Artificial intelligence (AI) is reshaping the operation of public organizations, yet many still struggle to harness its full potential amid institutional complexity and digital transitioning. Characteristically different from the private sector, the public sector requires the development and testing of specific models. However, most studies on the impact of AI have focused primarily on the private sector, leaving public organizations largely underinvestigated. This study examines how public sector agencies develop AI capacity and how this contributes to performance through key internal mechanisms, including organizational creativity and AI management. Using the resource-based view as a guiding framework, a structural model was developed to capture the effects of government mandates, citizen expectations, regulatory clarity, and public incentives on AI capacity and its organizational outcomes. Survey data from 225 senior managers employed in Vietnamese public organizations were analyzed using partial least squares–structural equation modeling. The findings illustrate that environmental context plays a crucial role in enhancing AI capacity, which, in turn, impacts both creativity and performance. However, AI management is not yet consistently translated into AI-driven decision-making, suggesting that public organizations are still in the early stages of embedding AI into their strategic processes. The study demonstrates that the performance benefits of AI are contingent on a comprehensive approach. Further, these gains are a function of not only technology but also key complementary factors, such as cultural readiness, managerial alignment, and institutional support. The study provides practical recommendations for leaders in emerging economies, outlining ways to strengthen governance and address complex challenges via more effective integration of AI into organizational operations.
Recent years have seen an acceleration in the digital transformation efforts of public institutions, with artificial intelligence (AI) playing an increasingly significant role (Legner et al., 2017). Governments worldwide have already adopted AI for administrative tasks to enhance both service delivery and operational efficiency (Misuraca et al., 2020). This is accompanied by a growing interest in the potential of AI to aid organizations in adapting to a constantly changing environment (Ayinaddis, 2025). However, these initiatives have produced inconsistent results. Some research efforts have suggested that AI is more effective when supported by visible leadership and adequate resources. When these conditions are met, AI has the potential to improve decision-making, encourage innovation, and strengthen overall institutional functions (Fosso Wamba et al., 2024; Mikalef & Gupta, 2021).
Early research efforts on AI adoption have focused on its economic and technological potential, but more recent studies have shifted their attention to institutional, ethical, and social dynamics. For instance, Obreja et al. (2024) investigated the influence of public awareness and cultural norms on AI adoption in public agencies. Ayinaddis (2025) highlighted that in resource-constrained settings, the success of AI projects often depends on environmental support and leadership commitment. This has been echoed by Madzík et al. (2024), who argued that organizations that lack strong learning processes struggle to achieve meaningful results from digital tools. Further, Lin et al. (2025) emphasized that aligning AI initiatives with existing governance structures increases the likelihood of success. This body of research thus demonstrates that an organization’s internal capabilities and external conditions are both critical for the effective use of AI.
While both public and private organizations have started adopting AI, public agencies often face unique implementation challenges (Berryhill et al., 2019). This is due in part to the stricter regulations and more rigid internal systems that govern the public sector, which can complicate and hinder the process (De Sousa et al., 2019). Despite frequent discussions on the potential of AI, insights into its connection to tangible outcomes, such as creativity and decision-making, in public institutions are significantly lacking (Campion et al., 2022). Furthermore, the specific ways in which governments use these tools have not been thoroughly explored (Jöhnk et al., 2021). These research gaps highlight the need for a deeper investigation of the mechanisms by which AI can effectively support innovation and improve effectiveness in the public sector.
This work employs the resource-based view (RBV) as its theoretical foundation. In this framework, critical resources, including both physical and intangible assets, are vital for developing AI capacity in organizations (Mikalef et al., 2023). Such capacity is considered a dynamic resource that helps public agencies enhance decision-making, stimulate creativity, and drive innovation. Nevertheless, the RBV has been criticized for its limited consideration of the external forces that enhance the value and deployment of organizational resources. To overcome this limitation, the present study also draws on the technology–organization–environment (TOE) framework, which underscores the role of technological readiness, organizational structures, and environmental conditions in promoting the adoption of technology (Ayinaddis, 2025; Gupta et al., 2022). Integrating the RBV with the TOE framework provides a more comprehensive foundation for analyzing how internal resources interact with contextual factors, including regulatory support, citizen demands, and government incentives, to strengthen the development and effectiveness of AI capacity in the public sector.
