Artificial intelligence adoption in a professional service industry: A multiple case study

This study explores the factors influencing AI adoption in professional service firms. Grounded in the Technological-Organizational-Environmental (TOE) framework, we employed a qualitative, multiple case study approach, investigating three auditing firms of varying sizes through interviews and secondary document reviews. Our findings reveal six factors influencing AI adoption, including technology affordances and constraints, the firm's innovation management approaches and AI readiness, the competition environment, and the regulatory environment. Noteworthily, these factors vary significantly among the three firms. Larger firms, often operating in an environment with high AI penetration, primarily perceive the operating affordance of AI rather than marketing affordance. This means their AI adoption encompasses greater scale and depth than smaller firms. However, this expansive adoption exposes them to a widening gap in regulatory frameworks, hindering AI adoption. Moreover, smaller firms are characterized by weaker AI readiness, positioning them disadvantageously to mitigate the constraints imposed by AI. This study contributes to existing literature by offering a more holistic perspective on AI adoption in professional services.


Introduction
The advent of Artificial Intelligence (AI) technology has brought revolutions to numerous industries.Among these industries, professional service stands distinct, primarily characterized by their high degree of customization, deep client engagement, knowledge intensity, and a specialized workforce (Maister, 2012;von Nordenflycht, 2010).Historically, these unique traits rendered professional service industries relatively insulated from technological upheavals (Autor et al., 2003).Nowadays, this landscape is shifting with the rising assimilation of AI into the knowledge domain and service delivery (Sampson, 2021).AI is increasingly adopted by professional service firms across different domains, such as accounting (Damerji and Salimi, 2021), legal (Remus and Levy, 2017;Susskind, 2017), and healthcare practices (Sun and Medaglia, 2019).However, AI adoption is an intricate journey.A recent study reveals that over half of AI projects fall short of delivering their promises to adopting firms (Vial et al., 2022).The failures are largely attributed to AI's dynamic and disruptive characteristics, which create complexities in its amalgamation with existing processes, structures, and managerial philosophies (Johnson et al., 2022;Sharma et al., 2022).Therefore, the need to better understand the adoption of AI in professional service firms can never be overstated.
On the one hand, the literature has extensively explored AI adoption across various domains, including public sectors (Maragno et al., 2023), manufacturing industries (Chatterjee et al., 2022), SMEs (Polisetty et al., 2023), and AgriTech firms (Issa et al., 2022).However, the findings from these studies cannot be directly applied to professional service firms due to their distinct characteristics, such as knowledge intensity and regulatory environment.These characteristics, when coupled with the evolving nature of AI, necessitate further investigation.
On the other hand, there have been some studies examining the adoption and use of AI within professional services, pointing to vital factors like the integration of a professional knowledge base into AI systems (Goto, 2023;Spring et al., 2022).However, an underlying assumption in these studies is that the determinants of AI adoption remain consistent across firms of different sizes.Counterarguments to this assumption can be drawn from other technological contexts.For instance, the research by Karunagaran et al. (2019) on cloud adoption in Germany underscored the divergent experiences of large firms and SMEs.While larger firms grapple with challenges stemming from established systems, interoperability conflicts, and inherent organizational complexities, SMEs predominantly embrace cloud computing for its economic advantages and have reported more pronounced benefits (p.871).Nevertheless, the extant literature exploring AI adoption is focused on either SMEs (Chatterjee et al., 2022;Lu et al., 2022;Sreenivasan and Suresh, 2023) or large firms (Bedué and Fritzsche, 2022;Chen et al., 2021), without a comprehensive understanding of factors influence AI adoption in different settings.Considering these gaps in the literature, we decided to adopt a multiple case study approach.To improve the comprehensiveness of our findings, as highlighted in Karunagaran et al. (2019), we decided to focus on firm size as the differentiating factor.Hence, we propose an exploratory study of factors influencing AI adoption in professional service settings: What are the factors influencing the adoption of AI in professional service firms?
The study is guided and informed by the technology-organizationenvironment (TOE) Framework (Tornatzky et al., 1990), which offers a holistic perspective to understand the AI adoption factors.To investigate the research question, we selected auditing, a professional service industry undergoing significant changes brought about by AI technology (Issa et al., 2016).We considered the Australian auditing industry as a good example for studying the adoption of AI in professional service industries because it is characterized by a high level of professional knowledge, customization, and client interaction.Technology disruption to the Australian auditing industry has sparked many debates, for example, between Kend and Nguyen (2020) and Seethamraju and Hecimovic (2020), who argued whether auditing professions have more acceptance or resistance towards AI.In addition, while other professional services are provided in very different ways in different contexts, like legal practice (Tetley, 1999), auditing services around the world are relatively more convergent (Tokar, 2005).Therefore, the findings in this context can be better generalized and interpreted by people in other contexts and countries.
This paper investigates AI adoption across three auditing firms of varying sizes.To our knowledge, this study is the first to holistically examine the contextual factors relating to AI adoption across professional firms of different sizes.The findings show six pivotal factors that influence AI adoption at the firm level, and the variation of the impact of these factors according to the size of the firm.These findings may require us to reconsider commonly held beliefs that advocate for a onesize-fits-all approach to AI adoption.Our findings bridge a gap in the existing literature, which has not thoroughly addressed how AI adoption differs between small and large entities.Moreover, we challenged the conventional paradigms of the TOE framework, highlighting its weakness in overlooking the strategic motivations that compel firms towards technology adoption.By integrating technology affordances and constraints into the TOE narrative, we introduce an enriched framework.This adapted TOE framework provides insights into technology adoption, linking the functional capabilities of the technology to the strategic visions of the adopting firms.
From a practical standpoint, our research offers insights for professional service firms keen on harnessing the capabilities of AI.Drawing from real-world cases, we show how AI adoption can either improve or constrain a firm's service quality, client interactions, and overall business reputation.Furthermore, we emphasize the distinct environmental considerations for firms of different sizes: while larger firms might need to prioritize regulatory compliance, smaller entities can capitalize on AI's potential to swiftly carve out competitive edges.Additionally, we spotlight the intrinsic organizational dynamics that can either bolster or impede AI adoption, underscoring the significance of linking agents and AI readiness.
The paper is structured as follows.In the next section, we provide an overview of AI usage in the professional industries, as well as our theoretical foundation.Next, we present the study's methodological approach to collect and analyze the data, followed by the findings.The discussion follows, critically analyzing the findings against the existing literature to identify and elaborate on the study's contributions and implications.

