorptive capacity will be researched. Also, as mentioned, the implementation of AI-based RPA’s has not been researched extensively. Information Technology Outsourcing (ITO) is a similar and widely studied section of professional service relationships but it differs from contracting suppliers for AI-based RPA’s (Lacity, Khan and Willcocks, 2009). In ITO, more responsibility is given to the third party and also it is likely that the organization outsources the services not because it is incapable to do it themselves but more from an economic standpoint (Goles and Chin, 2005). Furthermore, many academic papers have been written about the adoption of salesforce automations (SFA), enterprise resource planning (ERP) etc. but the implementation of AI-based RPA’s still forms a gap in the literature. As the implementation has been assumed to be more difficult due to its complexities and proposed embeddedness in the organizational structure, it calls for research. And finally, often in researches, the success of implementation could be partly attributed to the quality of the relationship between buyer and supplier, but why this is the case is left open for interpretation. The aim is to form a more comprehensive conclusion, supplement the evidence for the effects and offer practical explanations

So this research will investigate which factors contribute to a successful implementation of an AI-based RPA and this will consist of both social aspects as factors during the process. The research question is: how does the relationship between buyer and

supplier affect the implementation of a Robotic Process Automation based on Artificial Intelligence? It will be investigated whether the state of the relationship affects the eventual

(9)

9 capacity, and finally, it will aim to uncover why this is the case. The approach of the research will be from the clients’ perspective and will focus on the organizational level.

The results are of course important due to the enormous rise in interest in both RPA as in AI. Many firms are considering implementing and so it is worthwhile to investigate how this can be done most effectively. Traditional companies underestimate the need to prioritize AI and RPA in their strategic roadmaps (Kaivo-oja, Virtanen, Jalonen and Stenvall, 2015; Lauterbach and Bonim, 2016) and so research might be an effective way to show its possibilities. It is important for buyers seeking to find out what makes an implementation successful. This relates to the qualities of a supplier and to factors which must be given attention during the implementation process. But as the results will indicate which relational and procedural factors lead to a successful implementation, they might especially be important for suppliers. Since it is assumed that the suppliers market has still not matured to the full extent, it is important for suppliers to think about how they can differentiate in order to gain and sustain a competitive advantage. This research will hopefully illustrate how they should design the social side of the relationship, how much interaction there must be during the process, to what extent they must make the client understand the RPA and also give the supplier insights into why this is important.

The research consists of a mixed methods approach, specifically a sequential explanatory design (Cresswell and Cresswell, 2017). First, in the quantitative part of the study, a survey will be analyzed that consists of responses from 153 executives of firms that have recently implemented AI-based RPA’s. The questions refer to their relation with the supplier and the eventual success of the implementation. Constructs were created for the relevant variables, a confirmatory factor analysis was performed in order to check the reliability and validity, and finally via a Structural Equation Model (SEM) results about the effects are established. This part of the research will provide the main valid arguments for the conclusion,

(10)

10 but it will also serve as input for the qualitative part. This will consist of interviews that will be held with people that have recently been involved in the implementation of RPA’s and the relationship with the suppliers. Questions will be related to the outcomes of the quantitative part. This part serves multiple goals as it aims to increase the validity of the results, it enables some exploration and it gives practical insights into the reasons behind the results and effects. In the literature review, an overview of the existing research that has been conducted on the implementation of technologies will be given, the reasons for the research direction will be further explained, the hypothesized effects will be stated and a conceptual framework will be shown. After this, the data and methods section will consist of a further clarification of the data analysis techniques, a description of the way the data was collected and a section explaining how the constructs were measured. This will be followed by the results consisting of the outcome of the confirmatory factor analysis and the SEM results. Then, the qualitative results are presented in a post-hoc analysis. Both the results from the quantitative and the qualitative part will then be interpreted in the discussion which also describes the theoretical contributions, managerial implications, limitations and future research propositions. Finally, the research question will be answered in the conclusion which is followed by the reference list and appendices.

2. Literature review and hypotheses

2.1 Overview literature technology implementation

The goal of the research is to determine the influence of certain factors on the implementation success of AI-based RPA’s, and also why they are important. The factors that will be analysed both in the quantitative and the qualitative part are preselected based on theory, and so the

(11)

11 literature review will elaborate upon the articles and theories which explain possible factors that make the RPA implementation fulfil the organisation’s plans and expectations. While the focus is on the implementation of AI-based RPA, other literature describing the adoption of other technologies will be cited as well. The literature focussing specifically on the implementation of AI-based RPA’s is scarce and theories about general technology adoption are useful to understand the common aspects. First, a general overview will be given of the way in which theories try to explain factors leading up to effective and efficient implementations of innovations / technologies. Then, the subsequent sections will discuss the formed hypotheses with the corresponding literature.

Successful technology implementation can be explained in multiple ways; some explain it by looking at predictors on a personal level (Currie, Michell and Abanishe, 2008; McBride, Carter and Ntuen, 2012), other areas of research focus on the organisational level (Al Salti and Hackney, 2011; Cram, 2009) and some explain it in a sequential manner in which the adoption of automation systems must be seen as adoption in two stages and individual use follows from organisational adoption (Hausman and Stock, 2003; Parthasarathy and Sohi, 1997).

Personal adoption is an important factor for a successful implementation and major models that are used to explain this are the technology acceptance model (Davis, Bagozzi and Warshaw 1989; Venkatesh and Davis, 2000), the expectation-confirmation model (Bhattacherjee, 2001) and the theory of planned behaviour (Ajzen, 1991). In these models, the way in which the technology is perceived (including the additional effects) by the worker are considered to be the major drivers of the performance. Factors on a lower level that have been proven to influence the implementation performance are personality factors of workers (extraversion, intuition etc.) (McBride et al. 2012), hierarchical factors like the alignment between top management team (TMT) and supervisors (Cascio, Mariadoss and Mouri, 2010;

(12)

12 Homburg, Wieseke and Kuehnl, 2010; Meredith, 1987) and personal goal orientations (Jelinek, Ahearne, Mathieu and Schillewaert, 2006). These lower level factors have been proven to be influenced by factors as trainings, customer pressure and peer use.

