This paper aims to provide empirical evidence on adopting artificial intelligence (AI), including generative AI, in knowledge management (KM) processes and its impact on organisational decision-making. Specifically, the study addresses three key research questions: RQ1: How is (generative) AI adopted within KM processes in organisations? RQ2: What factors influence the adoption of AI in these processes, either facilitating or inhibiting it? RQ3: How does AI adoption in KM processes affect organisational decision-making?
An explorative investigation has been conducted through semi-structured interviews with KM and AI experts from a worldwide sample of 52 mostly private, large and for-profit organisations. Interviews have been analysed through a mixed thematic analysis.
The study provides an original framework in which the three investigated concepts are interconnected according to a dual relationship: linear and retroactive and 20 factors affecting AI adoption within KM processes.
The provided model guides managers in improving their organisational decision-making through AI adoption in KM processes. Moreover, according to the rational decision-making model, the authors propose a six-step systematic procedure for managers.
To the best of the authors’ knowledge, this is the first study that simultaneously addresses AI, KM and decision-making and provides an integrated framework showing the relationships between them, allowing organisations to better and practically understand how to ameliorate their decision-making through AI adoption in KM processes.
1. Introduction
In the past, organisations struggled with the lack of data; nowadays – and even more in the future – they (will) struggle to handle its copiousness because the capacity to generate and collect data is rising exponentially (Bag et al., 2021; Justy et al., 2023; Mortati et al., 2023; Rothberg and Erickson, 2017). However, on average, 55% of this data are “dark”: they go unused because they are unknown (Splunk, 2019). Thus, a growing need has emerged to help organisations understand what to do with this massive amount of data and how to use it best to make more informed and, theoretically, fairer decisions (OECD, 2022; Yunita et al., 2022).
In fact, all this data significantly impacts organisations’ knowledge management (KM), understood as the process of identifying, organising, storing and disseminating data/information/knowledge within an organisation (Beesley and Cooper, 2008; Heisig, 2009). Traditional KM cannot process and analyse this massive amount of data effectively, enabling better decisions (Chierici et al., 2018). Thus, organisations are called to revise their KM models and processes (Chirumalla, 2013; Liu et al., 2023) to better collect, transform and use the new and copious available data (Manesh et al., 2020), ensuring positive impacts on their performance and competitiveness (Del Giudice et al., 2023; Lee et al., 2020).
In this vein, artificial intelligence (AI) seems to help ameliorate KM processes (Dragičević et al., 2022; Leoni et al., 2022; Sanzogni et al., 2017); thanks to its ability to acquire, process and use knowledge to perform tasks as well as its capacity to reduce uncertainty and unlock knowledge that can be delivered to humans to improve decision-making (Gupta et al., 2022; Haenlein and Kaplan, 2019), with positive impacts on the overall firm performance (Mikalef and Gupta, 2021). In other words, AI may “support decision-making and knowledge management” (Brock and von Wangenheim, 2019, p. 115), creating more effective, accurate and flexible organisational decisions (Agrawal et al., 2017a; Metcalf et al., 2019; Wilson and Daugherty, 2018). This is even more true if we consider the emergence of generative AI tools (like ChatGPT, GitHub Copilot and AlphaCode), which are leading in a new KM era by revolutionising the way knowledge is generated, synthesised and applied (Alavi et al., 2024).
However, the relevance of KM literature and its integration with AI advancements has often been overlooked. It is observed that there exist certain unexplored areas regarding the results, obstacles and constraints associated with the implementation of AI within KM processes in contemporary organisations’ decision-making (Al Mansoori et al., 2021; Oppioli et al., 2023; Trunk et al., 2020). In fact, although numerous studies have highlighted the enhancements AI brings to KM in areas like knowledge acquisition and problem-solving, there is a significant gap in understanding the full spectrum of AI applications within all KM processes. This includes a lack of comprehensive strategies for integrating AI into KM processes (Taherdoost and Madanchian, 2023) and insufficient research on how AI adoption affects knowledge workers (Budhwar et al., 2022) both in terms of opportunities (e.g. improving individual performance, reinventing talent management practices) (Claus, 2019; Malik et al., 2023a) and challenges/concerns (e.g. risks related to well-being, bias, privacy issues, ethical dilemmas, etc.) (Budhwar et al., 2022). This underscores the need for a deeper exploration of how AI technologies, particularly generative AI – integrated with KM processes – operate at both individual knowledge workers and organisational levels (Davenport, 2007), influencing organisational decision-making and outcomes (Kudyba et al., 2020; Davenport et al., 2022).
Furthermore, many studies only discuss (generative) AI’s potential in KM (Liebowitz, 2001; Jarrahi et al., 2023; Alavi et al., 2024) and, with specific reference to knowledge workers, many of them (Budhwar et al., 2022; Malik et al., 2023b) provide frameworks belonging to systematic literature reviews, thus highlighting the need for more first-hand-based research. In this vein, Pauleen and Wang (2017) and Korzynski et al. (2023) call for further empirical research investigating AI adoption within KM processes to ameliorate organisational decision-making, along with the elements that contribute to or hinder such adoption (Kinkel et al., 2022).