AI and professional service industries
The most common types of AI technology discussed by researchers in the IS discipline are machine learning, machine vision, natural language processors, expert systems, and robotics (Collins et al., 2021).Instead of focusing on one of these categories specifically, this study takes a broad lens in considering the contemporary technologies in the AI spectrum.The reason for undertaking this perspective is to avoid a bias in research findings resulting from focusing on a particular type of AI (Sutton et al., 2016).This paper follows the definition by Russell et al. (2010, p 31), referring to AI as the technology that "enables the machine to exhibit human intelligence, including the ability to perceive, reason, learn, and interact."This definition is appropriate as it defines "what AI's capabilities are rather than strictly defining what it is (Collins et al., 2021, p 10)," allowing the examination of a wide range of AI technologies used in auditing.Therefore, the term AI is used in this paper as an umbrella term referring to a series of supervised and unsupervised machine learning technologies, such as neural networks and associative learning, as well as knowledge-based systems, such as decision-support systems and expert systems.
Corresponding to the increasing prevalence of AI adoption in various industries, the Information Systems (IS) and Technology Management research disciplines have witnessed a resurgence in interest in AI since 2017 (Collins et al., 2021).This renewed interest spans a vast spectrum of subjects.Extant research has delved into the potential of AI in value creation (Enholm et al., 2022;Vial et al., 2022), explored the dynamics of human-AI collaboration (Fügener et al., 2022;Melville et al., 2023), and debated on ethical considerations and trust issues surrounding AI (Ashok et al., 2022;Burton et al., 2020;Mahmud et al., 2022).Specifically, AI adoption has been studied across a myriad of professional service domains, including healthcare and medicine (Gupta et al., 2021;Johnson et al., 2021), legal service (Armour and Sako, 2020;Xu and Wang, 2021), banking and financial service (Bhatia et al., 2021;Rahman et al., 2021), and actuary (Richman, 2021).Through these studies, we discern the transformative potential of AI and gain insights into the myriad factors, such as data privacy concerns, AI readiness, perceived usefulness, that either propel or hinder AI adoption in the professional services context (Rahman et al., 2021).

The TOE framework
Various theories and frameworks from different disciplines have been customized to study the adoption of new technologies.Renowned theories, such as the diffusion of innovation theory (Rogers, 1995), institutional theory (Teo et al., 2003), resource-based view (Caldeira and Ward, 2003), and technology-organization-environment (TOE) framework (Tornatzky et al., 1990), provide theoretical foundations to explain the technology adoption at the firm level.Among these theories, we determined that the TOE framework offers a holistic and flexible perspective on technology adoption, which aligns with the exploratory objectives of the study.
Pioneered by Tornatzky et al. (1990), the TOE framework categorizes the myriad of factors influencing technology adoption into technological, organizational, and environmental constructs.However, the TOE framework is often characterized as a generic framework (Zhu and Kraemer, 2005), where numerous factors can be situated and analyzed within it (Baker, 2012).The variation of factors explored and examined in recent TOE studies further justifies its malleability and flexibility (Maragno et al., 2023;Neumann et al., 2022).Therefore, this framework is ideally suited for structured exploration without being tethered to predetermined constructs.In this research context, which explores the factors that influence AI adoption by professional service firms, the TOE framework is relevant because it can encapsulate the technological intricacies of AI, organizational dynamics, and the broader environmental context where these firms operate.Given AI's evolving and novel nature, we have foreseen uncovering previously uncharted factors influencing AI's adoption.The inherent malleability of the TOE framework can integrate these emergent insights.In addition, the intrinsic scalability of the TOE framework, adaptable from small businesses to large corporations, presented a compelling case for its use (Mahroof, 2019;Polisetty et al., 2023).

The TOE factors in AI adoption
We synthesized recent literature using the TOE framework to investigate the adoption of AI in various contexts.Appendix 1 shows the methodology of these studies, the type of AI and the focal context, the TOE factors investigated, and the main findings.The literature shows several discernible trends in the study of AI adoption.Compatibility (Polisetty et al., 2023;Sharma et al., 2023), complexity (Kinkel et al., 2022;Pan et al., 2022), and relative advantages of AI (Nam et al., 2021;Prasad Agrawal, 2023) are frequently emphasized under the technology dimension, highlighting the generic challenges and benefits to adoption, as well as the need to align with existing systems.Some technology factors specific to AI adoption are also mentioned, such as data quality requirements (Merhi and Harfouche, 2023;Mi et al., 2023), interoperable datasets and continuous learning processes (Maragno et al., 2023).These factors reveal AI's distinct technological characteristics and are more closely aligned with the argument for a rethinking of traditional theorizations of technology adoption and readiness within the AI context (Issa et al., 2022).
Organizational factors underscore the pivotal role of leadership, particularly top management support (Merhi and Harfouche, 2023;Sharma et al., 2023), and emphasize the interplay between operational capacities and strategic alignment in facilitating AI adoption (Mahroof, 2019;Nayal et al., 2022).However, the influence of firm size on AI adoption appears to be a subject of debate.For instance, a positive correlation between firm size and AI usage intensity is presented among international manufacturing companies (Kinkel et al., 2022) and the top 300 Indian companies (Prasad Agrawal, 2023).In contrast, such a correlation does not hold for the HR sectors in Chinese companies (Pan et al., 2022), pointing to the contradictory understanding of firm size's influence on AI adoption.It is worth noting that these studies, being quantitative, do not delve into the reasons behind the varying impact of firm size on AI uptake across different contexts.
While some studies suggest environmental factors are not the dominant factor in AI adoption (Merhi and Harfouche, 2023;Neumann et al., 2022), their influence is still notable.The literature recognizes elements like government support, competitive intensity, and environmental dynamics as pivotal.Governments, through policy settings, legal protection, and financial support, can facilitate AI adoption (Pan et al., 2022;Prasad Agrawal, 2023).Moreover, intensifying competition initiates AI adoption, with organizations recognizing AI as essential to secure or enhance their competitive advantage, particularly when their rivals have already implemented AI solutions (Mi et al., 2023;Nam et al., 2021).Beyond competition, with the challenges posed by rapid environmental changes, AI facilitates the agility necessary for organizations to adapt, underscoring its importance in navigating an everevolving business landscape (Prasad Agrawal, 2023).

Research methodology
The study used the multiple case study method (Yin, 2009).The data was collected using semi-structured interviews with 15 informants, complemented by supporting documents, such as firms' transparency reports and information from the vendors of AI applications.The data analysis was guided by Gioia's methods of analyzing qualitative data (Gioia et al., 2012).
To understand the adoption of AI in professional service firms, we used a case study approach with an aim to explore the relationship between the adoption of AI and its environmental context (Yin, 2009).The selection of the case study methodology was appropriate for capturing the emergent and evolving nature of AI adoption in professional services.The landscape of AI adoption within professional service industries is rapidly evolving yet remains nascent in academic literature (Spring et al., 2022).This emergent nature necessitates an exploratory approach.Therefore, the depth offered by the case study method allowed for an exploration of the complexities and multifaceted nature of AI adoption.The case study approach has been used in similar research when exploring nascent technologies, as seen in the domains of cloud computing (Karunagaran et al., 2019), fintech (Muthukannan et al., 2020) and blockchain (Chong et al., 2019), corroborating its suitability in our context.
The exploratory research design can also justify the selection of a multiple case study methodology.While a single case study can delve deeply into one specific context, the multiple case study method broadens the scope, capturing a range of perspectives and thus offering richer and more generalizable insights (Yin, 2009).The multiple case methodology was chosen to understand the adoption of AI comprehensively.To ensure a comprehensive view, we selected firms of varying sizes, recognizing that the adoption of the same technology could vary significantly among different-sized firms.Previous research has considered firm size as the differentiating factor in multiple case studies exploring technology adoption (Karunagaran et al., 2019;Pan and Pan, 2019).Adopting multiple case study design allows for both depth (within each case) and breadth (across cases) of our discussion.

