Robotic process automation and artificial intelligence capabilities driving digital strategy: a resource-based view Free

Viewed as powerful internal aspects of a company, organizational resources and capabilities have been used as a way to achieve competitive advantages (CAs) in the current business scenario. “The finding that CA rather than external environments are the primary source of inter-firm profit differentials between them, focuses attention upon the sources of CA” (Grant, 1991).

According to Grant (1991), companies can bring themselves a CA by possessing a scale-efficient plant, superior process technology, ownership of low-cost sources of raw materials or access to low-wage labor, among other resources. However, these types of resources no longer bring companies CAs for a single reason. Many years after the formulation of the RBV (Grant, 1991; Peteraf, 1993; Wernerfelt, 1984), these resources have become qualifying for competitiveness and are no longer able of improving it. Instead, companies are still looking for CAs but in a different domain.

Globalization and intense digitization have led to the emergence of an increasing number of new enabling technologies that are changing the business landscape as well as the way companies look for new capabilities. Adopting those technologies is an extremely challenging process because a firm must formulate new strategies that align with its culture and workflow (Intalar et al., 2021).

The global market size for one of the most prominent enabling technologies is expected to reach USD 25.66bn by 2027, expanding at a Compound Annual Growth Rate (CAGR) of 40.6% over the forecast period (Research and Markets, 2020). RPA is an IT artifact made up of a set of software, namely Robot, Code Editor and Orchestrator enabled to automate routine tasks performed by humans and capable of modifying the business processes embedded in the information systems infrastructure.

The “RPA is an umbrella term for tools that operate on the user interface of other computer systems in the way a human would do” (van der Aalst et al., 2018). RPA is more advanced than earlier Business Process automation tools because “robots” act like a human inputting and consuming information from multiple Information Systems (IS) (Davenport and Ronanki, 2018). “RPA is the least expensive and easiest to implement of the cognitive technologies and typically brings a quick and high return on investment” (Davenport and Ronanki, 2018).

RPA technology has been seen as important at the tactical and operational, but not on the strategic level of organizations (Lacity et al., 2015a; Sahli and Davenport, 2018; Santos et al., 2019; Šimek and Šperka, 2019; Suri et al., 2018). This view hinders the understanding of the RPA strategic potential, which improves groups of resources by changing business processes and leveraging the firm's performance. In addition, the study by Syed et al. (2020) stating that there is a scarcity of studies that explore the techniques underlying RPA, corroborates what also points the review by Ivančić et al. (2019) that the scientific literature on RPA technology remains insufficient.

By connecting the RBV and RPA technology literature, this paper aims to show the strategic importance of RPA from two perspectives. First, creating Artificial Intelligence (AI) capabilities by automating business processes and improving IS resources management. Second, showing how RPA can help organizations to reach a digital AI strategy.

Furthermore, according to the Systematic Literature Review (SLR), it was not possible to find other research that has studied the technology through the lens of the RBV, not even using the same methodology employed here ( Appendix 2 – Table A1). Thus, this study works toward showing the strategic relevance of RPA through an RBV perspective by suggesting the following main RQ: “How can RPA generate CAs to support an AI digital strategy through managing IS/IT resources and creating AI capabilities?”

For Melville et al. (2004) in their analysis of previous models of IT business value, “IT impacts organizational performance through intermediary business processes.” They also declare that “from the perspective of RBV, business processes provide a context within which to examine the locus of direct resource exploitation.”

This article unfolds in six sections. After the Introduction, Section 2 presents a comprehensive literature review, while Section 3 explains the research methodology. In Section 4, we present the results and their discussion. Section 5 presents the conclusion with the contributions and finaly the Section 6 brings the limitatios and suggestions for future work.

Strategic management inquiry leveraging the RBV has grown tremendously in recent decades to become firmly entrenched as a central perspective for understanding organizations (Helfat et al., 2023). Known as an important theory able to concretely explain the role of the firm's resources as the foundation of the firm's strategy (Grant, 1991), the RBV can be considered one of the most used theories in the field of IS/IT. Perhaps because it is well suited to address what causes firms to perform well on a sustained basis, the RBV has been one of the strategic management field's biggest successes (Helfat et al., 2023).

Many researchers have used the RBV as a theoretical lens to show how organizations can outperform their competitors by creating value from better management of their resource bundles. By defining an RBV's version from a digital perspective, Giustiziero et al., (2021) propose that firms acting in this domain have resource bundles with large advantages of scale, which makes the competitors' cost of opportunity very high. Also, Aydiner et al. (2019) contributed confirming the existence of a strong relationship between IS/IT capabilities, decision-making and business-process performances. Kindermann et al. (2021), built upon the RBV and showed that although there is broad evidence that digital technologies are strategically important for value creation, current literature lacks concepts of an organization's strategic orientation related to digital innovation and transformation initiatives.

Wernerfelt (1984) introduced RBV by proposing the development of tools for positioning resources and defining strategies for companies, helping them to balance profitability and resources as well as managing these resources over time. In addition, Grant (1996) argues that durable CA emerges from unique combinations of resources.