Thus, this paper aims to answer the following research questions:
How is (generative) AI adopted within KM processes in organisations?
What elements favour or inhibit (i.e. influencing factors) this adoption?
How does AI adoption in KM processes impact organisational decision-making?
To answer these questions, an explorative investigation has been conducted through semi-structured interviews with KM and AI experts – which represent the unit of analysis of this research – from a worldwide sample of 52 organisations, mostly private (86.5%), large (61.5%) and for-profit (92.3%) organisations, belonging mainly to the following sectors: ICT/IT service and consulting (19.2%), consulting (17.3%), financial service (7.7%), hospitality (5.8%), retail (5.8%) and transport (5.8%). Interviews have been analysed through a mixed thematic analysis.
The study provides an original framework in which the three investigated concepts are interconnected according to a dual relationship: linear and retroactive and 20 influencing factors related to AI adoption within KM processes. Accordingly, specific managerial guidelines are provided for improving organisational decision-making through AI adoption in KM processes.
The rest of the paper is structured as follows. In Section 2, the theoretical background is presented by focusing on the role of AI within KM processes and on the role of AI-empowered KM processes in companies’ decision-making processes. Section 3 reports the research design, data collection and data analysis of the proposed investigation. Section 4 is devoted to the findings, which have been used to develop an integrated framework synthesising the phenomena under investigation, as reported in Section 5, together with theoretical and practical implications. Finally, Section 5 presents the study’s conclusions, main limitations and possible future directions.
2. Theoretical background
2.1 Role of artificial intelligence in knowledge management processes
In today’s rapidly evolving era, characterised by uncertainty and dynamism, organisations increasingly recognise the importance of implementing digital technologies to effectively manage their knowledge processes (Al Mansoori et al., 2021; De Bem Machado et al., 2021). Indeed, KM processes – as identified by Heisig (2009) – focus on the processes related to the identification, acquisition, creation, storage, sharing and application of valuable knowledge for the organisation (Alavi and Leidner, 2001). Thus, organisations are exploring how KM processes can be combined with new digital technologies to ameliorate the processes themselves as well as the organisation’s overall performance (Gao et al., 2021; Husain and Ermine, 2021; Leoni et al., 2022). Consequently, technology has acquired an increasingly central role in the KM domain (Bhatt, 2001; Geisler and Wickramasinghe, 2015), enabling people to broaden the knowledge spectrum and make it accessible to everyone, anywhere, anytime (Al Mansoori et al., 2021).
In this context, AI stands out as the most prominently studied and focused upon digital technologies, garnering unparalleled interest for its role in enhancing KM processes (Husain and Ermine, 2021; Leoni et al., 2022). Basically, AI gives computers the ability to perform cognitive functions usually occurring within the human brain (such as reasoning and learning), solving complex problems previously tackled by human experts (Lei and Wang, 2020). By doing so, AI may reveal (or unlock) new knowledge from vast quantities of data that can be delivered to humans to improve their decision-making (Paschen et al., 2020; Vajpayee and Ramachandran, 2019).
Hence, according to Al Mansoori et al. (2021) and Bencsik (2021), there is a close mutual interaction between AI and KM, where AI adoption can effectively take KM to the next level. Indeed, AI can be used to recognise patterns and correlations between two or more data sets, and it may provide organisations with intelligent agents for various operations such as user profiling, pattern matching and text mining (Al Mansoori et al., 2021; Sundaresan and Zhang, 2022). Moreover, thanks to the prediction capabilities of AI-based technologies, it is possible to make assumptions about how future events might affect organisations (Ganesh and Kalpana, 2022; Jauhar et al., 2023), facilitating predictive analytics for risk assessment, machine learning algorithms to adapt to fluctuating market dynamics and intelligent automation to increase efficiency (Ivanov, 2023; Zamani et al., 2022).
The disruptive role of AI in KM processes is even more evident if we consider generative AI tools (Korzynski et al., 2023). These tools can take raw data and “learn” to produce various types of content, including text, imagery and audio (Harreis et al., 2023), enabling people to broaden the knowledge spectrum and make it accessible to everyone, anywhere, anytime (Al Mansoori et al., 2021), making KM processes potentially less expensive and more pervasive and powerful (Al-Emran et al., 2018).
However, although the enabling power of (generative) AI in KM processes is widely recognised (Agrawal et al., 2017b; Duan et al., 2019), many studies are centred exclusively on the potential role that (generative) AI may have in KM (Liebowitz, 2001; Jarrahi et al., 2023; Alavi et al., 2024), neglecting empirical evidence. In this vein, it is worth mentioning that some exceptions exist, but they focus on single KM processes. For example, Deng et al. (2023) investigate the use of digital technologies in facilitating KM only concerning the knowledge-sharing process, whereas Chin et al. (2024) address how AI–human interactions can be interpreted as a knowledge creation system.