Selection of auditing industry
Auditing is a professional service that provides assurance about the reliability of the information contained in the financial reports prepared by management in accordance with Generally Accepted Accounting Principles (Knechel and Salterio, 2016).Information technology has had a significant impact on the audit profession regarding the documentation, decision-making process, audit quality and productivity (Janvrin et al., 2008).Recent years also witnessed a growing focus on the adoption of emerging technologies in the auditing field, such as big data (Vasarhelyi et al., 2015), cloud computing (Sookhak et al., 2015), blockchain (Schmitz and Leoni, 2019), and AI (Munoko et al., 2020).The driving force to adoption is the challenges faced by the auditing profession in terms of vast volumes of unstructured data in auditing engagements (Kokina and Davenport, 2017).Appendix 2 synthesizes the discussions on AI technology automating, assisting and augmenting auditing tasks and processes.Although not exhaustive, the table indicates the research landscape and an overview of AI used in auditing.

Case selection
In order to ensure the comprehensiveness of the findings, case selection was based on a replication design that cases should have contrasting conditions (Yin, 2009), precisely, differences in their size.Thus, the key criterion for case selection is that cases should be from each of the large, medium, and small categories.As discussed earlier, we selected cases from the Australian auditing industry, which has the representational characteristics of professional service industries.In addition, because we were interested in the adoption status rather than the adoption decision, the candidate cases were those that had already adopted AI-enabled tools in their practice.
We began screening candidate cases from the Australian Financial Review (AFR), 1 which provided rich information about Australian auditing firms, such as the number of partners and annual revenue.The 1 An Australian business-focused, compact daily newspaper covering the current business and economic affairs of Australia and the world (https://en.wikipedia.org/wiki/Australian_Financial_Review-https://www.afr.com/).initial filter of cases was based on publicly available information regarding the adoption of AI, such as the firm's transparency reports, media releases, websites, and news.After obtaining ethical approval from the university, we contacted the candidate firms to confirm their adoption status.Subsequently, an invitation letter was sent to the initial contact within each of the candidate firms.Of the nine firms approached, six responded to a request to participate, and three agreed to participate.The number of cases selected for this study was similar to the convention for case study research in IS adoption (Poba-Nzaou and Raymond, 2011;Widuri et al., 2016).The three firms were one Big Four international firm (pseudonym: BIG), one second-tier firm (pseudonym: MID), and one small local firm (pseudonym: SML).Table 1 presents a comparison of the three firms, detailing their descriptions, the status of AI adoption at the time of the study, the type of AI technology they adopted, the range of auditing tasks AI can cover, their development strategies, and how they have integrated AI into their information systems.Among the three case firms, BIG has a comprehensive AI adoption, featuring advanced technologies like machine learning and natural language processing.Its AI applications are integrated into a grand auditing digital platform, covering a broad range of auditing tasks.MID has adopted a medium range of AI functions.Its AI systems, covering a few basic auditing tasks, are partially integrated into its audit platform.SML has a simpler AI focus, mainly targeting transaction scanning, and relies on stand-alone applications.

Data collection
We collected data from three participating firms after obtaining their participation consent.The primary data collection method was semistructured interviews with key informants from the three case firms.The selection of the interview method was driven by the research aim to obtain an in-depth understanding of AI adoption, where the aggregation of people's interpretation regarding the technology and the process can create knowledge at the firm level (Orlikowski and Gash, 1994).We also collected complementary evidence, such as firms' annual reports, promotion articles, information from the vendors, and professional  and Raymond, 2011;Trocin et al., 2023).An interview guide was developed from the literature and refined in a way that was pertinent to the context of AI and auditing.The TOE framework informed the themes in the interview guide.The semistructured interviews were guided by 23 open-ended questions exploring adoption-related factors (shown in Appendix 3).Interviews with 13 informants were conducted between April and July 2021.Within each firm, the interviewees were those who led, participated in, or influenced AI adoption and those who had extensive knowledge, experience, and understanding of the application of AI in their firms.After each interview, the interviewees were asked to refer to the colleagues who were in the best position to answer any questions that had not been fully addressed, as well as provide any useful documents that could enhance our understanding of adoption.To enhance the depth and reliability of our findings, we expanded our data collection methods, as informed by Eisenhardt (1989).Firstly, we conducted interviews with two auditing professionals not affiliated with the case firms to validate our findings.Secondly, we analyzed ten professional podcasts to provide a broader context and further validation.Such practices of augmenting  (Trocin et al., 2023;Young et al., 2018).Table 2 provides an overview of the key informants from each case firm, specifying their positions, the duration of the interviews, and the length of the resulting transcripts.The table also lists secondary documents collected from each firm.As supplementary data, validating interviews and professional podcasts from sources outside the case firms are also included in the table.

Mode of analysis
Guided by Gioia's methods for qualitative data analysis (Gioia et al., 2012), our data analysis process was intricately woven with the data " Fig. 1.Data analysis model.collection phase, a simultaneous approach to enrich the understanding of where to target the subsequent data collection (Urquhart et al., 2010).Gioia's method is particularly revered for its adeptness in visualizing the progression from raw data to discerned themes, a pivotal step in asserting analytical rigor (Jovanovic et al., 2021;Warner and Wäger, 2019).Delving into the specifics of our method, the collected data was initially read line by line in order to extract significant codes.In the initial round of coding, we identified 60 codes.Through subsequent comparison and consolidation, we refined this list to produce 20 firstorder codes.Pivoting from these codes, we discerned patterns and interrelationships, ultimately coalescing into six second-order themes.These themes were then nested under three overarching dimensions: technological (two themes), organizational (two themes), and environmental (two themes).The coding scheme and data structure were checked and agreed upon by all three authors to improve their rigor.Fig. 1 provides a graphic representation of how we progressed from empirical themes to the aggregate dimensions (Gioia et al., 2012).

Findings
This section presents the six factors influencing AI adoption across the three case firms: technology affordance, technology barriers, innovation management, AI readiness, regulatory environment, and competitive environment.The remaining section discusses how the constitutes and the impacts of the six factors vary among the three case firms.

Technological factors 4.1.1. Technology affordance
Technology affordance is "action possibilities and opportunities that emerge from actors engaging with a focal technology (Faraj and Azad, 2012, p. 5)."AI affords various opportunities to the three case firms, driving them to adopt the technology.Nevertheless, interviewees from three firms interpreted the affordance in different ways.In both BIG and MID, improvement in audit quality was the primary affordance, followed by better client experience and working efficiency.For example, interviewee A1 noted: "Improvements in the quality of our auditing and our audit product.We know that we standardize and automate; we know that quality improves (A1)." Similarly, interviewee B1 noted: "Probably the key drive at the moment is audit quality.It's a big drive from the corporate regulator idea to improve and become more consistent in our audit quality, so that's kind of the foundation of everything we try and do.And then when we're looking at specific tools, it's really around what's going to increase efficiency in the audit process (B1)."Nevertheless, in SML, decision-makers believed that AI primarily affords the firm to build a better business image, whereas better efficiency in the auditing process and the opportunities to provide valueadding services were the second-order affordances.Interviewee C1 noted: "[Using AI] is a marketing tool, market ourselves that we were different from other traditional firms at a similar size (C1)."Three firms recognized the affordance of AI differently, though they encountered the same kind of technology.The variation was caused by the different goal orientations within the three firms.BIG and MID had the primary objective to improve audit quality because they were subject to oversight by regulators who scrutinized the auditing quality of large and medium firms.By acknowledging the features of AI tools such as standardization and automation, decision-makers believed that there was a congruence between the technology features and their firmwide strategic goals.Therefore, the order of affordance they placed on AI reflected the strategic importance of improving quality, efficiency, and clients' experience.In contrast, at SML, the decision-makers prioritized brand building and client maintenance, as the biggest challenge it faced was high client turnover and extensive competition.They were aware that adopting AI would allow them to market the firm as being innovative, which was in line with their goal to improve their business image, attract new clients and keep their existing clients.