Also fundamental to the RBV to strategy formulation is maximizing rents over time. “For this purpose, companies should investigate the relationship between resources and organizational capabilities” (Grant, 1996). The author claims that creating capabilities is not simply a matter of assembling a set of resources, capabilities creation involves complex patterns of coordination between people and between people and other resources. He also created a practical framework (Figure 1) referring to a general perspective in terms of organizational and corporate aspects (Grant, 1991, 2018). The concept behind the framework unfolds from a three-step process for strategy formulation also highlighting the four main elements of the theory.

We selected RBV as the theoretical lens of this research because, by using Grant's (2018) theoretical framework, we will identify, in the articles analyzed, the evidence of the management of IS/IT resource bundles, the creation of capabilities for the implementation of AI technologies and the consequent promotion of competitive advantages. We will also be able to explain why these advantages can be considered sustainable using two fundamental RBV concepts. First, through the complementation of IS/IT resource bundles, and second, through the fungibility and scale-free concept of AI technology (Krakowski et al., 2022).

RBV describes the theoretical mechanisms through which resources are associated with competitive advantage (Barney and Hesterly, 2015). Strategic resources are seen as a core concept of the RBV and are distinguished from resources in general, such as cash, because they have certain attributes: they are valuable, rare, inimitable and irreplaceable (Barney and Hesterly, 2015). However, the use of AI technologies that aid human decision-making challenges our understanding of the relationship between resource management and performance improvement (Helfat et al., 2023). This can be explained by a basic premise of RBV where managers seek to build advantageous positions for their company, gathering unique and valuable combinations of resource bundles (Rumelt, 1984).

Capabilities resemble organizational competencies and are rooted and embedded in business processes and routines (Bharadwaj, 2000). Routines and business processes enable the integration of individual actions to create organizational capabilities (Grant, 2018). They refer to a corporation's ability to exploit its resources and consist of business processes and routines that manage the interaction among resources to turn inputs into outputs (Wheelen and Hunger, 2018). A capability is functionally based, which means it is possible to identify many organizational capabilities within each functional area: purchasing capabilities, engineering capabilities, information systems capabilities, etc. (Grant, 2018).

They can also be defined as the knowledge set that distinguishes and provides a CA (Leonard-Barton, 1992). Grant (2018, 1991) views organizational capability as the outcome of the knowledge integration of a team that develops complex productive activities. They are dependent upon their ability to harness and integrate the knowledge of many individual specialists. Competitive advantages are the result of a complex coordination of resources and capabilities to provide the organization with the ability to obtain extraordinary returns (Wheelen and Hunger, 2018).

Strategy is about achieving success and not a detailed plan or program of instructions. It is a unifying theme that gives coherence and direction to the actions and decisions of an individual or an organization (Grant, 2018). A firm's strategy is defined as its theory about how to gain CAs. A good strategy generates such advantages (Barney and Hesterly, 2015). Strategic management, on the other hand, is a set of managerial decisions and actions that help determine the long-term performance of an organization. It includes external and internal environmental scanning, strategy formulation, strategy implementation, evaluation and control (Wheelen and Hunger, 2018).

According to Topi and Tucker (2014), when referring to organizational units and capabilities, the terms “information systems” and “information technology” are, in practice, used interchangeably, given the inability of the literature to consistently separate them. The authors define information technology as the technological infrastructure and information systems as the application of technology in an organizational context, however, they explain that this “would impose excessively restrictive constraints, given the current practice in the fields.” Therefore, for this research, we follow the authors' suggestion to resort to IS/IT terminology when we use both terms interchangeably.

To make the RBV more useful for IS/IT researchers, Wade and Hulland (2004) were able to help better identification of IS/IT resources as assets. Earlier in his research Wade (2001) highlighted that a resource can be an asset, like an information management system, a patent, a brand or a capability represented by an action, a process, a skill or an ability. When assets, resources and activities are configured in unique combinations, they become generative elements of capabilities (Amit and Schoemaker, 1993). In addition, the RBV was explored to clarify the use of IS/IT by researchers to understand the importance of resource complementarity and moderating factors of the firm's performance.

The complementarity of IS/IT resources in a dynamic environment appeared to play more of an indirect supporting role in achieving a Sustained Competitive Advantage (SCA) (Wade, 2001). He also states that “IS/IT resources appear largely to be IS/IT capabilities supporting other areas of the firm rather than creating a CA directly.” Resources are naturally complementary, which means they are less productive when detached from their original home (Grant, 2018). From the perspective of other areas, these IS/IT capabilities work as resources that support them to get SCAs. Pan et al. (2019) extended the previous work of Wade and Hulland (2004) on the complementary effects of IS/IT resources on operational performance. They translate the IS/IT resources into the domains of the logistics area, showing how they generally act positively on a company's performance, mainly the resources from the inside-out processes.

RBV theory was initially developed to study the internal resources of a company. Little focus was given to researching the specific resources of the IS/IT area. Its capabilities, at first glance, rarely contribute to a direct influence on SCA. However, according to Wade and Hulland (2004), IS/IT resources form part of a complex chain of assets and capabilities that may lead to SCA. They identified that IS/IT exerts its influence on the firm through complementary relationships with other firm resources.

RBV also explores how competitive advantages are created through complementation by developing unique resource combinations (Newbert, 2007). It can be done by integrating existing resources and new ones into uniquely complementary resource bundles (Argyres and Zenger, 2012). This unique complementarity may arise in two different ways. First, through experimentation or serendipity (to find out by chance) in which already unique assets, resources and activities are creatively combined into new complementary configurations. Second, through deliberate investment firms can actively reconfigure resources in unique ways (Argyres and Zenger, 2012).