Thus, a significant gap exists in understanding the full spectrum of AI applications within all KM processes. Filling this gap represents the first aim of this investigation – RQ1: How is (generative) AI adopted within KM processes in organisations?
Furthermore, it is worth noting that despite the recognised benefits deriving from the application of AI in KM processes, there are still many difficulties that organisations encounter in carrying out the effective and efficient implementation of these new tools (Davenport et al., 2020), as well as numerous resistances, especially in terms of AI’s positive/negative impacts at both organisational and individual levels (Budhwar et al., 2022; Claus, 2019; Dwivedi et al., 2021; European Commission, 2020).
As clearly stated in the perspectives editorial by Budhwar et al. (2023), the outcomes generated by the adoption of AI can be interpreted as two sides of the same coin, i.e. positive and negative, powerfully highlighting the need to understand the broad spectrum of factors that influence its adoption and, therefore, the results achievable (Kinkel et al., 2022). For example, dealing with biased data (i.e. biased AI algorithms), cybersecurity and privacy issues and transformation processes related to the adoption of AI represent crucial challenges (Malik et al., 2021, 2023a).
Accordingly, providing an overview of the factors that promote or hinder AI adoption, with specific reference to KM processes, represents the second aim of this investigation – RQ2: What factors influence the adoption of AI in these processes, either facilitating or inhibiting it?
2.2 Artificial intelligence-empowered knowledge management processes in organisations’ decision-making
Recent studies highlighted how the integration of AI into KM processes not only accelerates the pace of knowledge dissemination but also enables organisations to ameliorate their decision-making processes (Alshadoodee et al., 2022; Caputo et al., 2023; Oppioli et al., 2023). In fact, AI tools are considered a powerful means to help organisations and individuals achieve better decisions (Nazeer et al., 2023), even more so when generative AI is adopted within KM processes to store, transform and distribute data, information and knowledge (Korzynski et al., 2023). In this vein, Trunk et al. (2020) stress the increasing importance of managers’ education in properly using AI in decision-making. Moreover, it has been recognised that when decision-making processes are AI-based, they are “more ‘informed’ because the exchange of information is rapid (often in real-time) [so, they are more] precise, punctual, efficient and valid” (Caputo et al., 2023, p. 2800).
Thus, AI-empowered KM processes improve efficiency in knowledge transmission, capture, storage, analysis, visualisation and interpretation (LaValle et al., 2010; Chen and Zhang, 2014), leveraging the vast amounts of data generated and disseminated through increased automation and big data usage (Carlucci et al., 2020; Iandolo et al., 2021) and revealing valuable insights on which to assist employees at various decision-making levels (McAfee et al., 2012; Meski et al., 2019), in both private and public sectors (Di Vaio et al., 2022). In sum, AI adoption within KM processes boost organisations’ decision-making abilities by making them more effective, accurate and flexible (Agrawal et al., 2017a; Dennehy et al., 2022; Duan et al., 2019; Metcalf et al., 2019).
Despite the above-mentioned positive aspects, it is worth mentioning that how KM processes and AI tools should be combined into KM processes to improve organisations’ decision-making processes seem to be still underestimated (Caputo et al., 2019; Chinnaswamy et al., 2018; Russo et al., 2023; Singh and Del Giudice, 2019). In this vein, according to Malik et al. (2023b), it is necessary for there to be an interaction between managers and AI to obtain valuable decisions. This interaction can be of different types, considering the numerous concerns (e.g. accuracy, explainability, data privacy/security) related to AI adoption. Thus, there is a need to establish a climate that facilitates human–machine collaboration within KM processes in organisations (EU, 2021; Liebowitz, 2021; Xiong et al., 2022).
However, no clear approaches on organising KM and AI for decision-making are currently provided (Trunk et al., 2020), raising a need for more empirical analyses (Di Vaio et al., 2022). This represents the third and last aim of this investigation – RQ3: How does AI adoption in KM processes affect organisational decision-making?
3. Methodology
3.1 Research design
This paper follows a qualitative approach as the best one for describing, interpreting and gaining in-depth insight into the phenomena under investigation (Azungah, 2018; Cao et al., 2021) and because it has already proven to be valid for similar studies (Mortati et al., 2023).
To collect primary data, semi-structured interviews at a distance (Saarijärvi and Bratt, 2021; Seidman, 2006) were conducted with KM and AI experts, who constitute the unit of analysis of this research, from 52 worldwide organisations between January 2021 and September 2023. Authors opted for semi-structured interviews as the right compromise between formality and informality. Indeed, this type of interview proves to be flexible, accessible and intelligible (Qu and Dumay, 2011), allowing a guided conversation between researchers and participants thanks to the interview protocol and the possibility for researchers to probe participants for additional details.