Technology constraints
Technology constraint refers to "ways in which an individual or organization can be held back from accomplishing a particular goal when using a technology or system (Majchrzak and Markus, 2012, p. 1)."Some technological features of AI created problems for three firms to achieve their goals to improve quality and efficiency, thus inhabiting their adoption.The major problems were the black-box problem, biases in development and use, and compatibility with clients' systems.The black-box problem refers to the issue that auditors cannot explain, interpret, or document the decision-making process of some AI tools (Dwivedi et al., 2021).The black-box problem first created a trust issue between the users and AI tools.For example, some data analytical tools were designed to learn from patterns in the raw data flow through the neural network.Because hidden layers were used in the network, auditors could not visualize the decision-making process of the tools that triggered the flags of risky items.Interviewee A3 noted: "Bear in mind that for all the designers or developers everything should be explainable.But in reality, it is not the case.[If not explainable, users] they don't trust and don't use it (A3)." The black-box problem is one of the most significant challenges for auditing firms to adopt AI-based tools, confirmed by the interview with an auditing professional outside the case firm: "I guess part of the challenge with anything with artificial intelligence, or any of that machine learning, is you can't quite see what's happening behind the scenes and so.Trying to communicate that and have that documented effectively on an audit file is probably one of the biggest challenges that comes through (E1)." The biases of AI tools were introduced by the people who participated in the development process and the data set used to train the algorithm.The bias interfered with the application of professional judgment and damaged the quality of decision-making, creating challenges to achieving better audit quality.For instance, Interviewee B2 noted: "A lot of the risks that we have to address around the actual completeness of the data, in order for it to be appropriate for decisions and judgments to be made based on that information (B2)." This view is shared by other auditing professionals, as Cobey expressed in the professional podcast (Corson, 2020): "And really right now what we're finding is that the performance of an AI, how accurate it is, is really dependent on a lot of different factors.How complete the data it has, the quality of that data, how it was trained, and how that is representative to know when it's operating within production." Because adopting AI created obstacles for the three firms to achieve their goals, they had to manage the abovementioned issues in their adoption process.BIG reported that the black-box problem required developers' extra effort on technology design and selection, and the compatibility issues needed data specialists to reformat the data.MID believed the biases of AI tools introduced compliance risks with quality standards.Thus, the national quality team led the roll-out of the tools and guided each engagement team on quality and compliance issues.
The involvement of the quality control team slowed down the adoption process.Nevertheless, not all the technological issues were solved effectively.For example, SML experienced a severe compatibility issue.AI held it back from accomplishing the goal of better profitability.It struggled to keep the cost of audits at the same level after adoption because of the high number of labor hours involved in converting the data from the client system to the AI tool.This was the primary reason SML discontinued the AI tool after a year.Interviewee C1 explained: "If the tools can't talk to MYOB [an accounting software frequently used by Australian small businesses], then it's really hard for us to apply at all, because otherwise, it's just so much time of manual work to adjust, reformat the data to be able to input into the tool (C1)." The finding reveals three technology constraints introduced by AI: the black-box problem, biases, and compatibility issues, which had a negative impact on the adoption within three firms.However, the impact varied among the three firms.The in-house development department in BIG and the quality control team in MID provided them with capacities and resources to manage the technology constraints.These resources, however, were not available for SML, which relied on the vendor to solve any technical issues.Therefore, the finding suggests that the impact of technology constraints was more substantial for SML, because it lacked sufficient resources and capacities to address those barriers.

Organizational factors 4.2.1. Innovation management
Incorporating AI into auditing is a process innovation; thus, innovation strategy and policies are important mechanisms that communicate innovation's necessity to subordinates (Baker, 2012).All three firms highlighted the importance of innovation in their strategy, mission, vision, and value statement, delivering a clear message to employees across different levels within the firm.
However, the three firms displayed differences in their approach to implementing their innovation strategy, stemming largely from their innovation policies and key linking agents of innovation.BIG was unique in having policies that fostered innovation, yet its stringent quality control regulations slowed adoption.MID focused solely on quality control, with no policies emphasizing innovation, while SML had neither innovation nor quality policies in place.We found that innovation policies greatly affected adoption across the firms.For example, BIG's pro-innovation stance led to collaborative efforts between auditors and in-house development teams, resulting in a grassroots adoption of AI tools throughout the firm, as highlighted by Interviewee A1: "We have citizen-led innovation, so that's from the grassroots up, where we encourage the innovation.We invest in skills; we give people the permission (A1)." In contrast, the lack of policies stimulating innovation in MID inhibited the bottom-up adoption of AI tools, as Interviewee B4 explained: "It's not like we have a policy where if you have an idea, raise it, and you'll have time to follow it up… in terms of managing adoption, it's from the center (B4)." In SML, the lack of innovation policy meant that the adoption was mainly driven by the top management, whose personal philosophy and experience primarily influenced the adoption process, as Interviewee C1 noted: "No (policies), I believe that's driven by the partners...He is innovative, forward-looking, and wants to bring changes technology-wise (C1)." The finding suggests that innovative policies enhance AI adoption from the junior level of employees, while excessive quality control policies slow down AI adoption; where no formal policies regarding innovation are in place, the senior management has a primary influence on AI adoption.Apart from policies, linking agents were also crucial in merging innovative strategies with AI tool adoption (Baker, 2012).These agents were champions who secure resources and drive AI tool diffusion, and gatekeepers who manage risks and ensure compliance.BIG and MID had a diverse set of linking agents, including senior partners and in-house teams in BIG, and appointed partners and directors in MID.In contrast, SML's sole agent was its managing partner, who was advocating disruptive technologies.While this streamlined adoption decision in SML, it also led to swift AI tool withdrawal due to compatibility costs.This suggests that having varied linking agents can facilitate AI adoption.