The influencing factors on an organizational level can be divided into a few segments. Internal and interorganizational social dimensions (trust, commitment etc.) are firstly major influences on the performance, since they have extensive reach in all parts of the process (Haried and Ramamurthy, 2009). However, the way they are considered to be important differs, as some consider them as independent variables in the process of actions and outcome (Ahn et al., 2016), some consider them as mediatoring variables (Ndbusi, 2011), while others consider them dependent variables of the actions (Flemming and Low, 2007; Gefen, 2004; Keilor, Bashaw and Pettijohn, 1997). Of course, many of the constructs are correlated with each other so this compares to a story of ‘the chicken or the egg’. Secondly, the actions, or behavioral dimensions (participation, information sharing etc.) of the organisations during the implementation process have substantial influence on the success (Lee and Kim, 1999). And a final consideration lies in the attitudes and ways of doing things in organisations that may be incompatible with the implementation of the automation (Hayes and Jaikumar, 1991).

2.2 - Hypotheses

The hypothesized effects will be explained in the following sections. First, a general explanation of the conceptual framework will be given after which the argumentation and theoretical foundation for the individual hypotheses will be described in the sequential sub-alineas.

In the framework, it is hypothesized that good relational factors between buyer and supplier lead to a better information exchange during the process and this betters the ability of the supplier to transfer valuable knowledge of the RPA to the buyer. These factors will result

(13)

13 into a better implementation performance. The two social constructs are the trust the buyer has in the supplier and the informal relationships between them. The choice for these variables will be explained later. The results will thus indicate if the (proposed) effect that the social constructs might have on the implementation performance can be explained by a successful transfer of knowledge, and if so, to what extent this can be explained by a qualitative exchange of information. The social constructs are considered as the independent variables, the quality and information exchanged and desorptive capacity as the mediators and the implementation performance as the dependent variable. The desorptive capacity is in this case not approached as the universal quality of the supplier but seen in the relational context. Both the direct effects of the social constructs on the dependent variable and the mediating variables, as the indirect effects via the mediators will be tested. From now on, the success of the AI-based RPA implementation will be defined as implementation performance. The informal social relationships with the supplier and the trust of the buyer in the integrity of the supplier are occasionally jointly defined as the social constructs. The hypotheses are shown in the conceptual framework and will now be explained.

Framework 1

(14)

14

2.2.1 Desorptive capacity’s relation to implementation performance

The term desorptive capacity was formed in 2004 by Lichtenthaler, Lichtenthaler and Ernst as a complementary concept to absorptive capacity and it is used as a construct that describes to what extent a party is able to effectively make the other party understand certain technological knowledge. They had formulated a theory that states that organisational mechanisms influence a firm’s outward technology and knowledge transfer performance through desorptive capacity (Lichtenthaler and Lichtenthaler, 2010). The term ‘desorptive capacity’ can be seen as a bundling of previously existing terms. Al Salti and Hackney (2010) reviewed literature about ITO and distincted four factors that affect the knowledge transfer of information systems between companies: knowledge-related, recipient-related, source-related and relationship-related factors. The source relationship-related and recipient relationship-related factors can be seen as precursors to desorptive capacity and absorptive capacity respectively. Here, the source related factors had a great positive influence on the success of the ITO. Since the term has received a more prominent position in business studies, authors have been using it as a predictor for performance in supplier-buyer relationships. Ahn et al. (2016) concluded that desorptive capacity has a direct positive influence on implementation performance since it enables firms to crystallize their external knowledge. It has also been tested where desorptive capacity stands in the relationship between actions and results. Ahn et al. (2016) concluded that the relation between a firms’ openness and firm performance was mediated by desorptive capacity; Kim et al. (2014) stated that it moderates the relationship between R&D intensity and firm performance. In all researches, the desorptive capacity of the supplier positively influenced the successful transfer of technology, and so this will also be expected in this research.

H1: There is a positive direct relationship between the desorptive capacity of the supplier and

(15)

15

2.2.2 The relation between the quality of information exchanged and the desorptive capacity.

The positive and direct effect of information sharing on implementation performance has been researched and confirmed many times (Erffmeyer and Johnson, 2001; Gregory, Beck and Prifling, 2009; Haried and Ramamurthy, 2009). However, the path of this effect can still be open to interpretation and multiple articles state that this effect is due to the influence it has on the transfer of knowledge. Rashed, Azeem and Halim (2010) distinguished between the sharing of information and the transferring of knowledge. Information was described as day to day operational particulars which will be processed differently by each individual, while knowledge represented the know-how. Information will transform into knowledge as critical thinking, review and reflection are applied. They found that information that is both timely and relevant to the context of the business has a strong direct impact on the knowledge developed from its use. The positive effect of the qualitative sharing of information existed because it led to better individual know-how. The assumption of the indirect effect was confirmed in a different way by Lee (2001) who stated that organisational capabilities moderate the relationship between knowledge sharing (implicit and explicit) and implementation performance. Multiple researches on outsourcing found that knowledge sharing processes (implicit and explicit) lead to a higher level of shared knowledge which in its turn improved the outsourcing performance (Blumenberg, Wagner and Beimborn, 2009; Lau, Tang and Yam, 2010). Since the introduction of its concept, the desorptive capacity has been researched a few times examining the influence of information sharing. Ahn et al. (2016) found that the effect of open innovation, the use of both purposive inflows and outflows of knowledge, on performance was mediated strongly by desorptive capacity. Also, the factor of knowledge management in relationships had no significant direct effect on firm performance but was also mediated by the desorptive capacity, while integration, search capacities, and desorptive

(16)

16 capacity all had significant direct effects on firm performance. Meinlschmidt, Foerstl and Kirchoff (2016) quite literally found that the explanatory power of desorptive capacity lies in how firms can facilitate knowledge application at the recipient which is based on the sharing of information. Based on the literature, the second hypothesis is formed.