In the same vein, complementary resources act by improving the incumbents' visibility to technological opportunities, and if the complementary assets they possess increase their ability to create value, they will have the incentive to enter a new market in the first place (Wu et al., 2014). Likewise, the firm that initially invested in the wrong technology, but which also invested in fungible complementary and scale-free resources, will be more likely to adapt to technological change, perhaps because it has the motivation to continue investing and overcome failure-induced biases (Eggers and Park, 2018).

Fungibility can be defined as the ability of a good or asset to be exchanged with other individual goods or assets of the same type then, a resource with high fungibility can be applied to many domains (Krakowski et al., 2022). A good example of fungible resource is money, where a $1 bill can be transformed into four-quarters or ten cents. Whereas, resources are treated as having a scale-free property when their value is not reduced due to the magnitude or scale of the operations to which they are applied (Levinthal and Wu, 2010). These two properties, working together with the concept of complementarity of resources, greatly facilitate the introduction of AI technologies in all areas of organizations, substantially increasing their scope of implementation. To explore and evaluate the use of RBV by IS researchers, Wade and Hulland (2004) developed a Typology of IS/IT Resources demonstrating that capabilities (as previously noted, a subset of the firm's resources) held by a firm can be sorted into three different types of processes (Table 1: Inside-out, Outside-In and Spanning). The Inside-out refers to capabilities that react to market requirements, threats and opportunities in an inside-out manner, focusing on internal attributes. On the other hand, Outside-in capabilities have an external orientation, emphasizing the anticipation of the market conditions. Acting in both directions, spanning capabilities are used to integrate the last two.

RPA is defined as “the software configuration to do the work previously done by people” (Lacity et al., 2015b). It has been seen as an umbrella term for tools that operate on the user interface of other computer systems in the way that a human would (van der Aalst et al., 2018; Collins et al., 2021). Also, RPA is a tool that executes “if, then, else” statements on structured data, usually using a combination of Graphical User Interface (GUI) interactions (Cewe et al., 2018). However, besides automating repetitive tasks, these need to be typically rule-based and well-structured (Syed et al., 2020).

Unlike traditional workflow technology, the systems remain unchanged (van der Aalst et al., 2018) because RPA operates through the user interface (Bosco et al., 2019; Lacity et al., 2016). In addition, RPA is seen as part of IS/IT that will gain importance in the modern business environment where information is processed on an unprecedented scale (Siderska, 2020). RPA is the “least expensive and easiest to implement of the AI technologies and typically brings a quick and high return on investment” (Davenport and Ronanki, 2018). Even though its level of cognition is the smallest compared to other types of AI.

We have observed two distinct approaches to the use of AI both in organizations and in academia. An incremental approach (Davenport, 2018; Davenport and Kirby, 2016; Davenport and Ronanki, 2018) moderately accelerates the evolution of AI utilization, while the radical approach (Iansiti and Lakhani, 2020a, b) expands like a chain reaction, completely changing the nature of the firm and its traditional business models. Based on these two approaches, we argue that for incumbent firms, the more appropriate way to initiate a cognitive journey in search of AI technology utilization is the incremental option. Perhaps because of the uncertainty of the radical perspective that can bring more risky situations to the incumbent business.

Starting with Davenport and Ronanki's (2018) approach, which analyses RPA within a more holistic view of AI, the authors compare other technologies with higher degrees of cognition. It is possible to observe the development of a path for the safe development of a digital AI strategy. They also highlight that “it is useful for companies to look at AI through the lens of business capabilities rather than technologies.” And complete by saying that there are three important “business needs,” which can be supported by different types of AI technologies, namely, process automation, cognitive insight, and cognitive engagement.

Process automation is the most common type of business need and refers to digital and physical tasks. RPA is more advanced than earlier business-process automation tools because the “robots” act like a human inputting and consuming information from multiple IS/IT systems, without touching their software. Cognitive Insight uses algorithms to detect patterns in vast volumes of data and interpret their meaning. It can be thought of as “cognitive perceptions” that are identified and offered to facilitate the analysis of a huge amount of data. Machine learning is a good example typically used to improve performance on jobs only machines can do. Finally, cognitive engagement is used in projects that engage employees and customers using natural language processing chatbots, intelligent agents and machine learning as well. In Table 2, we have summarized all the ideas developed by Davenport and Ronanki (2018) based on its incremental strategy.

It is understandable that theoretically connecting two different concepts requires an effort that involves searching for knowledge and critical thinking. Therefore, it would be valuable to put together the concepts and theories that were presented in the literature review section. To facilitate the readers' understanding of how the main literature that comprises this work was connected, namely, RBV and RPA, we created a workflow that details the theoretical lens (TL) used to amalgamate these ideas. This workflow of concepts represents the idea of the connection to be achieved. Its main function is to organize ideas and untangle the content analysis process, clarifying the concept behind the connection (Figure 2).