AI readiness
AI readiness refers to a firm's capacity to deploy and use AI in ways that add value to the firm (Alsheibani et al., 2018;Holmström, 2022).Our finding identified several crucial factors that influence this readiness, including financial resources, management's AI literacy, employees' digital skills, and the robustness of the firm's data infrastructure.
We observed a marked variation in available resources among the three firms, significantly shaping their respective AI acquisition strategies.BIG stood out with its considerable financial resources, which drove its decision to primarily invest in an in-house development approach.This strategic choice not only fortified BIG's autonomy, allowing it to tailor tools best suited to its needs, but also enabled it to develop advanced software beyond what was available on the market.On the other hand, MID, constrained by its middling financial and developmental capacity, opted for partnerships with tech companies, and often implemented tools devised by their affiliated networks.While this strategy appeared cost-effective, it had its drawbacks.Reliance on external entities meant MID's functionality of the tools was bound by the developing capacities of those parties.This dependency further introduced complexities, particularly in licensing and local customization, leading to adoption delays.Meanwhile, SML faced significant resource restrictions.With a lean IT team and a tight budget, it was compelled to gravitate towards off-the-shelf software.This route, while pragmatic, confined SML to tools that addressed only a very narrow scope of auditing tasks.In sum, the breadth and depth of AI tool functionalities adopted by these firms can be largely attributed to their financial resources.
Data infrastructure, which encompasses the systems and tools collecting, storing, processing, and managing data, serves as the prerequisite for AI adoption.The state of data infrastructure varied across the three firms.While BIG and MID were considered to have robust data infrastructures in place prior to AI adoption, SML's adoption was hampered by its absence.SML lacked a standardized data model to extract client data for AI tools.As Interviewee C3 highlighted, this is a common challenge in small firms: "I think it's harder for (SML) to start from zero and then solve AI and data competency at the same time.If I were to say what is the ideal customer profile for our adopters, it's usually the higher-end, top 100 audit firms, with data competency already enhanced of some description (C3)." Nevertheless, even with sturdy data infrastructures at BIG and MID, inhibitors to adoption arose from a different quarter: senior management's limited AI literacy.Interviewee B2 shared: "One of the barriers for adoption probably sits at my level on the partnership table, as they don't fully understand software or how the evidence is derived from that, and don't trust it on their audit files (B2)." A point of contention among the firms was the perceived adequacy of employees' digital skills.BIG saw a significant gap, with Interviewee A1 emphasizing the need for a new breed of auditors: "The auditor of the future will be a different skill set to the auditor of today, for sure.We have undertaken a huge digital up-skilling program across BIG.We recognized three years ago that investments in technology would not be fully leveraged unless we had an investment in digital skills (A1)." In contrast, MID's senior management felt that their auditors' digital skills were generally in line with current AI tools, with only ancillary skills like data preparation needed.These insights propose that as AI further permeates auditing, the demand for digital competencies intensifies.Notably, AI adoption was most extensive at BIG, moderate at MID, and minimal at SML.Consequently, the gap in digital skills among auditors was most pronounced at BIG and least at SML.Moreover, the lack of data infrastructure was especially problematic for SML, possibly due to resource constraints preventing the simultaneous adoption of AI and building data infrastructure.

Environmental factors 4.3.1. Regulatory environment
Regulatory environment refers to the laws, regulations and standards imposed by government agencies (Baker, 2012).Case data revealed that the regulatory environment influenced AI adoption.The auditing standards lagged behind practices, meaning that they neither prescribed nor prohibited the use of AI in auditing practice.However, silence from auditing standards created different obstacles for the three case firms.BIG had the most substantial concern over the compliance of auditing standards when adopting AI tools.Interviewee A2 explained: "So, I would say that one of our biggest constraints to adoption would be audit methodology and regulation because there are some principles embedded in the regulation and methodology that lend itself better to a human-delivered, sample testing approach than a machine-delivered full population testing approach (A2)." In contrast, MID had moderate concerns over the silence from the auditing standards, witnessed by the conflict of perceptions among decision-makers.Interviewees B1 and B4 believed that the lack of backup from standards reduced the scope of AI usage and slowed down the adoption process.Still, Interviewees B2 and B3 believed that auditing standards were flexible and thus were more facilitating than inhibiting.For example, Interviewee B2 noted: "I would pretty much say that they're more on the facilitator side, on the basis that all of our standards are principles-based.So, we don't have rules on what you can and what you can't do.So, you apply principles and make judgements (B2)." In SML, all interviewees agreed that the impact of auditing standards was neutral towards SML's adoption of AI tools.Interviewee C1 explained: "I don't think the auditing standards are actually encouraging or discouraging applying AI tools to audit.I don't see any impact either way because auditing standards are really guideline basis (C1)." Although the same set of auditing standards regulated all three firms, they had a different understanding of the influence of standards with AI.The primary reason is that the three firms had various degrees of AI involvement in the auditing practice.Across three firms, as the depth and breadth of AI applications increased, the lack of backup from standards increasingly became a concern.

Competitive environment
A competitive environment is the dynamic external system in which firms operate and compete with each other (Cornett et al., 2019).The Competitive Environment had a different impact on adopting AI by three firms.Competition had the most substantial impact on SML adopting AI, a moderate impact on the MID, and little impact on BIG.SML experienced intensive competition and high client turnover, driving it to adopt AI tools to attract and maintain clients.Most of its competitors relied on traditional auditing tools in its fragmented market; thus, adopting AI can differentiate SML from its competitors.C1 explained: "We're the pioneer of the market, so competitors need to look at us.But AI certainly helps us during the auditing tendering process … the market is very competitive.So, it's really some clients maybe going out for tender and try something new.It's hard to maintain the relationship and easy to lose clients (C1)."MID faced competition from large and medium firms in terms of audit pricing, as Interviewee B2 explained: "With audit, there's a lot of commercial pressures on audit fees.And so, as a result of that, then it's important that we have the most efficient processes to achieve the level of work that we need to do (B2)." Accordingly, adopting AI can release the competitive pressure, corresponding by the transformation of the pricing strategy from a traditional time-based model to a value-based model.Interviewee B3 explained: "We should be selling value, and time should have nothing to do with it, because when you buy at our firm, we don't charge you based on the number of hours it took.It's a value pricing strategy…It's really important that we do factor in the cost of the technology pieces and affect pricing models (B3)." Nevertheless, adopting AI can hardly be used as a marketing tool because firms with similar or bigger sizes have all adopted it.Interviewee B1 explained: "If we're going up against the big four, everybody's got basically the same slices in there, it talks the same talk.So again, it doesn't really stand you out (B1)."However, none of the interviewees from BIG believed that the adoption of AI was driven by competition.Interviewees A1 and A2 claimed that the primary driving force of adopting AI was to respond to regulators' concerns about audit quality and provide better services.Because all leading auditing firms have done extensive marketing on developing and using AI tools, it can no longer be a selling point to win clients.Interviewee A1 explained: "We think there's quite a lot of hot air in the market on this at the moment.Our challenge to the market and the other firms is to show us the technology and how it's going to be used on a reliance basis.So, not just using it to look at in front of a client, not just using it as a marketing tool, actually using it day to day on audits (A1)." Interviews with professionals outside the case firms also highlighted that some firms prioritized marketing their use of AI over its actual usage: "So, I think a lot of practices will promote that they're using AI when, in reality, it's nothing more complicated than an Excel spreadsheet than it's doing some basic what-if scenarios, so something like that.(E2)".
The finding suggests that adopting AI brought SML competitive advantages to differentiate itself from competitors, most of which were non-adopters.But this kind of advantage did not exist in BIG and MID, where many competitors had already adopted AI.In these two firms, adopting AI brought competitive efficiency and service quality advantages.