The semi-structured interview protocol has been validated through a pilot test (Malmqvist et al., 2019) with three experts. The pilot tests aimed to assess the suitability of the questions and offer early insights into the feasibility of the research. In addition, it allowed the authors involved in the data collection to gain experience in conducting in-depth, semi-structured interviews by further developing their interviewing skills and understanding of conversation flow dynamics.
3.2 Data collection
Two authors have conducted the data collection process due to their direct access to elite informants. To select the interviewees, an email was sent to all the professionals from the two authors’ networks who, due to the role held, should have had adequate experience and knowledge in the field of AI, KM and organisational decision-making. The email explained the research project and its main objectives and asked the professional if they were interested in participating in the project. All those who showed interest received a second email containing the basic structure of the interview to verify ex ante the adequacy of their position and knowledge concerning the topics that would have been addressed in the interview itself. After this second email, 52 professionals confirmed their willingness to join the project and were interviewed. According to Guest et al. (2006), concept of data saturation – i.e. the idea that additional interviews become redundant if they only reiterate what is already been learned from prior interviews – the authors felt to have reached this threshold with the 49th interview. Nonetheless, three extra interviews were conducted to confirm that no additional insights were missed.
Recognising the potential for self-selection bias – where companies willing to participate might systematically differ from those unwilling – we implemented proactive measures to mitigate this risk. Invitations to participate were extended to various experts and incentives were offered to encourage participation. These incentives included the promise of a report on the main research findings and a reserved place at a closed-door dissemination workshop. This approach aimed to minimise the likelihood that only certain types of experts would respond, thereby enhancing the diversity and representativeness of our sample (Groves et al., 2009).
The interview protocol was organised into five parts (Castillo-Montoya, 2016): participants were informed about the study aim; authors interviewed respondents starting with demographic questions (e.g. their specific job role, the year of experience); the interviews shifted towards more study-objective specific questions; participants were encouraged to provide examples; and participants were invited to provide additional comments.
All interviews were conducted online, digitally recorded, transcribed and translated (when necessary) from Italian into English by the authors to ensure consistency and avoid language bias. Moreover, all transcripts were merged to build a unique data set and triangulated with corporate documents, press articles and other publicly available materials.
Each participant signed an interview release form in which they specified whether the authors were authorised to use their organisation’s name. Moreover, each of them received the transcript of their interview “principally to validate what was said during the interviews […] or discover and correct errors or inconsistencies” (Mero-Jaffe, 2011, p. 236).
It is worth underlining that, of the 52 interviewed, most of them belong to large (61.5% with more than 1,000 employees), private (86.5%) and for-profit (92.3%) organisations. In comparison, only six (11.5%) belong to public organisations and only one (1.9%) to an intergovernmental agency. In terms of years of activity, the average is 15.2 years, whereas, in terms of role, most of them are knowledge manager (23%), digital innovation/solutions/transformation manager/head (15.4%) and CEO/found/owner (15.4%); thus, they demonstrate adequate experience in terms of KM, AI and decision-making. Finally, the participants belong to organisations mainly in the following sectors: ICT/IT service and consulting (19.2%), consulting (17.3%), financial service (7.7%), hospitality (5.8%), retail (5.8%) and transport (5.8%).
Table 1 provides further details on the type of informants and related organisations.
3.3 Data analysis
A verbatim transcription process was carried out for all the interviews to improve the research’s reliability, validity and trustworthiness (MacLean et al., 2004). The analysis of transcripts – and the triangulation with other sources – has been conducted through a mixed thematic method (Braun and Clarke, 2006; Clarke and Braun, 2017; Yin, 2014), drawing from both deductive analysis, in which communication messages are categorised according to an initial codebook, and inductive analysis, where new themes are allowed to emerge. According to the research questions and the theoretical foundation of this study, the already identified KM processes have been used as the initial codebook for the deductive analysis.
Three authors took part in the analysis process, identifying and categorising all the relevant information in the interviews’ transcripts according to specific steps (Knox et al., 2021). In the first step, each participant’s transcript was individually examined to obtain the first-order codes (Spiggle, 1994). A “multiple rater” approach was used (Scandura and Williams, 2000); thus, the authors analysed transcripts individually, and when disagreeing, they deepened the analysis to find a shared vision. In the second step, the authors grouped codes into second-order constructs, i.e. categories and categories into themes based on content analogies and affinity, to identify common patterns and differences. In the final step, the authors extrapolated a unified model synthesising how (generative) AI adopted within KM processes affect organisational decision-making. An extract of how the thematic analysis was conducted is shown in Figure 1.
4. Findings
4.1 Artificial intelligence adoption in knowledge management processes
All the different interviewees confirmed the crucial role that AI is currently playing in KM processes within their organisations. They state that AI is nowadays indispensable to carry out KM processes for three main reasons: the number of collectable data and information has increased in a potentially infinite way, it is unthinkable that humans can perform the analysis, integration, dissemination and storage of this infinite variety effectively and efficiently in space and time and the organisation’s overall performance is ameliorated thanks to the possibility to make better decisions through the application of AI-enhanced knowledge. Thus, according to interviews, (generative) AI pervades all KM processes, playing a crucial role in organisational decision-making (Table 2).