Our theoretical lens workflow was designed based on the backbone of RBV (Barney and Hesterly, 2015; Grant, 2018; Helfat et al., 2023; Peteraf, 1993; Wernerfelt, 1984), which is represented by the four main elements of the theory. Thus, it is possible to observe that its basic foundations were maintained, with resources being the forerunners, followed by capabilities that together promote sustainable competitive advantages, responsible for creating strategies capable of strengthening organizations in the face of their competition. In a way, it is quite noticeable that the greater interaction between the main concepts presented in the previous sections is concentrated in the IS/IT resources.

This interaction starts with the implementation of the simplest technologies, namely, RPA, and advances to the more complex ones after consolidating the previous implementation, which looks to minimize the uncertainties faced by the incumbent and mitigate the risks on the current business (Davenport and Ronanki, 2018). In this way, we argue that RPA is a fungible scale-free technology (Eggers and Park, 2018; Giustiziero et al., 2021; Krakowski et al., 2022; Nason and Wiklund, 2018) that can be easily applied in many domains and situations. Thus, RPA improves business processes increasing information systems efficiency and productivity, an inside-out infrastructure resource (Pan et al., 2019; Wade, 2001; Wade and Hulland, 2004).

The implementation of RPA ends up working as a spark that promotes the opportunity for organizations to acquire new knowledge about AI technologies. The development of new knowledge provides firms with the creation of new AI capabilities (Grant, 2018; Leonard-Barton, 1992). These AI capabilities will grow up as companies confidently accelerate toward the use of new and bolder AI technologies (Davenport and Ronanki, 2018). Given that the combinations of resource bundles are the generating elements of capabilities (Amit and Schoemaker, 1993). Thus, we postulate that organizations will be able to generate AI capabilities.

In addition, RPA integration with the infrastructure of IS/IT creates uniquely complementary resource bundles and highlights the possibility of managers to build advantageous positions by creating unique and heterogeneous valuable combinations of resources and capabilities (Amit and Schoemaker, 1993) and this heterogeneity will generate sustainable competitive advantages.

Finally, we have come up with an AI strategy formulation that can be originated by combining the firm's new unique resource bundles created from the automation of IS/IT infrastructure and AI capabilities generated through new learning development acquired from the RPA implementation. These new resources and capabilities will increase market opportunities and minimize internal firms' weaknesses (Grant, 2018; Wheelen and Hunger, 2018).

However, according to Grant (2018), two aspects draw attention to the role of resources and capabilities as the basis for a strategy. First, as business environments have become too unstable, internal resources and capabilities, not external markets, provide a more secure basis for strategy. Second, SCA rather than industry attractiveness emerges as the main source of profitability. Therefore, strategy formulation must be seen as a consequence of better coordination of resources and capabilities to create SCA.

An SLR is an explicit and reproducible project to identify, assess and interpret the existing body of recorded documents (Fink, 2014). According to Seuring and Müller (2008), it generally aims at two specific objectives. The first is to summarize existing research and identify patterns, themes and issues, while the second seeks to identify the conceptual content of the field and contribute to the development of theory.

The RQ that guides our study seeks to answer “how RPA can generate CAs to support a digital AI strategy through the management of IS/IT resources and the creation of AI capabilities.” As this RQ is very comprehensive, we decided to break it down into four specific RQs to facilitate the search and understanding of the answers (Collins et al., 2021). See below the list of four RQs, built from the main one.

To answer them, this study employed a hybrid approach, combining bibliometric and content analysis. This combination is designed to identify trends in the literature evidenced in the most frequently discussed topics and fields and any gaps that may exist in the literature (Gomes et al., 2019; Takey and Carvalho, 2016).

Our data sample was drawn from the two self-titled main databases available to scholars, Web of Science (WoS) and Scopus. They were chosen due to their advanced analytics and technology in one solution capable of saving process time and ensuring data accuracy. They also offer a set of metadata that can be easily collected, including data on abstracts, authors, institutions, number of citations, number of pages, number of cited references and impact factors of journals, essential for carrying out a consistent bibliometric analysis. In addition, they also have a structure for updating data that allows constant updates, besides keeping a wide range of collections ready for use (https://www.elsevier.com/solutions/scopus) and (https://clarivate.com/products/scientific-and-academic-research/).

We define a group of strings that would constrain the information to be reached in the databases. As we will rely on the RQs to determine the search strings, we have established an organization that groups the strings by each specific RQ. On the one hand, this action results in a higher-than-expected number of duplicated articles. On the other hand, this information allows us to visualize the string combinations that resulted in no articles. This brings us to one more relevant fact that attests to the originality of the topic under study. See the combination of the string in the RQs in Table 3.

Before start collecting data at Scopus and WoS, we refine the search by applying the same filter criteria in both databases: “Search within,” “Document type,” “Period” and “Language.” Search within represents the field where the search will be performed, therefore we selected “Title,” “Abstract” and “Keywords.” For Document type, we selected “Article” and “Review,” as these types of documents undergo a peer review process and offer the most complete set of metadata in the Scopus and WoS databases (Gomes et al., 2019). The period defined was “up to 2022” since our search was performed in November 2022. Finally, we select the filter language “English” because it represents the most used language among researchers. This selection criterion resulted in 548 documents, representing 335 from Scopus and 213 from WoS.