H2: There is a positive direct relationship between the quality of the information exchanged

between buyer and supplier and the desorptive capacity of the supplier.

2.2.3 The relation between the social relationships and implementation performance.

First, the choice for the social constructs will be explained. Explanations for implementation performance can be traced back to theories like the relational view, the transaction cost theory, the social exchange theory or the commitment-trust theory (Blau, 1968; Grover, Cheon and Teng, 1996; Kern, 1997; Klepper, 1995; McIvor, 2005; Morgan and Hunt; 1994). These theories play a part in most of the literature related to all sorts of technology adoption with a third party involved and researches derived from these theories have resulted into sets of relational factors that are generally accepted to improve implementation performance. However, some social constructs still produce mixed results. For example, having a relationship based on mutuality was found to be both positively as well as negatively associated with performance (Koh, Ang and Straub, 2004; Lacity et al. 2009; Lee and Kim, 1999). Other relational factors such as flexibility, adaptations, cultural compatibility and coordination have been found to be positively associated with performance, but sometimes the results were not significant or the effect was not that high (Fleming and Low, 2007; Goles and Chin, 2005; Haried and Ramamurthy, 2009). However, the relational dimensions of trust and good (informal) social relationships are generally accepted and confirmed to be great predictors of the success of the implementation.

(17)

17 The definition of a good social relationship is often interpreted differently but in this research the term relates to personal relations, informal interacting time and frequency of communications. Attention is given to articles that explicitly separate good social relationships from other social constructs like trust, commitment etc. Good social relationships have been termed many times as crucial in processes of ITO and implementation of technologies (RPA’s, Salesforce Automations, Enterprise Resource Plannings) supplied by third parties (Bush, Moore and Roco, 2005; Gefen, 2004; Li and Li, 2009; Willcocks et al. 2005). Multiple arguments have been given as to why social relationships have a direct positive effect on implementation performance. Rashed et al. (2010) found that buyer-supplier relationships with closer contact make suppliers search for better ways to meet the needs of the buyers. In what stage good social relationships are important was researched by Athaide, Meyers and Wilemon (1996) who found that effective management of seller-buyer relations requires close cooperation (product co-development, seeking feedback etc.) throughout all phases of the commercialization process. This was also confirmed by Kronawitter, Wentzel, Turetschek and Papadaki (2009) who found that having active and adequate social relationships was a success factor throughout all stages (from preparation to post-deal) in the ITO process. This in contrast to factors of acceptance, fitting cultures and commitment which were only important in particular stages. A good social relationship seems to promote more face-to-face meetings (Hanna and Daim, 2007), mitigate any raised risk (Frisanco, Anglberger, Ang and Onu, 2008), provide a better insight into the requirements of the client (Qi, 2008) and enable better negotiation processes to lessen misunderstandings (Rhodes, Lok, Loh and Cheng, 2004). However, the importance of proper social relationships between buyer and supplier seem to differ regarding the type of technology implementation. Burgess and Gules (1998) found that better informal buyer-supplier collaboration is more important for soft technology than for hard technology implementation which would be due to the systematic nature of soft technology.

(18)

18 And so, as AI-based RPA’s can be defined as soft technologies, the third hypothesis can be formed.

H3: There is a positive direct relationship between the buyer-supplier social relationships and

the implementation performance of the RPA.

2.2.4 Trusts’ relation to implementation performance

As stated, the second constructs that is steadily found to be positively related to implementation performance is the trust of the buyer in the integrity of the supplier. The direct effect of trust on the success of the implementation of technology provided by a third party has been confirmed many times (Kern, 1999; Lee and Kim, 1999; Lindgreen, 2003; Mao, Lee and Deng, 2008; Niu, 2010; Sabherwal, 1999) and trust has been defined as the most important of the variables in relational exchange theory (Blau, 1968; Homans, 1985). For a successful incorporation of a technology, trust is needed within an organisation (Handfield and Bechtel, 2002; Susan and Holmes, 1991) as well as between organisations (Hart and Saunders, 1997; Hausman and Stock, 2003). In the research about software outsourcing by Kanawattanachai and Yoo (1997), teams that were outperforming others succeeded in both establishing trust in the beginning, as well as maintaining this trust throughout the project. Zaheer, McEvily and Perrone (1998) investigated the role of trust in interfirm exchange and concluded that inter-organisational and interpersonal trust are two distinct constructs and influence the performance in different ways. It is described as the expectation that a party will act predictably, fulfill its obligations, and behave fairly even when the possibility of opportunism is present. Trust fulfils multiple functions: it allows focussing on long-term objectives, it suppresses opportunism, it enables risk taking, reduces conflicts, improves responses to crises and enables a reduction in transaction costs (Klepper, 1995; Rousseau, Sitkin, Burt and Camerer, 1998). Also, the buyer

(19)

19 is not able to monitor and control all the details in every exchange, so regarding the implementation of technologies with third parties involved, trust is a way to alleviate part of this obstacle (Mao, Lee and Deng, 2008). Sabherwal (1999) explained that trust limits the need for structure because the urge to protect itself is lessened. However, a balance between structure and trust in organisations must always be kept. Lander, Purvis, McCray and Leigh (2004) explain that when knowledge comes from a source which is trustworthy, the application of this knowledge tends to be less challenging. Trust is also one of the major drivers to evolve the relationship from a short commitment to a scenario of being long term partners (Zviran, Ahituv and Armoni, 2001). Taken into account that all these theories point in the same direction, the fourth hypothesis can be formed.

H4: There is a positive direct relationship between the trust of the buyer in the supplier and the

implementation performance of the RPA.