Adopting (generative) AI within KM processes is far from simple. In this respect, interviewees emphasise that, to choose an AI tool, the organisation must make careful and multiple analyses without being “exclusively” influenced by current trends. In other words, what kind and how many AI tools an organisation has to adopt can be understood as a wise balance between multiple influencing factors (as detailed below).
4.2 Influencing factors
4.2.1 Benefits.
The first factor that strongly emerged from the interviews and which “justifies” the increasingly massive adoption of AI by organisations refers to the benefits that this use brings (or at least should bring) to the organisations. According to the different interviews, six benefits can be identified, as reported in Table 3.
What interestingly emerges from Table 3 is that using AI improves organisations from a double aspect. From an operational point of view, AI within KM processes allow organisations to reduce the time and cost of carrying out tasks. In this vein, when tasks are performed in real time, organisations can identify patterns, trends and anomalies as they emerge, making decisions on up-to-date information. According to interviewees, this real-time processing enhances operational efficiency, allowing organisations to know immediately where they need to allocate resources; improves customer experience, allowing organisations to anticipate customers’ needs and/or adjust their product/service accordingly; and ameliorates risk management, allowing organisations to identify potential risks before they impact business operations. From a cognitive point of view, AI adoption can act as a stimulus for humans, allowing individuals/employees to increase their potential (by learning more and better, improving their creativity, and the like), which in turn benefits the organisation.
4.2.2 Enablers.
Interviews highlighted three main AI enablers (Table 4) and the most important one is the employees’ – especially managers’ – commitment. In this respect, several managers’ crucial activities have been underlined, such as the importance of organising the “AI journey” according to a double perspective, i.e. top-down and bottom-up, to ensure the organisation as a whole has the opportunity to understand better how (the method) and why (the scope) the new AI tools will be adopted, as well as the possibility to express their opinion on the matter. It is also essential that managers are fully aware of the time required for the transformations linked to using AI tools to be fully implemented and thus bring about the expected results. In other words, an agile organisational structure is essential.
It is worth mentioning that, among the enablers, the number of customers also emerges. In this respect, the increase in the number of customers provides a larger data set for an organisation, enhancing the effectiveness of AI adoption that can enable more accurate predictions and personalised recommendations. In addition, a more extensive customer base often means more resources, facilitating investment in AI technologies and expertise. Finally, leveraging AI with a more extensive customer base enables organisations to improve decision-making, enhance customer experiences and gain a competitive edge in the market.
4.2.3 Barriers.
Respondents acknowledge AI’s relevance for KM processes but point out many barriers to its adoption (Table 5). Some of these barriers are typical of any other change/transformation affecting an organisation, such as the “conservative” organisational culture or the lack of adequate financial resources. Different barriers, however, are closely linked to the intrinsic characteristics of AI. In this sense, many interviewees emphasised that one of the most significant barriers to implementing the appropriate AI tools is the presence of unqualified personnel. In this respect, the critical need for organisations to invest in training and development programs to upskill existing employees or recruit qualified professionals emerges. Failure to address this barrier may result in suboptimal utilisation of AI, hindering the organisation’s innovation, efficiency and competitiveness.
Another barrier that deserves specific attention regards the organisational size. In this respect, interviews stress how the small company size hinders AI adoption. This could lead to two main consequences: larger organisations – with greater access and capability in using AI – may solidify their dominance in the market, potentially leading to a less competitive landscape, which could result in fewer choices and higher prices for consumers; smaller organisations may struggle to keep up with technological advancements, risking their competitiveness and survival in the market. Therefore, policies and initiatives promoting equal access to technological resources and supporting the development of smaller businesses are crucial for fostering a more balanced and competitive market environment, ultimately benefiting consumers.
4.2.4 Concerns.
Participants also pointed out numerous concerns related to AI adoption (Table 6). These concerns are multiple as AI mixes with ethical, moral and legal aspects, making justifying its use and outcomes particularly complex. Several interviewees highlighted how AI can trigger vicious cycles regarding human creativity, destroying it rather than stimulating it. All this brings out again the importance of adopting a conscious approach to the use of AI to exert its benefits to the maximum, reducing threats and challenges to a minimum.
Other concerns that deserve attention are “liability” and “overestimation”. Concerning the first aspect, as AI becomes increasingly autonomous, determining responsibility for errors or damaging outcomes becomes complex, potentially leading to legal disputes and challenges. This is even more true if we consider the current absence of clear regulations and legal frameworks. At the same time, liability concerns could hinder innovation and adoption of AI, slowing down progress in various sectors.
Concerning the second aspect, overestimating AI capabilities can lead to unrealistic expectations and reliance on AI without proper oversight or human intervention, leading to incorrect investments in organisational resources.