Continuing with the process, first, we started a refinement of the 548 articles' sample by applying a detailed process that can be seen in Figure 3. As an exclusion criterion, we started removing duplicated articles. This action resulted in 221 articles from both databases. Next, although some articles addressed themes explored by defined strings, there was a possibility that they did not address RPA as a main focus (Gomes et al., 2019; Takey and Carvalho, 2016). Thus, after analyzing the Title, Abstract and Keywords with attention to this criteria, 156 articles remained.

In addition, as we also decided to develop a content analysis, we applied one more article exclusion criterion, however, this does not mean that the articles removed in this last criterion will not be used in the bibliometric study. By selecting the theme, we understood that it would be more fruitful to use, in the content analysis, articles that applied a case study approach. This is because this approach is used when researchers intend to empirically investigate a contemporary phenomenon within its real-life context (Yin, 2015). In this way, we reviewed the methodology section of the 156 articles and selected those that applied a case study approach (47) from the others that used other approaches (109). To finalize our refinement process, we removed 7 more articles due to the unavailability of full text, leaving 40 articles for content analysis.

After collecting and refining our sample, we performed a bibliometric and content analysis (Braz et al., 2018; Gomes et al., 2019; Takey and Carvalho, 2016). Whereas bibliometrics analysis aid in understanding the publication patterns in the main databases, the content analysis focuses on the surveyed articles and helps to develop the conceptual framework (Takey and Carvalho, 2016).

Our bibliometric analysis was developed using a quantitative approach to provide an overview of the evolution and current state of the RPA literature. Thus, we prepared a set of visual elements that show important characteristics of the articles and journals that are in our expanded sample (156 articles). With this sample, we prepared the elements showing the number of publications per year, the number of publications per subject area, the number of publications per journal and a list of the most cited articles.

As one of the goals of bibliometric analysis is to count the number of times journals cite a particular article to yield a measure of an article's influence (Gomes et al., 2019), we created this list to provide not only this information, but we also sophisticate it by creating new information that, along with the number of citations, could bring more decision flexibility to academics who look for journals with relevance for publishing their articles. Besides the number of citations, we included the WoS Journal impact factor (JIF) and the article impact factor (AIF). AIF was first developed by Carvalho et al. (2013) and it means the number of times an article has been cited since its publication and its journal impact factor (JCR). AIF values were calculated as AIF = Citation * (JCR + 1).

In addition, as the article's main focus is content analysis and given the space limitation constraints, we decided to move the bibliometric analysis results to the appendix section. For more information on bibliometric analysis, see  Appendix 3.

Unlike the bibliometric analysis, we used a qualitative approach in our content analysis for the sample of 40 articles. Although we adopted an iterative process for our content analysis, we can point to three distinct steps (Gomes et al., 2019). In the first step, we defined a process text analysis following the steps of the theoretical lens (TL). Based on the structure of the main elements of the RBV, we create a coding frame (CF) to organize and guide the search for the answers for the specific RQs in our sample (Table 4). We completed the CF defining the coding scheme for each respective RQs namely: Resource, Capability, CA and Strategy Formulation. For each specific RQ code, we define what should be found in the sample text clearly and descriptively.

In the second step, following the concepts of the TL, we organize the answers in a systematic approach, using a Microsoft Excel spreadsheet, containing: the articles' titles, authors' names, publication years, references representing the specific RQs answers, and the page it was found. These worksheets were essential to keep the information organized and safe for future investigations. In the third and final step, based on the results, we built a proposal for a new theoretic framework for AI digital strategy formulation.

This section presents the content analysis results of the sample of 40 articles by searching for the answers to each RQ. In that vein, we based on the construction of our TL (Figure 2) in association with the proposed coding frame (Table 4). On the one hand, the TL will trace the path of how the connection between the RBV and RPA literature will be made, on the other hand, the coding scheme will drive the search for the responses of the RQs that matched the evidence obtained within our sample.

RQ1 looks for identifying how RPA improves IS/IT resources management after automating a digitized business process or integrating two systems supporting different areas. TL shows that automating the business process or integrating systems is how RPA improves IS/IT infrastructure resources. Therefore, we set a coding scheme as “Resources” and search citations on the article's sample for all business processes automated or systems integrations examples.

We brought some examples in Table 5 to help understand how can improvements achieved by RPA after its implementation helped in the management of IS/IT infrastructure resources through the automation of business processes, as were the cases of (Wojciechowska-Filipek, 2019; Hartley and Sawaya, 2019; Kokina and Blanchette, 2019; Flechsig et al., 2022; Soeny et al., 2021). Examples of systems integration were also identified, such as (Thainimit et al., 2022; Carden et al., 2019; Šimek and Šperka, 2019). In addition, the fungibility of the RPA technology also appeared through the many different domains shown in the publications per subject area ( Appendix 3 - Figure A1).

During the analysis of the data coded as “Resources,” we observed important aspects that help us to build an understanding of how the implementation of RPA to automate business processes or integrate systems, involves several areas of the company. This involvement creates CAs that are often only observed in these areas.

For example, Wojciechowska-Filipek (2019) shows the automation of a particular process in the financial sector, which aims to prepare confidential information to be delivered to authorities and institutions authorized by the government. This process has the involvement of various areas of the organization such as operations, legal, public relations and customer management, among others. It is where the fungible complementary aspect of RPA will promote unique sustainable competitive advantages.