2.2.5 The relation between the social constructs and the quality of information exchanged.

Although the effect of the social constructs on the performance has been agreed upon in the literature, some articles concluded this to be an indirect link. The direct effects of the social constructs on the quality of information exchanged will be discussed jointly. Social capital in a relationship between buyer and supplier, which is intangible has been found to enable effective information and knowledge transfer (Rottman and Lacity, 2008). Kotlarsky and Oshri (2005) researched if social ties and knowledge sharing contributed to a successful collaboration in distributed Information System (IS) development and found that if the buyer lacks trust in the supplier, the information sharing would decline. Later, Oshri, Kotlarsky and Wilcocks (2007) found that socialization is important for globally distributed teams and their success was mainly based on enough frequent face-to-face meetings and on the activities and processes

(20)

20 employed before and after these meetings. These interpersonal ties stimulated the workers to exchange tacit knowledge informally. Many of the difficulties in transferring information between the source and the recipient have been found to stem from human resource issues (Ranft and Lord, 2002). These difficulties in the transfer of knowledge were proven to be more apparent in companies with high turnover rate as this hampered the development of the building of trust and social relationships. Companies seemed to benefit from positive social settings as this improved the informal sharing of information. It has been proven that a qualitative informal partnership increases the effective sharing of tacit knowledge (Blumenberg et al., 2009; Lee and Qualls, 2010). This literature supports the formation of the fifth set of hypotheses.

H5a: There is a positive direct relationship between the buyer-supplier social relationships and

the quality of information exchanged.

H5b: There is a positive direct relationship between the trust of the buyer in the supplier and

the quality of information exchanged

2.2.6 The relation between the social constructs and the desorptive capacity.

Sufficient desorptive capacity is often defined as an objective quality of a supplier that is constantly able to effectively transfer knowledge, but this is in reality dependent on the context. And so, social dimensions between companies have been found to have a significant effect on the desorptive capacity (Kim et al. 2014). It has also been confirmed that a firms’ openness and inclination to communicate has a direct positive influence on the desorptive capacity (Ahn et al. 2016). Dell’Anno and Del Giudice (2015) examined the role of absorptive and desorptive capacity in technology transfers and concluded that good interaction was a necessary condition for developing organisational capabilities that improved the sharing of knowledge. This was also found by Meinlschmidt et al. (2016) who stated that managers should build relationships

(21)

21 with suppliers because this would better the desorptive capacity as it would allow the efficient transfer of sustainable knowledge and avoid exchanges of un-useful information. The social aspects were also proven important when choosing intermediaries and had to be taken into account as these influence the desorptive capacities of organizations (Ziegler, Ruether, Bader and Gassmann, 2013). Based on these researches, a direct link between the social dimensions and desorptive capacities can be expected.

H6a: There is a positive direct relationship between the buyer-supplier social relationships and

the desorptive capacity of the supplier.

H6b: There is a positive direct relationship between the trust of the buyer in the supplier and

the desorptive capacity of the supplier.

2.2.7 The mediation effect of desorptive capacity and the quality of information exchanged.

The last two hypotheses are aimed at uncovering possible mediation effects of desorptive capacity and the quality of information exchanged. Besides researches claiming that the quality of information exchanged and the desorptive capacity increase implementation performance and are improved by social constructs, they have also been regarded multiple times as mediators. There have been researches concluding that the main reason why trust and a good social relationship lead to increased performance was the benefit of free communication / better information sharing (Holton, 2002; Khan and Khan, 2013; Rottman and Lacity, 2008). In some of the mentioned articles, the desorptive capacity acted as the mediator which acted as the final and only influencer on performance (Kim et al., 2014) or it was concluded that openness (trust between parties, readiness to collaborate etc.) only affected implementation performance in an indirect way via desorptive capacity (Ahn et al., 2016). And so, the final hypotheses expect

(22)

22 both the quality of information exchanged and the desorptive capacity to mediate the effect between the social constructs and the implementation performance.

H7a: The effect between the buyer-supplier social relationships and the implementation

performance is mediated by the quality of information exchanged and the desorptive capacity of the supplier.

H7b: The effect between the trust of the buyer in the supplier and the implementation

performance is mediated by the quality of information exchanged and the desorptive capacity of the supplier.

3. Data and methods

3.1 Data analysis

The research will take on a mixed methods approach by which a qualitative part will be used to validate and further explore the findings of a quantitative model. This approach allows enjoying both the benefits of the quantitative as the qualitative method: it is useful for the validation of quantitative results, the output of the quantitative part serves as input for the qualitative part, it allows viewing the research subject from multiple angles, it enables a better clarification of the results and explains possible causal effects, while it still offers a certain level of generalizability (Cresswell and Cresswell, 2017; Neuman, 2013). However, discrepancies might arise between the quantitative and the qualitative results but this can improve the understanding when looked at possible reasons for the differing results. The qualitative part consists of multiple interviews that can lead to a more substantiated conclusion and can help explain peculiarities. The method will be discussed further in the post hoc analysis. The quantitative analysis is performed using a SEM. This model has generally been

(23)

23 accepted as a suitable instrument for testing relationships between multiple latent variables (Bollen and Long, 1993). Wu and Zumbo (2008) explain that using a SEM model has multiple advantages. SEM incorporates multiple observed indicators to represent the latent variables and this approach would make the latent variables measurement error free, thus creating a more realistic view of the strength of the association. Secondly, it is more flexible when incorporating multiple independent variables and mediators into a single model. Finally, testing mediations using an SEM provides informative goodness-of-fit indices for the models. However, the coefficients of the SEM must be interpreted through the lens of established theories as they do not explain truly causal effects. Thus, the sequential qualitative part is again useful to enhance this interpretation. In this research, AMOS 21.0 was used to analyse the data and run the SEM (Arbuckle, 2014). A confirmatory factor analysis (CFA) was performed to test the latent variables as to ensure a certain rate of validity and reliability. Furthermore, using the CFA, the ‘goodness of fit’ of the model is tested.