4.3 Decision-making
Once AI tools have been identified and the related influencing factors analysed, they are implemented and run accordingly within the different KM processes, aiming to ameliorate the possibility of making decisions that are as accurate, aware, informed and right as possible. In this vein, one of the most exciting aspects that emerges is that – according to participants – decisions can be classified according to the problem they wish to solve (from simple to complex ones). Simple problems deal with every day and frequently occurring difficulties that have established procedures, whereas complex problems encompass multiple relationships, and there is no obviously correct choice or established procedure. Based on that, the interviewees agree that AI can make the final decision, provide the decision maker with possible alternatives and provide simple support to the decision maker. This leads to three main decision-making (Table 7): automated decisions: AI has the agency to make decisions autonomously and humans can benefit from AI decision-making scalability, speed and consistency (e.g. automated planning and scheduling); augmented decisions: AI recommends one or multiple alternatives, reducing complexity; thus, humans can rapidly analyse and make decisions (e.g. simulations); supported decisions: AI provides valuable insights but are humans – through their common sense, judgement and expertise – that identify the best decision (e.g. hiring processes). The latter circumstance is particularly true when referring to decisions of a strategic nature that – according to interviewees – must always be made by humans, with AI just supporting them.
Thus, the increasing use of AI within KM processes leads to a redefinition of decision-making dynamics within organisations, which entails redefining the relationship between human agency and AI agency. In this vein, it clearly emerges from the interviews that – even though both humans and AI can make decisions independently – the best results come out when they collaborate, i.e. when humans and AI interact and co-create the final decision, allowing organisations to reach their goals:
It is the intersection between the two [man and AI] that creates the real added value for the organisation. If both are left free to express their full potential, this allows us to co-create the “perfect” experience for our customers [Interview #24].
Yes, the best decisions are those based on AI and humanity, but their relationship is not fifty-fifty. I mean, although autonomous AI makes many decisions, we keep the decision of what decisions to offload and what to maintain [Interview #33].
Sometimes decisions are made by AI, sometimes by humans; it is not easy to find the balance, but it is necessary [Interview #48].
5. Discussion
Findings from the interviews have been used to develop an integrated framework synthesising the results obtained through this investigation (Figure 2). The model relates the three investigated concepts, i.e. AI, KM processes and decision-making, highlighting how there is not only a linear relationship between them but also a retroactive one, as graphically shown in Figure 2 and emphasised by many interviewees:
It's not just about the [AI] tools themselves. It is about understanding how they shape decision outcomes and how those outcomes, in turn, inform our future choices regarding AI [Interview #26].
It's a loop: our results inform our future choices that will influence our future results, and so on [Interview #38].
AI tools are adopted and improve KM processes, which has a positive impact on the decisions we make. At the same time, the success – or failure – of the results achieved through those decisions will influence how we perceive AI in terms of benefits or problems related to its use. This, in turn, will impact whether and how we re-adopt AI within our KM processes and, therefore, ultimately influence our decisions. This mechanism repeats itself infinitely, self-generating and self-influencing. [Interview #42]
According to Figure 2, there is a linear relationship between the different elements that compose the framework: based on the perceived benefits, but also of the identified barriers and all the other influencing factors, the organisation chooses its AI tool(s) and within which KMP adopt it; this generates an ad hoc decision, according to the complexity of the problems that need to be solved, which will lead to specific results (positive/negative). At the same time, the obtained results will influence all the other elements of the framework thanks to the existence of retroactive feedback: based on the results derived from the decisions made, the perceived influencing factors change, as well as the choices made regarding the AI adopted and the related KMP. The relationship between AI, KM processes and decision-making underscores the importance of successful AI integration in KM processes, as it directly influences decision results as well as the need to carefully evaluate AI’s effectiveness in achieving desired outcomes, as perceptions of its benefits or drawbacks can shape future adoption decisions.
In other words, the provided framework shows that AI, KM and decision-making are interconnected and mutually influence each other, in line with previous contributions that look at organisations as complex systems (Holland, 2006; McElroy, 2000), in which decisions are formed based on the interactions among multiple factors and actors (Cristofaro et al., 2021; Parasuraman et al., 2000). In this specific case, these interactions are made even more complex by the human–AI interplay (Di Vaio et al., 2022). In this vein, the framework suggests that AI mainly play a complementary rather than competing role with humans (Budhwar et al., 2022), identifying three specific decision-making typologies (i.e. automated, augmented, supported) according to the problem that needs to be solved and the related complexity.
Moreover, the proposed framework includes a feedback loop that helps the organisation learn and improve its knowledge base over time (Argyris, 1990; Argyris and Schön, 1997). In fact, the outcomes of decisions are fed back into the system as “additional knowledge”, allowing the organisation to store explicit knowledge from decision-makers and make it immediately available within the overall system (Nemati et al., 2002).