We highlight another example in the medical field that explores RPA for eye screening (Thainimit et al., 2022). The research develops a glaucoma exam based on a mobile application integrating several systems with RPA and Machine Learning. The RPA processes activities in several areas such as collecting patient history data, payments, requesting clinical visits, acquisition and transferring images and medical data among the central ERP, intermediary systems, and the developed mobile application.

RQ2 aims to understand how can AI capabilities be built through RPA implementation. According to our TL, the incremental strategy (Davenport and Ronanki, 2018) to implement AI technologies can solve most of the business needs a firm can face. The simplest and lowest-risk option is to start through the RPA implementation. However, to do that, firms must develop new knowledge about these AI technologies (Grant, 1996; Leonard-Barton, 1992) thus generating new AI capabilities. This will also drive the complementation of AI capabilities and IS/IT resources (Argyres and Zenger, 2012; Eggers and Park, 2018; Krakowski et al., 2022). Therefore, we set a coding scheme to track back in our sample, citations of references that demonstrate the search of firms for new knowledge and learning development about new AI technologies and complementation of resources and code them as “Capabilities.” The results of this coding scheme are shown in Table 6.

After analyzing and discussing the results, we highlight relevant aspects that point to the path of continuous growth of knowledge and the complementation of resources by companies. Carden et al., (2019) showed in their single case study, the example of a large technology company that sought to increase its efficiency and productivity by automating its business processes with RPA. The strategy was to seek, through innovation, to develop the necessary capabilities for the future and chose to start this project via RPA. In this way, it initiates a continuous development of knowledge, creating a path for the automation of intelligence related to the cloud, process robots, visualization, cognitive computing, advanced analytics and Blockchain.

In Table 6, it is possible to see examples of searching for knowledge creation (Cooper et al., 2019; Lacity et al., 2021; Carden et al., 2019; Lacity and Willcocks, 2016). We also identified examples of complementation of resources (Sobczak, 2019; Sobczak and Ziora, 2021).

Nonetheless, we emphasize that as academic studies on capabilities are vast and permeate different thematic areas, so is its definition. We have a multitude of definitions that make it somewhat difficult to understand when we do not consider other factors such as context, objective, thematic area, etc. Therefore, after coding and collecting data and defining capabilities in the literature review, we contextualize capabilities based on our findings.

Considering the context, we can look at capabilities in two different ways. There is a group of capabilities that are intrinsic to the technology. For example, RPA has great flexibility and can be duplicated almost instantly during a sudden increase in demand and return to the initial condition after the surge. The other group is underlying capabilities related to the organization that is created to react to a given situation. In our case, the implementation of AI technology to meet a business need generates a movement of coordination and management of IS/IT infrastructure resources and other resources from related areas that create new capabilities through scale-free fungibility and complementarity, in this case, AI capabilities. In our research, we are focused on this last group of capabilities.

Cooper et al., (2019), who studied an RPA implementation case in public accounting, observed that after automating work tasks, a different set of skills were demanded of accountants, although the use of third-party software programs allows accountants to create bots without Extensive coding knowledge and experience. Another example of knowledge creation in automation by RPA is Kokina et al., (2021). In this article, the authors explore the roles, skills and competencies accountants will need to develop for automation. They also point out that some required skills will be coding (programming), management and data structure.

RQ3 aims to discover how automating a business process using RPA can generate SCA. Grounded on TL step 4, we argue that while IS/IT infrastructure resources cannot generate sustained competitive advantages automating business processes with RPA makes it possible through scale-free fungible complementary resources. This is possible, as most business processes involve activities from different areas and their results and benefits appear mainly there, despite automation being applied to an IS/IT infrastructure resource. Therefore, we elaborate a coding scheme to look for citations on the samples identifying results found by using RPA which improves competitiveness, efficiency or performance of the firm in other areas. Then we codify these citations as “CAs.” In Table 7, we highlight several citations that were coded as “CAs” and all confirm the argument of our TL.

In a case study, Šimek and Šperka (2019) researched the implementation of RPA through a pilot project to understand its phases and required activities. In addition to several contributions to the study of RPA, the authors highlighted that shortening the processing time by transferring the workload to robots in the HR onboarding process, opened additional space for recruiters to take care of applicants and serve them more quickly, which is a CA in the labor market. Despite the IS/IT infrastructure has received direct improvements of automation, the results of the project were appropriated by other areas of the firm. In this specific case, as we are discussing the implementation of RPA in a business process services market firm, it improved the HR department, which in turn also optimize the sales and customer relationship area delivery.

Another example of SCA creation can be seen by Helm et al. (2022) which follows the same rationale as the former. The study aims to research cognitive automation, an RPA automation attended by AI, in three legacy firms. It proposes to apply cognitive automation to extract unstructured data from documents and convert it to structured data to be used in big data analytics. The results of the automation of three different business processes were, on average, 80% in error reduction and 20 min per document in processing time. This automation reduced the time spent with repetitive tasks and enabled workers to return to value-creating work. This was important information for the management to create added value in the final areas, said the authors.

RQ4 aims to understand to which extent the development of an AI digital strategy can be promoted by the automation of business processes with RPA. Paved in the TL, strategy formulation represents the final result of the process postulated in the RBV. While defining key resources and capabilities and having established firm strengths and weaknesses, we can devise strategies where a firm can maximize its strengths and minimize its vulnerabilities. Therefore, we define a coding scheme to find in the sample, citations of references that show the creation or change in the firm's strategy given the implementation of RPA. Then we codify it as “strategy formulation” (Table 8).