Discussion
Our findings identified six factors influencing the adoption of AI in three case firms.Technologically, the affordances and constraints of AI emerged as pivotal.Organizational dynamics were underscored by firms' innovation management and their readiness to adopt AI.On the environmental front, the competitive and regulatory landscapes were significant determinants.Interestingly, while these six factors were consistently observed across the case firms, there was a marked variation in how they manifested in each case.Table 3 provides a succinct summary of our key findings, emphasizing these six core factors and illustrating their variations among the three firms.It lays the groundwork for our discussion and interpretation of the findings.
When it comes to the affordances of technology, it is evident that all three firms recognized the competitive advantages offered by AI adoption.However, the manner in which the competitive advantages were built, what we term "affordances" (Burton-Jones and Volkoff, 2017), varied based on the firms' size and competition environment.For small firms, the competitive landscape remains dominated by competitors using traditional auditing tools.In this competitive environment with low AI penetration, the introduction of AI can differentiate firms from their competitors.However, the competitive advantage is built, not on using AI per se, but on marketing AI.Thus, the most cost-effective strategy is to adopt some AI tools, not necessarily with depth, but extensively market them to clients.We named this affordance as "marketing affordances." On the other hand, middle-sized and large firms may find themselves in an environment with high AI penetration.For these firms, merely adopting and promoting AI does not equate to a competitive edge since many of their competitors have already done so.Their path to competitive advantage lies in truly harnessing the power of AI to enhance operational efficiency, streamline auditing processes for a better client experience, and minimize human error to ensure superior service quality.These affordances, named 'operating affordances', spanning efficiency, client experience, and quality assurance, demand a comprehensive, in-depth AI adoption.Based on these findings, the relationship between firm sizes, AI-penetration in a competitive environment, perception of affordances, and AI adoption includes: • Smaller firms often operate in a lower AI-penetrated environment than larger firms.• The AI-penetration in the competitive environment influences the perception of AI affordances.• Smaller firms adopt AI mainly driven by marketing affordances, resulting in a lower depth and width of adoption.• Larger firms adopt AI mainly driven by operating affordances, resulting in a higher depth and width of adoption.
The findings show that the larger firms planning for an extensive and in-depth AI adoption may perceive the regulatory environment as a restrictive factor.In contrast, for smaller firms that envision a more limited scope of AI adoption, this inhibiting impact might not be as relevant.An interesting observation from our study is that, while auditing regulations have remained neutral towards AI, they have not affected firms of different sizes in the same manner.The largest firm has expressed the most significant concerns about adhering to these auditing standards during AI adoption.This apprehension appears to wane as the firm size decreases.One plausible explanation is that firms like BIG aim for a broader and deeper AI adoption.Hence, they expect the auditing standards to be more affirmative rather than just being neutral.However, for smaller firms like SML and MID, the extent of their planned AI adoption is less ambitious, making the neutrality of these standards less of a hurdle.
Contrary to previous research arguing that the lack of regulatory guidance on AI hampers adoption (Seethamraju and Hecimovic, 2022), our findings suggest a more sophisticated perspective.For larger firms with ambitious AI plans, a neutral regulatory stance could be seen as a stumbling block.In contrast, for smaller firms with less extensive AI goals, such neutrality might be less impactful.Based on these insights, we highlight that: • Larger firms aiming for larger-scale AI adoption perceive the neutrality of regulation as an inhibitor.• As the planned scale of AI adoption reduces, smaller firms do not perceive the neutrality of regulation as an inhibitor.
The findings also reveal the constraints imposed by AI, including data-led bias, the opaque nature of the AI, and incompatibility with existing systems.Interestingly, these constraints, while generally deterring AI adoption, did not produce a uniform result across all firms.The extent of their impact is moderated by the firm's AI readiness (Alsheibani et al., 2018).For instance, larger firms, with their wellestablished data infrastructure, substantial financial resources, and strong development capabilities, are positioned more favorably to navigate these constraints.Taken BIG, for example, their development team has strategically integrated a spectrum of machine learning algorithms in their AI software, balancing the interpretability and accuracy of AI outcomes.Unfortunately, it is observed that AI readiness diminishes in smaller firms.Lacking adequate resources and foundational data infrastructure, the constraints imposed by AI are magnified for smaller firms.To an extreme, such constraints have culminated in the termination of AI adoption.In light of these findings, we highlight the relationships between firm size, perception of AI constraints, firms' AI readiness, and their impact on adoption: • AI adoption is inhibited by the constraints imposed by AI technology, including data-led bias, black-box issues, and system incompatibilities.• The degree of AI readiness correlates with the size of a firm; larger firms typically demonstrate greater AI readiness.• Smaller firms experience a more severe adverse impact from AI constraints than larger firms, due to their limited AI readiness.
In our case studies, we consistently observed that innovation management is pivotal to facilitating AI adoption.Innovation management comprises three pillars: policies promoting innovation, quality control during innovation, and the role of linking agents.Policies that foster innovation are critical in energizing grassroots initiatives, driving AI uptake directly (Do et al., 2018).However, the very nature of innovation brings risks.Firms must therefore, balance their push for innovation with policies that manage quality and risk (Omrane, 2022).While the quality control policies might seem like a brake on AI adoption initially, it is a safeguard ensuring the long-term success and sustainable use of the technology.Linking agents are the connectors bridging broad innovation strategies with practical AI tool adoption (Baker, 2012).While senior managers provide direction and secure resources, they might not always grasp the specific use of AI tools in practice.Thus, it is vital to also have linking agents from the broader workforce, ensuring that ideas at the practical, individual level can be uplifted to the firm level.Our findings indicate that the comprehensiveness of policies and diversity of linking agents are important measures of innovation management.However, a disparity emerges when comparing the innovation management approaches among firms of different sizes.Larger firms, like BIG, tend to have well-rounded policies and a diverse set of linking agents.Conversely, smaller firms, exemplified by SML, often operate with fewer innovation policies and rely heavily on senior managers as their primary linking agents.This structure possibly makes AI adoption more challenging for smaller firms.In summary, we highlight the relationship between firm size, innovation management approach, and AI adoption: • Larger firms have a more comprehensive innovation management approach, spanning thorough policies and varied linking agents.• A firm's comprehensiveness of innovation management is positively related to its AI adoption.
To conclude, this study utilized the TOE framework to delve into the factors influencing AI adoption within professional services firms.Employing a multiple case study approach, we examined three auditing firms of varying sizes, gathering insights through interviews and supplementing with secondary documents.Our comprehensive data analysis yielded six salient TOE factors influencing AI adoption, revealing noteworthy variations among different-sized firms.