In sum, decision outcomes not only influence future AI adoption and KMP but also reshape the organisation’s perceptions of AI benefits and challenges. Moreover, as decisions inform future choices and outcomes, organisations accumulate knowledge and refine their approaches to AI adoption and decision-making. This dynamic interplay suggests a more nuanced understanding of how the investigated elements interact over time, ameliorating organisational agility and responsiveness to changing environments.
5.2 Theoretical implications
The results of this study expand the literature that links AI and KM (Liebowitz, 2001; Sanzogni et al., 2017; Kinkel et al., 2022) by providing empirical – and not just potential (Alavi et al., 2024; Jarrahi et al., 2023) – evidence of its adoption by organisations and of the underlying motivations – in terms of benefits, enablers, obstacles and concerns – which lead knowledge workers to increase their awareness of integrating AI tools into their daily practices and decision-making processes (Kudyba et al., 2020).
The relationship identified between AI adoption, KM processes and decision-making emphasises the interdependence between these topics, highlighting their iterative nature and enriching current theoretical discussions on them (Cao et al., 2021; Duan et al., 2019; Dwivedi et al., 2021). In particular, findings support recent studies that stress the importance of establishing a clear human–AI agency relationship (Chin et al., 2024; Kang and Lou, 2022; Sundar, 2020), as strongly advocated by Budhwar et al. (2022), to promote fruitful human–machine collaborations to get the best from both when acting synergistically (Jarrahi et al., 2023; Xiong et al., 2022; Wilson and Daugherty, 2018), especially in terms of decision-making (Bader and Kaiser, 2019; Metcalf et al., 2019). At the same time, our results mirror those of Bigman and Gray (2018) and Haesevoets et al. (2021), revealing that people are more willing to accept AI involvement in managerial decisions only if machines play a supportive role.
Moreover, the provided results also contribute to the human resource management literature (Malik et al., 2023b) by providing empirical evidence on how AI adoption is transforming the nature of work, workers and the workplace, highlighting both positive (e.g. ameliorating employees’ evaluation/training/learning) and negative (e.g. unskilled human resources) aspects.
Furthermore, findings support recent studies (Caputo et al., 2023; Paniccia et al., 2024) that stress the importance of considering the time variable as well as the capacity of the organisation to make proper estimations in terms of time needed for the AI implementation. In this vein, AI adoption allows organisations to perform specific tasks of KM processes in real time, enabling more informed and timely decisions (Tien, 2017).
Finally, we corroborate and advance previous studies that attempted to identify the factors influencing the adoption of AI (Cho et al., 2023; Kinkel et al., 2022) by providing a list of 20 influencing factors (benefits, enablers, barriers and concerns) of the AI adoption in KM processes, offering a more comprehensive understanding of the complex dynamics surrounding such adoption.
5.3 Managerial implications
Managers can view the provided model as a guide to verify how their organisation performs in decision-making through AI adoption within KM processes. In this vein, the relationship among investigated topics stresses the importance of adopting a holistic approach. Accordingly, we propose a revisited version of the six steps of the rational decision-making model (Schoenfeld, 2010) to allow managers to fully grasp and exploit the interconnectedness between AI, KM processes and decision-making by providing structured guidelines on how AI can effectively support the decision-making process by enhancing KM practices:
Identifying the problem(s) to solve with AI. Before implementing AI, organisations need to clearly understand what problem(s) they would like to solve through its adoption. This may involve identifying specific knowledge gaps or inefficiencies that AI can address. Valuable questions in this first phase include: Why do we want to adopt an AI tool? What problem do we solve by using it? Is the problem a simple or a complex one?
Generating and evaluating alternatives. In this second phase, the organisation is called to create a four-quadrant matrix in which benefits, enablers, barriers and concerns concerning AI adoption are listed clearly and exhaustively. This will facilitate the evaluation of the most suitable AI tool(s) among the many alternatives and, according to the problem typology previously identified, be capable of enhancing KM processes and improving knowledge flow within the organisation. In this vein, organisations have to recognise that the use of technology across the organisation varies considerably according to the tasks different knowledge workers perform (Davenport, 2011). Moreover, integrating AI into KM processes necessitates precise governance mechanisms to ensure the ethical and responsible use of AI technologies (Chin et al., 2024). This may involve establishing AI supervision committees and implementing robust data governance frameworks. In addition, organisations should focus on talent management initiatives to cultivate a new knowledge-based workforce (Kudyba et al., 2020) equipped with the necessary skills to leverage AI tools effectively (Kudyba and Cruz, 2023). This may entail reskilling and upskilling programmes to enhance digital literacy and promote a culture of continuous learning. Valuable questions in this phase include: Are data (in terms of sources, quality, and quantity) available? Do we have sufficient financial resources? Are employees skilled for this/these specific tool/s? Have we established clear protocols (for example, to handle ethical or regulatory concerns) for adapting the AI tools as needed over time?