By discussing the above results, we can underline research from Hartley and Sawaya (2019) that studied digital transformation in the supply chain business process by investigating digital technologies like RPA, ML and Blockchain. The study concluded that those technologies can provide automation and data, enabling supply management professionals to stop doing transactional activities to initiate thinking strategically about increasing business value to generate SCA.

Another example of strategy formulation can be seen by Mishra et al. (2019). The authors investigated the business process management industry to understand technology deployment challenges that impact automation through RPA. With this aim, they argue that the automation strategy is more of a business strategy to solve an organization's critical business problem in the digital era rather than a tool that just makes automation possible. They also affirm the focus on RPA to enlarge its project scope to automate end-to-end processes using new AI technologies.

RPA literature pointed out that about 37% of the executives interested in implementing AI programs, do not understand these technologies or how they work (Davenport and Ronanki, 2018). The same survey also shows that a 56%of the managers were screened out because they were not knowledgeable about AI technologies or their companies and were not active with them. Based on that and the results discussed, we develop a conceptual framework for AI strategy formulation (Figure 4). The main idea behind the creation of this conceptual framework came from the analysis of the literature, the concepts behind our theoretical lens and an exhaustive discussion of this paper's results.

The first modification applied to the original framework for strategy formulation was the change from a workflow to a cycling shape. Despite the original framework also working in a cycling manner, it did not look like physically a process that needs to do it. Thus, by changing to a cyclin shape, the new framework format stimulates the user to work constantly.

We also added a new initial step (1-) that alerts the user to understand AI technologies more deeply. The following change was the inclusion of the IS/IT resources bundles. The goal of this modification was to incentivize the user to assess IS/IT resources, create a list of resources available and understand their usefulness. We also provide a list of Business needs characterized by the incumbent incremental strategy approach for AI technologies. The user can look at the table and choose the best technology option for their needs following the incremental strategy. Then, by choosing the required knowledge to be acquired, users can create new AI capabilities through knowledge development. The third and fourth steps are an adaptation of the original steps of RBV to the context of digital transformation. After presenting the modifications, we unveil its functionality.

The process starts with an alert, to firms interested in developing an AI strategy, to seek an understanding of the main AI technologies and the concepts behind them. Main vendors and their service quality indicators, examples of applications, strengths, weaknesses and limitations. Also, it is necessary to identify the human resources available and the need for training.

Next step, we must elaborate an inventory of all the IS/IT resources required for implementing AI technology. At the same time, it should be also necessary to inventory the AI capabilities according to the business needs and the respective AI technology. The IS/IT typology (Pan et al., 2019; Wade and Hulland, 2004) table will serve as an example for practitioners. However, users should not limit themselves to them, making an internal assessment in search of other resources not highlighted in this framework. If the user finds different IS/IT resources, it is a positive sign that the company will be more likely to create sustainable competitive advantages that are difficult for competitors to copy or imitate. The same process should be done with AI and other capabilities, but the user must look for the business needs first to identify the alternative of AI technology that fits their specific needs and then develop a corresponding AI capability.

In the third step, the firm must evaluate IS/IT resources and AI capabilities in terms of the strengths and weaknesses of the competitors and their strategic importance. In the fourth step, they should analyze the best strategy to maximize the strengths and minimize the weaknesses of the firm.

We achieved the main purpose of this research by clearly explaining the strategic side of AI technologies, more precisely RPA. Therefore, we theoretically grounded our study on the RBV theory, which even after 4 decades of its creation is still the main theoretical basis for research on business strategic management. After preparing these bases, we contributed theoretically and practically to the literature by building the theoretical lens that amalgamated all the other concepts used and the conceptual framework for formulating AI strategies. We also successfully evaluated the role of RPA technology at each stage of the original RBV workflow, uncovering how technology can support incumbent companies. In addition, we carried out our research following the academic rigor required for a study with this purpose.

As a contribution to the literature, the theoretical lens workflow (Figure 2) clearly shows how AI technologies are capable of generating sustainable competitive advantages. For this, we update and shed light on the concept of complementarity, fundamental for managing and creating bundles of unique complementary resources. We also understand that RPA, as a technology under the AI's umbrella, can be considered a fungible and non-scalable technology resource. These types of resources adapt perfectly to the characteristics of AI technologies and, when acting in conjunction, transform resource bundles into unique and inimitable ones, creating means for the development of strategies to achieve sustainable advantages.

In addition, from the theoretical lens emerged a central contribution to our research. By following the foundational RBV workflow with its four main elements, we model a new innovative conceptual framework for AI strategy formulation (Figure 4). With this framework, we propose a path for incumbent organizations aiming to embark on a cognitive journey for AI and look for a safe way to do so. Besides using the RBV foundational workflow, we also incorporate all the concepts discussed in this manuscript. We also show, for all these reasons, that RPA becomes the safest and most advantageous option to start an AI digital journey, adopting a safe incremental approach to incumbent business protection.