3.2 Data collection

The data of the quantitative model comes from a survey that was filled in by 153 executives of companies that recently adopted an RPA that was based on AI. The data were collected by dr. A. Alexiou, prof. I. Oshri and dr. S. Khanagha in 2016 via a professional firm. The dataset has no missing values and the surveys were filled in by one respondent per firm. The targeted firms ranged in the number of employees, the smallest having ten or fewer employees and the largest having over a million, but the majority (57%) had between a 1000 and 10.000 employees. Descriptives of the numbers of employees can be seen in graph 1 in the appendix. As the bulk had a size in this range, the results are mainly representative for larger firms. They operate in different sectors (‘financial services’, ‘manufacturing’, ‘retail, distribution and transport’, ‘public sector’, ‘IT/computer services’ and other commercial sectors). As shown in graph 2 in

(24)

24 the appendix, each of these sectors was represented by between 20 and 30 firms which makes the results representative for all of these sectors. As there is a possibility that the effect between trust / social relationships and performance might be different depending on the size of the firm and industry, they might influence the results of the regression, and so these two variables were held as control variables. Some requirements were set for the respondents of the survey and their associated organisations. The organisation must have contracted a third party provider to automate a task previously performed by a human, the automated solution needed to have artificial intelligence/learning abilities, and the respondent must have been familiar with how the automated solution was implemented in the organisation.

3.3 Measures

The survey started with descriptive questions about the organisation and automation. Then, questions relating to the implementation and relationship with the supplier were asked which were segmented per variable. For all of the variables, unless mentioned otherwise, statements about the construct were presented and the respondent had to react according to a five point Likert scale and thus explain to what extent they agreed or disagreed.

First, the social relationships were measured using the scale developed by Zimmermann, Oshri, Lioliou and Gerbasi (2017). Example items are: ‘we maintain close social working relationships with some members of the automation solution supplier organisation’ and ‘we know some members of the automation solution supplier organisation on a personal level’. Trust was measured by the scale of Doney and Cannon (1997) and exampling statements are: ‘our automation solution supplier keeps promises it makes to our firm’ and ‘our automation solution supplier is not always honest with us’. Two statements were asked in a reversed scale to avoid the tendency to rate all questions with similar values, which would meddle with their true attitudes (Couch and Keniston, 1960). For the concept of desorptive capacity, an approach

(25)

25 has been chosen in line with the general consensus of the definition using nine statements relating to the ability of the supplier to provide the buyer with enough knowledge of the RPA and to what extent the supplier provided the buyer with a clear understanding of the possibilities of adaptations. Example items are: ‘our automation solution supplier successfully provided us with a clear understanding of the benefits of Robotic Process Automation (RPA)’ and ‘our automation solution supplier successfully provided us with a clear understanding of the key underlying components of Robotic Process Automation (RPA)’. As the individual respondent in this survey is not able to judge whether its supplier can successfully and consistently transfer the knowledge and as such, is limited to his cognition in this particular case, the data of desorptive capacity are contextually dependent on the aspects of the relationship. The quality of information exchanged between the buyer and supplier was measured using the scale developed by Krishnan, Martin and Noorderhaven (2006) which consists of three statements. Two example items are: ‘our automation solution supplier firm has provided relevant information whenever we asked them for it’ and ‘we are promptly notified by our automation solution supplier whenever any major change occurs at their firm’. And finally, for the performance variable, the interviewee had to indicate to what extent the RPA implementation fulfilled their organisation’s plans and expectations. They had to rate three aspects; the quality of the solution (compared to expected technical specifications and performance), the duration of implementation (compared to the original plan for implementation), and incurred costs (compared to the original budget for implementation). These three items needed to be rated according to a rating scale ranging from 0 - 10% to 90 - 100%. All the questions and their descriptives are presented in table 8 in the appendix.

(26)

26

4. Results

4.1 Descriptives of the measures

Table 1

Descriptive statistics, correlation coefficients and scale reliabilities

Variables M SD Correlations

Soc.Rel Trust DCAP Info Q. Performance Firm size Industry

Model Soc. Rel. 3.5 .8 .692 Trust 3.7 .7 .497** .796 DCAP 3.9 .7 .513** .571** .889 Info quality 4.0 .8 .461** .541** .541** .690 Performance 7.2 1.9 .234** .425** .422** .374** .841 Firm size 7.0 1.9 -.052 -.097 -.074 -.006 -.091 Industry 3.6 1.7 -.077 -.039 -.031 -.058 -.144 -.144 * p < 0.05 level., **p < 0.01 level

Notes: Bold values on the diagonal represent the Composite reliability

In table 1, the means, standard deviations, correlation coefficients and scale reliabilities are shown. What is noticeable at first, is that all variables are significantly correlated to each other. As recommended by Baron and Kenny (1986), for a mediation effect to be proven, mediators firstly have to be significantly correlated with both the predictor variable(s) and the outcome variable(s). In coherence with the expectations, trust and social relationships are correlated positively with performance, quality of information exchanged and desorptive capacity which makes the mediation effect plausible. Furthermore, the quality of information exchanged is positively correlated with desorptive capacity which in its turn is positively correlated with performance. However, it is evident that some differences in strengths of correlations exist. It

(27)

27 can be noticed that trust is correlated stronger with desorptive capacity, quality of information exchanged and performance. There is especially a significant difference between the correlations of social relationships - performance and trust - performance.

Two questions were removed to improve the reliability of the model. The pitfalls of these questions were the reversed scales, which were apparently not well noticed. After finalizing the model, the scale reliabilities of the constructs were assessed using their Composite Reliabilities (CR). This is a reliability instrument to measure the internal consistency and is chosen for being a better alternative to the Chronbach Alpha which has the limitation that it assumes the same factor loadings for all items. CR considers the varying factor loadings of each item which is why it is often preferred (Shook, Ketchen, Hult and Kacmar, 2004). The threshold for CR is at 0.7 for the construct to be defined as truly reliable, but certain articles have appointed 0.6 also as acceptable (Awang, 2012; Hulland, 1999). The variables of ‘trust’, ‘desorptive capacity’ and ‘performance’ show high reliability and the constructs of ‘social relationships’ and ‘quality of information exchanged’ show a slightly lower one. But as 0.6 is considered acceptable for newly developed scales, and since the two values are relatively close to 0.7, this is not seen as a major limitation (Hair, Black, Babin, Anderson and Tatham, 1998). It does mean to some extent that the items relating to the constructs of ‘social relationships’ and ‘quality of information exchanged’ are slightly less measuring the same construct.