Theoretical contributions
The theoretical contributions of our study are twofold: firstly, the emergence of the notion of firm size as a critical factor for understanding technology adoption; and secondly, the integration of the affordance lens with the TOE framework.Our findings bridge the gaps in existing literature, offering a more holistic perspective on technology adoption.
Regarding firm size and its influence on technology adoption, prior studies often fall into two camps.Some have overlooked its impact, presuming that the factors influencing technology adoption remain consistent across varying firm sizes (Mahroof, 2019;Sharma et al., 2023).Others recognize firm size as a mere organizational factor, acknowledging its influence on adoption but neglecting the interconnected impacts it has on other adoption-related determinants (Kinkel et al., 2022;Pan et al., 2022).Both perspectives prove problematic.A plethora of research from diverse domains indicates that the factors influencing the adoption of emerging technologies diverge considerably between firms of different sizes, evidenced by studies on big data (Raguseo, 2018) and cloud computing (Karunagaran et al., 2019).Our work aligns with these findings, rejecting a generic one-size-fits-all model for technology adoption and underscoring the paramount importance of accounting for firm size.Absent this consideration, we encounter conflicting results in existing literature.For instance, while some argue that auditing standards encourage the adoption of AI and data analytics (Salijeni et al., 2019), others posit that these very standards act as barriers (Cao et al., 2015;Seethamraju and Hecimovic, 2022).Our study illuminates that factoring in the firm size can reconcile these contrasting views.Moreover, we challenge the viewpoint that merely lumps firm size with other organizational factors on AI adoption (Neumann et al., 2022;Prasad Agrawal, 2023).Echoing Baker (2012), we contend that firm size should be identified as a fundamental cause for the differences observed in adoption-related factors.It is essential to recognize that firms of different sizes exhibit varied levels of AI readiness and operate in distinct AI-penetrated environments (Alsheibani et al., 2018).As a result, different-sized firms experience diverse perceptions of technology affordances and constraints, as well as varied influences from both regulators and competitors.Therefore, our study underscores that a holistic understanding of technology adoption must consider the variation of contextual factors within and outside differentsized firms.
Incorporating the affordance lens into the TOE framework stands as another theoretical advancement in our study.Prior adoption studies predominantly emphasized perceived benefits and barriers as core technological factors (Oliveira and Martins, 2011;Pillai et al., 2022;Polisetty et al., 2023), limiting their scope to a mere technical dimension.This restricted perspective led to perplexing outcomes, such as that perceived benefits have little impact on adoption (Pan et al., 2022;Polisetty et al., 2023).Our study takes a socio-technical perspective, underscoring that firms, beyond evaluating intrinsic AI benefits, assess the alignment of AI's technological attributes with firms' strategic objectives.This realization gravitates us towards the affordance lens, which posits that the essence of a technology for a firm lies not in its standalone features, but in its synergy with the firm's strategic direction.The affordance lens is further enriched by distinguishing between perceived barriers and technology constraints (Majchrzak and Markus, 2012).While perceived barriers are the difficulties anticipated by firms during adoption (Oliveira and Martins, 2011), technology constraints refer to the problems that adopting technology limits the firms' ability to achieve firmwide goals (Leonardi., 2011).Our findings across three case firms reveal that the challenges hindering AI adoption were not perceived barriers to adoption, but rather the issues AI introduced in achieving enhanced efficiency, quality, and client experience, the key objectives for all firms.Thus, technology constraints, a concept emphasizing the intertwining between technological features and firms' goal orientation, may be more appropriate to study the hurdles to adopting technology.Aligning with recent scholarship (Mamonov and Benbunan-Fich, 2021;Shin and Park, 2019;Steffen et al., 2019), we assert the indispensability of the technology affordances and constraints for comprehensive adoption studies.

Practical contributions
While our research is contextualized within auditing, the findings and practical implications may be transferable to a wider array of professional service industries, such as legal and medical practices.These industries share similarities such as organizational structures, client interactions, professional workforce, and the balance between applying professional judgment and adhering to strict regulations (Sampson, 2021;Spring et al., 2022;von Nordenflycht, 2010).As such, other professional services firms may experience similar drivers, barriers, and facilitators of AI adoption as we found in this study.For example, much like auditing firms might leverage AI to bolster service quality and streamline operational processes, legal, consulting, and medical practices may similarly harness AI's capabilities to augment service quality and achieve elevated operational efficiencies.On the flip side, the impediments that auditing firms encounter, such as the lack of guidance from regulators, and the 'black box' and biases introduced by AI, could be reflective of challenges that permeate other professional service domains.
At its core, our study endeavors to furnish professional service firms with actionable insights to gauge their AI adoption prospects and readiness.One pivotal starting point is to assess the AI penetration within their competitive environment.In a market where most competitors rely on traditional approaches, even elementary AI adoption can offer firms a strategic edge.Publicizing AI-enabled services can serve as a significant market differentiator in such contexts.Conversely, in markets awash with AI innovations, firms must elevate their AI adoption scale and depth, embedding AI deeply within their operational paradigms to maintain a competitive stance.Next, aligning AI's potential with the firms' strategic objectives is pivotal.If the overarching ambition is operational efficiency, then solutions like Robotic Process Automation might be the most cost-efficient.However, if the strategic needle tilts towards transformative quality enhancement, diving into more sophisticated AI platforms, particularly those anchored in deep learning, becomes imperative.Finally, firms must evaluate their AI readiness.The multifaceted construct of AI readiness demands attention to not just financial capacity but also the AI literacy of leadership, the digital competencies of its workforce, and a conducive innovation management infrastructure.An integrated assessment of these dimensions ensures a streamlined trajectory towards successful AI adoption.

Limitations and future research
This research offers insights into the adoption of AI in professional service industries, particularly in the context of auditing.However, several areas of limitations exist, providing avenues for future research.The first limitation is the generalizability of our findings.While we have tried to present and discuss our findings in a way that can be applicable to a wider context, there are specific dimensions about some professions (such as unique knowledge bases and distinct governance frameworks) that may influence AI adoption.Consequently, firms in different professional service industries may experience subtle differences in factors influencing AI adoption.Hence, further empirical research is required in domains beyond auditing to validate or challenge our findings.
The second limitation is the exploratory nature of our study.Our study adopts an exploratory approach, aiming to provide a broad overview of AI adoption in professional service firms.Consequently, we were not able to delve into every specific detail.For instance, we identified various AI technologies, such as Robotic Process Automation, Natural Language Processing, and Predictive Modelling, as being adopted by firms.However, what influences the adoption of these specific types remains uncharted.Considering that different types of AI are characterized by varying levels of complexity, the factors influencing their adoption may also differ.Future research should aim to disentangle these AI technologies and meticulously study the factors influencing their adoption.
The third limitation is the qualitative approach we adopted.This approach is effective in exploring potential relationships between variables like firm size, technological affordances and constraints, innovation management approach, and the impact of regulation.Yet, the robustness of these identified relationships requires validation through quantitative examination.Future quantitative studies can provide statistical rigor and might confirm or reject the relationships proposed by this study.Finally, the scope of our study is restricted to the factors influencing AI adoption at the firm level.While it provides an initial step towards understanding AI's disruption to professional services, there are other crucial dimensions left unexplored.Future research should pivot towards examining other facets like the implementation, actual usage, and, most critically, the impacts of AI on professional services.

Declaration of competing interest
None.Financial justification in terms of ROI is the primary determinant affecting the adoption of AI and robotics in hotels.The participants pointed out the importance of cost savings, revenue increase, or customer experience improvements as the key factors that drive the adoption of AI and robotics in hotels.Internal IT expertise, resistance by employees, competition with rivals, and legal issues such as privacy are identified as third-order influencers.The study also found that AI would be, if it was not already, an effective tool to improve hotel operations
(continued on next page) J. Yang et al. (continued )

AI technology Auditing tasks Sample applications
Intelligent expert system and decision support system Internal control assessment (Changchit and Holsapple, 2004;Changchit et al., 2001), fraud risk assessment (Lombardi and Dull, 2016), going concern assessment (Chye Koh and Kee Low, 2004), materiality assessment (Comunale and Sexton, 2005) Embedding audit rules into an expert system shell that is widely used in the medical field, which can assist auditors in making fraud risk assessments.The system was created to address the current risk environment and a variety of contexts of interest, as the risk environment changes (Lombardi and Dull, 2016).