Moreover, we observed that the growth rate of publications is constantly falling from 2019 to 2022 ( Appendix 3). Although it is normal that the number of publications on a subject after considerable growth stabilizes and starts to decline, the RPA is still a very new subject that improves its usability constantly and, therefore, should be further studied, mainly if we consider the innovations that are being recently applied alongside this technology, cognitive automation for instance. By showing the strategic side of RPA by looking at RPA from the perspective of RBV, we expect to encourage the academics interested in this technology, contributing to increasing the interest in the publication and making them gain traction and rise again.

Despite being theoretical, the conceptual framework for AI strategy formulation unfolds in another relevant practical contribution. From a managerial perspective, incumbent organizations interested in adopting an incremental approach may use this framework to create a new strategy or help change the current one intending to deploy AI technologies. Our objective was to propose a basic path for incumbent organizations that intend to embark on this journey and are looking for a safe way to do so.

This academic work has important limitations that need to be pointed out. This is a purely theoretical research work and therefore requires further empirical verification, particularly concerning the conceptual framework for strategy formulation. Although our results were obtained by consulting other academic research, we minimized this limitation by using data collected in articles that approached RPA with due rigor and that used a case study method, the most appropriate for reporting empirical research. In addition, we searched for articles for the theoretical foundation in journals of great prominence in the classification rankings of global academic production (ABS, JCR, CiteScore, etc.).

Although we address the subject of artificial intelligence in this work, another limitation is the focus on RPA technology. RPA is a technology that is under the umbrella of artificial intelligence. This includes numerous other tools such as machine and deep learning, neural networks and Chatbots all with different attributes.

As recommendations for future research, we suggest an empirical study on RPA implementation, to test the conceptual framework and the steps of the developed theoretical lens. We also propose to validate the existence of the IS/IT resources influences and relationships discussed in this study. Another further suggestion can be to investigate how RPA operations management impact could contribute to overcoming the organizational challenges of the current context of digital transformation. Furthermore, how could RPA technology providers assume the role of complementors in the AI ecosystems and check how RPA technology could contribute to sustainable development by leveraging the Circular Economy approach.

RPA:

Robotic Process Automation

AI:

Artificial Intelligence

IS/IT:

Information Systems and Information Technology

RBV:

Resource-Based View

CA:

Competitive Advantage

CF:

Coding Frame

SCA:

Sustainable Competitive Advantage

WoS:

Web of Science

JCR:

Journal of Citation Report

JIF:

Journal Impact Factor

AIF:

Article Impact Factor

TL:

Theoretical Lens

RQ:

Research Question

SA:

Subject Area

ML:

Machine Learning

HR:

Human Resources

EIS:

Evaluation Index Systems

FNN:

Fuzzy Neural Network

We started our bibliometric analysis using the expanded sample of 156 articles that were collected based on a study of keywords and grounded on the research questions (Collins et al., 2021). In this section, we show results representing the state of research in RPA in the context of IS/IT from the perspective of RBV.

Analyzing the sample, we observe publications in a varied number of subject areas (SA) due to the pervasiveness of AI technologies (Teece, 2018). Because it is pervasive, AI technologies can have applications in many areas. Although the SA with the greatest number of publications is Business, Management and Accounting with a total of 50 publications, our expanded sample has 15 different SAs in total, as can be seen in Figure A1.

We also emphasize that the sample contains SAs that belongs to all four core SAs, namely: Health sciences, Life sciences, Physical sciences and Social sciences, according to Scopus classification. Therefore, besides it is explained by the pervasiveness of AI technologies, this variety of SAs can also be explained by the complementarity of these technologies. Resorting to Teece (1986), the complementarity of IS/IT resources can be explained as when one resource can influence another and how the relationship between them affects a firm's performance.

Publications on RPA have increased considerably over the years. An investigation of this evolution also shows some important points to be highlighted. Despite many articles defending hype on RPA (Syed et al., 2020), the percentage of publications growth pointed out in Figure A2 showed an important decrease since 2019. It has steadily fallen at a very important rate, pointing to percentages such as 233%, 55%, 45% and 7%.

Working in the direction of showing articles and journals with the greatest impact and influence in the topics explored in this research, we prepared another two visual elements to be seen together, the number of publications per journal and the table of articles and journals' impact. As we said, we intend to support academics interested in having their research published in journals with high impact at the academy. We performed a mapping of the publications based on the 156 articles of the expanded sample. These 156 articles were all published in 116 different journals, however, 95 journals published just one manuscript in this period.

We prepared a graphic with the number of publications per journal up to 2022 (Figure A3). In this graphic, we see 21 journals with two or more articles published. Although it is very important to look at the first seven journals with more publications as a good alternative for publishing, academics should also pay attention to the impact of these journals and articles as well. For this reason, we also prepared Table A2 presenting the most cited articles and their journal's impact factor and articles' impact factor. Both pieces of information can be if read together, a good source of information to help academics interested in publishing about RPA.

By deepening our analysis of Figure A3 and Table A2 together, it is possible to clarify important situations that need academic attention. Journal Computers in Industry, for instance, placed in the first position of the most cited articles with the highest impact factor (Table A2) and has also a good position (3rd) on the number of publications (Figure A3). MIS Quarterly Executive follows the rationale, placed in second and seventh positions, respectively. However, the Journal of Emerging Technologies in Accounting and Sustainability, second and fourth in several publications (Figure A3), has no position within the 20 most cited articles list (Table A2). Therefore, it is necessary to observe both positions to have a better understanding of the publication's possibilities.