4.2. Confirmatory factor analysis

4.2.1 Validity and reliability

SEM is used to test the strengths of the effects, but before the hypothesized structure is tested, the quality of the measurement constructs must be assessed further (Anderson and Gerbing, 1988). A CFA was performed to see how well the proposed model accounts for the correlations

(28)

28 between the variables in the dataset and to further assess the levels of validity and reliability. Table 2 shows the average variance extracted (AVE), maximum shared variance (MSV) and on the right, a factor correlation matrix with the square root of the AVE on the diagonal.

Table 2

Convergent and discriminant validity table with a factor correlation matrix

Using this table, the convergent and discriminant validity can be evaluated. First, the convergent validity relates to the consistency of an observation variable that measures a latent variable, in other words, how the different measurements relate to each other under the same construct. For a construct to have sufficient convergent validity, the AVE should be above 0.5 (Hair et al., 1998). In this model, only the construct of performance has an AVE of above 0.5 which harms the convergent validity as it means that the latent factors are not well explained by the observed variables. This counts as a limitation of the model. Then, sufficient discriminant validity would indicate that the measurement model of a construct is free from redundant items, or in other words, it would indicate that constructs that are seemingly unrelated are in effect truly unrelated. This is proven when, either the MSV is smaller than the AVE, or the square root of the AVE is greater than the inter-construct correlations (Hair et al., 1998). The discriminant validity thus also forms a limitation as, again, only the ‘performance’ construct meets these criteria. The risk with a lack of discriminant validity is that variables in

Variables AVE MSV Soc.Rel Trust DCAP Info Q. Performance

Soc.Rel .363 .624 .603

Trust .400 .731 .790 .632

DCAP .473 .557 .668 .746 .688

Info Q. .427 .731 .670 .855 .709 .654

(29)

29 the constructs correlate more with variables outside the parent factor, which means that the latent variables can perhaps be better explained by a different factor (Bagozzi and Yi, 1988). Eventually, this lower validity makes the conclusions perhaps less legitimate.

4.2.2 Model fit

The indicators relating to the goodness-of-fit of the model are shown in table 3 which refers to how well the proposed model accounts for the correlations between the variables in the dataset. There are many indices that can be calculated to determine the goodness-of-fit, but Hooper, Coughlan and Mullen (2008) report that while there is no golden rule, some indices are essential because they reflect different aspects of the model fit, and thus form a more comprehensive model. The model Chi-square together with its degrees of freedom and associated p value should be reported at all times. Further, the Comparative Fit Index (CFI), Tucker Lewis Index (TLI), Root Mean Square Error of Approximation (RMSEA) would have to be included according to Hu and Bentler (1999). The Goodness of Fit Index (GFI) and Adjusted Goodness of Fit Index (AGFI) are not presented since these are often only added for historical reasons rather than for sophistication (Hooper et al., 2008; Sharma, Mukherjee, Kumar, & Dillon, 2005). Also, the p of Close Fit (PCLOSE) is added as it can help understand the sampling error in the RMSEA.

(30)

30 Table 3 Goodness-of-fit statistics Model Level of acceptance Literature

Chi-square / df 1.565 <3.0 Wheaton et al. (1977)

P-value .000 >.05 Wheaton et al. (1977)

RMSEA .061 <.08 Browne and Cudeck (1993)

PCLOSE .059 >.05 Hu and Bentler (1999)

CFI .903 >.9 Bentler (1990)

TLI .89 >.9 Bentler and Bonett (1980)

Although the preparation of the model enabled the indices of CFI and RMSEA to also attain the level of acceptance, two indices are still short. First, the absolute fit indices determine how well a model fits the sample data and gives the best demonstration of how well the proposed theory fits the data (McDonald and Ho, 2002). The Chi-square is the traditional measure to evaluate the overall model fit, and it is divided by the degrees of freedom in order to minimise the effect of the sample size. Wheaton et al. (1977) rate a value of below 5.0 as acceptable and below 3.0 as good, so the value of 1.565 means an adequate model fit. The corresponding p-value, however, does not attain the level of acceptance which means that the null-hypothesis would need to be rejected and there would be an association between the variables of the model. Then, the RMSEA and corresponding PCLOSE tell to what extent the model, with unknown but optimally chosen parameter estimates, would fit the populations covariance matrix (Hooper et al., 2008) and both attain the recommended thresholds of below .08 and above .05 respectively.

Then, the incremental (or relative) fit indices, unlike absolute indices, rely on the comparison of the chi-square with a baseline model (null hypothesis) and for these models, the null hypothesis is that all variables are uncorrelated (McDonald and Ho, 2002). The CFI is a

(31)

31 measure that takes into account the sample size and this measure was attained. Finally, the TLI was with its .89 slightly under the level of acceptance of .9.

As two of the six indices did not attain the recommended value, there is inconsistency between the correlations proposed and the correlations observed for the model (Awang, 2012). Although the model has been revised by removing some unengaged responses and by excluding questions from the analysis, still the model does not have the best fit. Reasons could be a too small sample size or too extensive constructs (Hooper et al., 2008). This serves as a limitation of the model. However, this problem can be contained to some extent, as the qualitative part lends itself for a further validation.

4.3. Results SEM

The results of the regressions are presented visually in framework 2 and the corresponding statistics are presented in table 4.

Framework 2

Results SEM

(32)

32 Table 4

Direct effects

Consequent

Info quality DCAP Performance

Antecedents Coeff. SE p Coeff. SE p Coeff. SE p

Soc. Rel .151 .071 .036 .198 .069 .001 -.249 .218 .255 Trust .557 .356 .002 .341 .086 .000 .664 .279 .019 Info quality - - - .219 .078 .006 - - - DCAP - - - .662 .255 .010 Constant 1.154 .356 .002 1.054 .349 .003 2.626 1.112 .020 R2 .415 R2 .443 R2 .260 F(4, 148) = 26.235, p = .000 F(5, 147) = 23.419, p = .000 F(6, 146) = 8.550, p = .000

It is shown in framework 2 that almost all hypothesized relationships are significant except the direct and indirect links of social relationships with performance. First, the direct effects of table 4 will be elaborated upon.