Natural language processing
Internal audit quality assessment (Boskou et al., 2019), continuous auditing (Fisher et al., 2016), fraud detection (Fisher et al., 2016;Goel et al., 2010), audit planning (Li and Vasarhelyi, 2018), risk assessment (Fisher et al., 2016;Zhaokai and Moffitt, 2019), substantive tests (Zhaokai and Moffitt, 2019), compliance assessment (Khan et al., 2021) A natural language processing-based text mining tool that can automate the textual analysis of disclosures of internal audit mechanism in annual report.The results show that classification models developed using text analysis can be a promising alternative proxy in assessing internal audit quality (Boskou et al., 2019).

Machine learning
Going concern assessment and bankruptcy prediction (Ding et al., 2019), fraud detection (Gray and Debreceny, 2014;Perols, 2011), assertion-based testing (Sifa et al., 2019), auditing document analysis (Sun and Vasarhelyi, 2018) A machine learning-based peer selection method with financial ratios that can select benchmark peers in audit.The method applies K-medians clustering to identify peer firms, where significant deviations between a firm and its peers may indicate potential anomalies (Ding et al., 2019).

Robotics process automation
Substantive test (Appelbaum and Nehmer, 2017;Cooper et al., 2019;Moffitt et al., 2018), analytical procedures (Moffitt et al., 2018), internal control assessment (Moffitt et al., 2018) Robotics process automation can automate rule-based auditing tasks that are repetitive and manual.Its use can repurpose the role of the auditor by replacing perfunctory tasks such as analytical procedures, internal control assessment and substantive test (Moffitt et al., 2018).

Table 1
Case description.
MID is a second-tier accounting firm with annual revenue of more than $300 million and nearly 200 partners from ten offices nationwide in 2020 SML is a Sydney-based boutique firm with five partners and 16 employees from three offices.The audit division generated $1.3 million in revenue in 2020 Adoption Status The use of AI in the audit division started in 2017 in the global network The Australian branch announced the adoption of AI in audit in 2017 Ongoing new applications development with 2-6 months development cycle Initial attempt for in-house development in 2016 (withdrawn in one year) Partnership with tech-company in 2017 Firm-wide pilot in 2021 • Full-population scanning • Risky transaction scoring Development Approach • Primary in-house development • Within-network tech sharing • Partnership with tech-company • Off-the-shelf application • Primary within-network tech sharing • Partnership with tech-company • In-house development • Off-the-shelf application • Off-the-shelf application System Integration Microservices/plug-in modules All integrated into the grand audit platform Some are not integrated into the grand audit platform Individual application a The classification was informed by prior research discussing the typology of AI Kokina, J., & Davenport, T. H. (2017).The Emergence of Artificial Intelligence: How Automation is Changing Auditing.Journal of Emerging Technologies in Accounting, 14(1), 115-122.doi:https://doi.org/10.2308/jeta-51,730,ibid.podcasts, to triangulate and validate the interview data (Poba-Nzaou

Table 2
Data collection and sources.
(Tankersley and Johnston, 2023)ng the Potential Impact of Data and Analytics on BIG (Internal report, 3800 words) An Overview of BIG's AI-enabled Auditing Platform (Online article from official website, 1359 words) BIG Responsible AI and Trust (Broacher,932 words) BIG Transparency report (2021 & 2022, 19,058 words) BIG's Award-winning Audit Product Enabled by AI (News release, 878 words) Case firm MID Key informants and positions Partner, Audit and assurance (B1) Partner and national leader, audit quality (B2) Director, Strategy and business development (B3) Senior Manager, Analytics, innovation and transformation, Audit quality (B4) Interview duration and transcript length The average duration of interviews is 55 mins, with a transcript length of 20,454 words.Secondary documents MID's Strategy, Mission, Vision and Value Statement (Broacher, 4345 words) MID Audit Innovation (Online article from official website,2444 words) MID Digital Audit Suite (Online article from official website, 2080 words) MID Transparency report (2021 & 2022, 23,856 words) Case firm SML Key informants and positions Manager, Audit and assurance (C1) Vendor, APAC Head, Analytics & industry insights (C2) Vendor, Chief Technology Officer (C3) Vendor, Vice President, AI and Product (C4) Interview duration and transcript length The average duration of interviews is 52 mins, with a transcript length of 28,152 words.Secondary documents The Control Points of SML's AI-enabled Auditing Tool (Internal document, 5976 words) SML's Strategy, Mission, Vision and Value Statement (Online Broacher, 478 words) Marketing Material of SML's Auditing Service (Online Broacher, 461 words) AI-driven Auditing Summary (Online article from vendor's website, 1715 words) Audit Automation with AI (Online article from vendor's website, 1924 words) Explainable AI in Audit (Online article from vendor's website, 2795 words) AI as a Driven Growth Force for Auditing Firms (Online article from vendor's website, 1004 words) Supplementary data sources Interview with auditing professionals to validate the findings of the study: • CPA Australia: How Data Analytics, AI & Automation are being Integrated into External audit (Grayston and Vasarhelyi, 2019); Duration: 91 mins; Length of transcript: 12853 words.•EY:HowArtificialIntelligence is Disrupting the Finance Function (Corson, 2020); Duration: 22 mins; Length of transcript: 4419 words.•Deloitte:TheTransformativePower of Humans with AI (Ammanath, 2020); Duration: 34 mins; Length of transcript: 5250 words.•RiskInsights:WhatAuditorsNeed to Know about Using and Auditing Artificial Intelligence (McGarrity, 2020); Duration: 18 mins; Length of transcript: 3266 words • KPMG: Artificial Intelligence (AI) -What does This Mean for Audit and How Can Businesses leverage Technology?(Smart and Campbell, 2022); Duration: 14 mins; Length of transcript: 2702 words • KPMG: Innovation in Audit(Bradley, 2021); Duration: 28 mins; Length of the transcript: 4300 words • CPA Practice Advisor: The Truthiness of AI(Tankersley and Johnston, 2023); Duration: 17 mins; Length of the transcript: 2818 words • Bloomberg: Technology Means Transformation for Audit Sector (Iacone, 2019); Duration: 21 mins; Length of the transcript: 2800 words data collection are well supported by prior research

Table 3
Finding summary.
Organizational competency boosts AI adoption, while complexity hinders it.Readiness enhances AI's perceived value but not its user-friendliness, possibly due to cultural aspects.Compatibility improves perceived utility, but not ease of use.A competitive advantage and partner support encourage adoption.Crucially, strong leadership elevates both the perceived value and ease of AI integration.Findings emphasize AI in risk reduction, supply chain integration, and information sharing.However, Technological, Organizational, and Environmental influences were discounted.This could be due to AI's novelty, a disconnect between academia and industry, or managers focusing on internal challenges over external environmental impacts on AI adoption.