Choosing an AI tool. According to Steps 1 and 2, it will be possible to select the “best” AI tool(s) for the organisation's KM processes’ goals. In this phase, the initially identified problems need to be more detailed, explicitly considering the KM processes in which the organisation intends to introduce AI and the type of problem the organisation is addressing. Strategic imperatives stemming from this understanding include prioritising resource allocation towards AI capabilities that enhance decision-making effectiveness and KM efficiency. This may involve investing in AI technologies that support data integration, analytics and natural language processing to facilitate KM processes and decision support. Helpful questions include: What specific KM goals do I want to achieve? Can the identified AI tool(s) be adopted in multiple KM processes? If yes, how?
Implementing AI tool(s). Once the best AI tool(s) is identified, setting up a pilot project to try the tool(s) on a small scale before making more significant investments is useful. Once trained, the AI tools can be put in place without forgetting that they require constant monitoring because “everything changes over time” (e.g. business requirements, technology capabilities, data). Moreover, many AI tools can self-learn, which may require several modifications/adaptations at both organisational and operational levels. In this vein, the interconnectedness between AI, KM processes and decision-making suggests a need for flexible and adaptive organisational structures and processes. Traditional hierarchical structures may inhibit the flow of knowledge and impede decision-making agility. Instead, organisations may benefit from flatter structures that promote cross-functional collaboration and knowledge sharing. Valuable questions in this phase include: How often and in what ways (i.e. identify specific metrics or KPIs) do we monitor AI tools’ effectiveness and efficiency post-implementation? What are processes in place to detect and respond to issues or anomalies in AI tool outputs? How will implementing AI tools impact our existing KM processes and infrastructure?
Evaluating decision effectiveness. In this phase, the decisions taken are assessed to understand if they could solve the identified problem(s) as expected. The results of this evaluation will have repercussions on all the previous steps, modifying them if necessary. Organisations should adopt a strategic approach aligning with their AI adoption roadmap and organisational objectives. This may involve balancing investments in foundational AI infrastructure with targeted solutions that address specific KM and decision-making challenges. Valuable questions in this phase include: Was the decision correct/helpful in solving the identified problem? What worked well/poorly and why? Was the problem clearly defined? How did the use of AI impact our KM processes and decision-making outcomes? What lessons can be learned to improve future KM practices and AI integration?
6. Conclusions
This study simultaneously addresses AI, KM processes and decision-making to provide an original model of the relationships among them, allowing organisations to understand better how to ameliorate their decision-making through AI adoption within KM processes.
Despite theoretical advancements and practical implications provided by the investigation, it is also possible to derive some limitations of this work, which also opens the doors to future studies. The interviewees were chosen for reasons of appropriateness rather than representativeness. However, the variety of investigated realities made it reasonable to believe that they are a representative sample of companies of our time, making findings sufficiently generalisable. By using structured case studies or ethnographic research, future studies could offer deeper insights into the nuanced dynamics of AI integration in KM processes and decision-making within specific organisational contexts. Moreover, as most organisations analysed are private, future studies could verify the model’s validity by exclusively focusing on the public sector. In addition, longitudinal studies could investigate the long-term impact of AI adoption within KM processes on organisational performance and decision outcomes. Furthermore, comparative studies across different industries and regions could provide valuable insights into the contextual factors influencing AI adoption and its implications for decision-making. Another limitation lies in the qualitative nature of our research. Although the chosen interviewees provided valuable insights, their selection prioritised appropriateness over statistical representativeness. Future studies could address this limitation by employing quantitative methods. In this vein, a survey-based approach could be implemented to measure the broader impact of AI on KM processes and decision-making across a larger, more representative sample.
In addition, the framework identifies three decision-making typologies (automated, augmented, supported) based on the problem complexity and the role of AI. In this vein, future research must delve deeper into understanding how the human–AI interplay influences decision outcomes. Understanding these nuances is essential for designing effective decision support systems and optimising human–AI collaboration (Di Vaio et al., 2022). Finally, due to the complexity of the phenomena under investigation, we call for future research to adopt a collaborative and interdisciplinary perspective to understand better how people (within and outside organisations) can use and should be educated in using AI tools. In this vein, quantitative analyses could be instrumental.
To conclude, adopting AI is crucial for KM processes within organisations and can help improve their decision-making. At the same time, the phenomenon deserves further investigation, and researchers and practitioners are called to “keep their guard up” in the face of excessive enthusiasm and optimism in adopting AI-related practices.
The authors gratefully acknowledge all the experts for their time, devotion and precious contributions to this study. Furthermore, the authors are grateful to Prof. Eric Tsui for the precious advice provided on an earlier version of this work and the reviewers for the precious suggestions provided to ameliorate the overall quality of our investigation.
Authors’ contribution: Conceptualisation: All authors; Data curation and investigation: Luna Leoni and Ginetta Gueli; Formal analysis: Luna Leoni, Ginetta Gueli and Marco Ardolino; Methodology: Luna Leoni and Marco Ardolino; Validation: Mateus Panizzon and Shivam Gupta; Writing – original draft preparation: Luna Leoni and Marco Ardolino; Writing – review and editing: All authors.