It shows that the first hypothesis H1 can be confirmed since desorptive capacity is associated with a significant positive direct relation with the performance of an AI-based RPA implementation (β = .662, p < .05). An increase of 1 on the scale of desorptive capacity is associated with an increase of .662 of the implementation performance. Then, the second hypothesis H2 can also be accepted as a higher quality of information exchanged is related to an increase of the desorptive capacity of the supplier during an implementation process (β = .219, p < .01). It was hypothesized in H3 that good social relationships between the buyer and supplier were associated with an improved implementation performance, but results show that this cannot be confirmed as the effect between social relationships and performance is negative and insignificant (β = -.249, p = .255). On the other hand, trust in the supplier is related to the highest significant increase of implementation performance (β = .664, p < .05). The fourth hypothesis H4 can thus be accepted. Then, the direct effects between the independent social

(33)

33 constructs and the mediators were all found to be significant. The hypothesis 5a claimed that good social relationships relate to a better quality of information exchanged and this has proven to be a small but significant effect (β = .151, p < .05). The same goes for the hypothesis 6a stating that good social relationships relate to a better desorptive capacity of the supplier (β = .198, p < .01). The effects between trust and the mediators have proven to be stronger. Hypothesis 5b can be accepted and it can be stated that higher trust in the supplier relates to a substantial increase in the sharing of information since the coefficient is relatively high (β = .557, p < .01). Finally, H6b can be accepted as an increased trust in the supplier is associated with a significant increase of the desorptive capacity of the supplier (β = .314, p < .01). The hypothesis 7a claimed that the effect between social relationships on performance and implementation performance is mediated by the quality of information exchanged and the desorptive capacity. Baron and Kenny (1986) stated four criteria for a mediation model to be valid. The independent variables must be significantly correlated to the dependent variable (1), the independent variable must affect the dependent variable in a regression (2), the mediator variable must affect the dependent variable in a regression of both the independent variable and the mediator on the dependent variable (3), and the effect of the independent variable on the dependent variable in (3) must be less than in the second criterion (4). As the variable of social relationships does not meet the second criteria due to the insignificant relationship, the hypothesis 7a cannot be confirmed. To confirm the hypothesis 7b, additional statistics are needed and presented in table 5 and table 6.

(34)

34 Table 5

Indirect effects of trust

95% Conf. Int.

Effect SE Lower Upper

Total indirect effect .483 .195 .103 .863

Trust → Info Q. → → Perf. .176 .164 -.159 .494

Trust → DCAP → → Perf. .226 .097 .050 .432

Trust → Info Q. → DCAP → Perf. .081 .051 .013 .209

Table 6

Total, direct and indirect effects of trust

Performance 95% Confidence Interval

Effect of trust Coeff. SE p Lower Upper

Total effect 1.146 .233 .000 - -

Direct effect .664 .259 .012 - -

Indirect effect .483 - - .103 .863

Table 5 shows the specifics of the indirect effects and this allows a better interpretation to test the last hypothesis. In this table, the indirect effects through the individual mediators as well as the serial indirect effect through both the mediators is shown. The individual effect through the info quality variable is also shown to form a more comprehensive image of the contributions the individual paths make to the total indirect effect.

The first indirect effect of trust in table 5 explains that trust in the supplier would relate to a better quality of information exchanged which in its turn relates to a better performance. This effect is positive with a mediation effect of .176 but it is insignificant since the lower part of the bootstrap 95% confidence interval is below zero. The point estimates of the bootstrap interval show the mean over the number of bootstrapped samples and if zero is not present between the resulting confidence intervals, the mediation effect can be defined as significant (MacKinnon, Lockwood and Williams, 2004). On the other hand, the indirect effect of trust in

(35)

35 the supplier relating to a better desorptive capacity resulting in a better implementation performance is significant and positive with an effect of .226. This coefficient of .226 reflects that a substantial part of the total indirect effect is because trust relates to a higher capacity of knowledge transfer which in its turn relates to a better implementation. The indirect effect of trust going through both the mediators and relating to an improved implementation performance is also positive and significant. The indirect effect that trust has on performance is thus substantial (.483). It must be stated that the effect is mostly mediated by desorptive capacity since this takes up the biggest piece of the pie. Table 6 shows the total, direct and indirect effects of trust on performance and it can be stated that the indirect effects take up a substantial proportion of the total effect. As the direct effect is significant and the indirect effect is substantial and significant it can be stated that the effect between trust and the implementation performance is mediated and H7b can thus be confirmed.

5. Post hoc analysis

5.1 Data and methods

One characteristic of the SEM is that no truly causal effects can be confirmed as you can only conclude from the results that associations exist between the latent variables. The qualitative data provide extra insights and allow a better interpretation of the direction of the effects. And so this part has multiple functions; it strengthens the validity of the total research, it gives a certain sense of direction to the effects, it serves as a means to find out why the results are the way they are and it also enables finding out the reasons for certain peculiarities that are found in the quantitative model. It allows the research to dig deeper into the relationships of the variables and lets buyers share their experiences into why certain constructs are important or not.

(36)

36 The qualitative part consists of ten interviews that were conducted with workers from ten different companies that had just recently implemented an RPA which was supplied to them via a third party. The interviews ranged between 35 and 60 minutes and all of them were done in person, except one which was conducted over the phone. An initial interview was conducted as a pilot interview and this was used to test the approach of questioning. After this, the question list was adapted to enable key informants to answer in a more informative way. Most of the interviews were one on one, but two interviews were conducted with two informants of the firm. The respondents had differing functions in the companies ranging from IT-worker to director of the firm, but one of the criteria was that they had to be directly involved in the implementation process and the relationship with the supplier. The companies were present in differing industries such as logistics, insurance, finance and the firms ranged in size from 35 to 80.000 employees. Finally, an interview was conducted with a consultant that was specialized in the implementation process of AI-based RPA’s and was involved in ensuring good relations between supplier and buyer.

The interviews were semi-structured, as a template with questions was arranged but the interviewers were free to deviate from the list and ask further question when interesting insights came up. The questions were formed in a deductive way since they originated from the quantitative part but the importance of refraining from steering the respondents in a certain direction was kept in mind. The questions mainly focused on finding out how the constructs were interrelated in their implementation process and why this was the case. The respondents were also reminded to answer the questions on behalf of the company they worked for and not from a personal perspective as the respondents of the survey were also answering on organizational level.

The interviews were recorded, transcribed and then analyzed using Nvivo 10.2.2. This was done via content analysis in which the data were coded and classified. This method is used

(37)

37 to make sense of the data and highlight the important findings (Neundorf, 2002). Again this was approached in a more deductive way since the relationships which were the subject of analysis were already apparent and the results of the quantitative part were leading. Nodes were first formed which described the relationships between the constructs and then, quotes in the transcripts relating to these relationships were noded. Since the informants often do not explicitly mention the constructs, these had to be interpreted. The way that the variables were measured in the quantitative analysis was leading when interpreting the constructs mentioned in the transcripts. For example, performance comprised of required time, overall quality, and costs of implementation. Both quotes that confirmed and contradicted the relationships were noded.

5.2 Results

In table 7, the results of the analysis are presented and representative quotes are mentioned. In this analysis, the focus was on explaining the direct effects of the variables, and thus, the indirect relations are not present in the table. The effect reflects the general consensus among the respondents relating to a more positive or negative relationship.

Table 7

Relationships and representative quotations

Effect Quotations

Soc. Rel → Performance +- “It is not necessary to have a good social relationship [A] outside of the professional relationship for the business relationship to work [E].” “Sometimes the informal relationship [A] helps to get things done quicker [E]. But even then, such a relationship only works if the supplier finds you important.”

“And eventually, his partner, on a social human level [A], has pulled me into the group of which I am pretty enthusiastic about [E].”

“ “ → DCAP + “So I think it is very important that you understand it (RPA) [C] and feel heard, and that begins on a personal level [A].”

“ “ → Info Quality + “And because of that open culture [A], you quickly notice that the

threshold is lower to discuss a problem [D].”

(38)

38 “Once you are in such an informal relationship [A], you are much more approachable. You will much quicker ask questions [D].”

Trust → Performance + “And that process takes longer [E] because you have to test everything, and also because there is that distrust [B], and that stays.”

“When the trust is good [B] you can blindly follow and things will go as planned [E].”

“ “ → DCAP + “So I think that trust forms the base [B], because you have to trust that the knowledge that the supplier transfers to you [C], that it is correct.”

“ “ → Info Quality +- “Well, because this is just the exchange of information, when I do not

trust him [B], he can just as well exchange the information [D].”

DCAP → Performance + “It matters whether they were successful in transferring the knowledge

[C]. And because we knew about it, the implementation was more successful [E] and certain mistakes were prevented.”

“At the next robotization it is important [E] that the knowledge arrives at us quickly [C] because we are going to work with that package. But also, nobody knows how Microsoft Word works under the hood but everybody can work in Word.”

Info Q. → DCAP + “Those were all those learnings that were not well fixed in the beginning [C], and so the team needed to adjust very frequently, and that just requires a very good and open communication back and forth [D].”

[A] Social relationships, [B] Trust, [C] DCAP, [D] Quality info exchanged, [E] Performance

5.2.1 Social relationships

The opinions about the effect of good social relationships with the supplier on the implementation performance were mixed. When asked about the general influence of social relationships on the performance, multiple complementary positive effects were given. First, it was stated that the process of implementing AI-based RPA’s still comes with new insights because of its novelty and there is no ‘written book’ of the pitfalls which means that no fully complete contract can be written. So because all the actors are still learning, people must also informally work well together as workers constantly bump into new things and this offers more flexibility. Furthermore, when the social relations are good, they know each other better so they understand the mutual situations better, they ‘speak the same language’ and they feel that they are more being heard. Then also, when there is less social distance, workers feel that they work on the same goal which improves the inner motivation of the firm itself resulting in a

(39)

39 better implementation. Finally, a good informal relationship has also led to the supplying firm sending in better people for the implementation.

However, others stated that only the professional relationship was important in the sense that the supplier helps when problems arise but an informal relationship would not help when implementing. Others are in the middle and state that the importance depends on external aspects. The company culture was defined a factor, as a company with an informal atmosphere would also benefit more from good informal relations with the supplier. Then, the type of contract seemed to matter because a relationship based on a licensing agreement would involve more social interactions than simply buying a version of an RPA. Also, the informal relations could help with arranging processes in a more efficient way but then the supplier must also consider the buyer as important. And finally, it was mentioned that it helps to have a good relation with someone but only because it leads to more trust.

When asked specifically for the effect of good social relationships on the transfer of knowledge, the effect was deemed positive and also differences between larger and smaller firms seemed to exist. Especially at larger firms, many different sectors of an organization need to come together when implementing RPA’s and it was stated that no single supplier exists that can provide all the necessary knowledge. That is why project teams are sometimes formed with workers from multiple disciplines and for this to be a success, the informal relations need to be good. Certain large firms send their workers on trainings together with workers of the supplying firm, with the intention of both improving the technical skills as well as the inter-organisational social relationships. Also, the approaches of the suppliers differ between the firms as they all have their own types of security and specific settings and a better informal cooperation helps to transfer this specific knowledge. However, at smaller firms, it was seen more as buying a software package which does not particularly require informal relationships for the knowledge to transfer.

Implementing robotic process automations based on artificial : the role of the buyer-supplier relationship