Vietnam represents an ideal case for studying AI implementation in government, given its national initiatives to drive digital transformation and integrate AI into public services. However, the public sector faces well-documented structural and institutional hurdles, including limited infrastructure and a rigid regulatory environment; such challenges mirror those present in digital transformation projects globally (Al Halbusi et al., 2025; Atobishi et al., 2024; Fosso Wamba et al., 2024; Neiroukh et al., 2024). Given that cultural and institutional characteristics shape the dynamic interaction between organizational capacity, the environment, and innovation (Sahoo et al., 2024), a deeper investigation is needed. Nevertheless, empirical studies on the development of AI capacity in Vietnam’s public sector are limited.
This study explores the complex relationship between AI capacity and environmental context, along with their joint effect on organizational creativity, decision-making, and performance in the public sector. As a secondary goal, the study aims to identify and address the primary challenges that hinder the successful adoption of AI, thereby providing actionable insights for public sector leaders. The findings of this work contribute both theoretically and practically to the development and implementation of AI strategies in the public sector, ultimately supporting the creation of a more professional, modern, efficient, and effective public administration. To achieve these objectives, the following research questions were posed:
RQ1 To what extent does environmental context influence the development of AI capacity and its associated internal mechanisms in public organizations?
RQ2 How do AI capacity, organizational creativity, AI management, and AI-driven decision-making directly and indirectly affect organizational performance?
The RBV, originally proposed by Wernerfelt (1984) and later expanded by Barney (1991), posits that an organization’s internal resources and capacity are critical for achieving and sustaining superior performance. Resources that are strategically valuable, difficult to mimic, and not easily substitutable create a foundation for attaining a long-term competitive advantage.
In the context of AI adoption and digital transformation, the RBV offers a robust perspective for elucidating how AI capacity can serve as a strategic organizational resource (Neiroukh et al., 2024). AI capacity encompasses a combination of tangible resources (data and technology infrastructure), human skills, and intangible organizational competencies that together enhance innovation, decision-making, and operational effectiveness (Fosso Wamba et al., 2024; Mikalef & Gupta, 2021). By employing advanced data analytics and AI-driven processes, organizations can leverage these resources to foster creativity, accelerate the decision-making process, and improve service quality (Do et al., 2022; Neiroukh et al., 2024).
According to previous research, organizational capacity often acts as a link between internal resources and performance results (Belhadi et al., 2024; Lou & Wu, 2021). In this regard, AI capacity is conceptualized as a dynamic bundle of resources that enables organizations to respond effectively to environmental demands, promote innovation, and improve organizational performance (Chatterjee et al., 2021; Mikalef et al., 2023). Moreover, by combining the RBV and dynamic capability perspective, recent studies have emphasized the need to continuously develop AI-related competencies to sustain adaptability and strategic advantages amid continuous environmental change (Almheiri et al., 2024; Fosso Wamba et al., 2024; Krakowski et al., 2023).
While recent studies have highlighted the importance of revisiting the RBV to better address the challenges of AI-driven and dynamic environments (Helfat et al., 2023; Krakowski et al., 2023; Neiroukh et al., 2024), the framework remains appropriate for this study. The RBV provides a solid theoretical basis for examining how AI capacity, when strategically developed, can function as a critical resource to foster innovation, support effective decision-making, and enhance organizational performance in the public sector (Eng’airo, 2024). Guided by this perspective, this study examines how AI capacity contributes to creativity, decision quality, and overall performance in public organizations.
At the same time, the RBV has been criticized for its limited focus on external pressures. Because resource value depends on both internal development and contextual conditions, it is important to consider complementary perspectives. The TOE framework highlights how technological readiness, organizational factors, and environmental conditions enhance technology adoption and capacity development (Gupta et al., 2022). By integrating the TOE framework with the RBV, this study accounts for both the internal strategic role of AI capacity and external forces, such as regulatory support, citizen pressure, and government incentives, that influence its development (Ayinaddis, 2025). This combined perspective provides a more comprehensive theoretical foundation for understanding how organizational resources interact with environmental contexts to influence creativity, decision-making, and performance in the public sector.
AI capacityAI capacity refers to the ability of a business to select, establish, and use AI resources to help achieve its strategic and operational goals (Mikalef & Gupta, 2021). It represents a composite construct that comprises multiple resource dimensions, including tangible resources, such as data assets and AI-related technological infrastructure; intangible resources, including organizational culture, agility, and risk orientation; and human resources, such as the technical expertise and managerial competencies necessary to develop and implement AI solutions (Al Halbusi et al., 2025; Mikalef & Gupta, 2021).
As a high-order and dynamic resource, AI capacity enables firms to process large volumes of data, combine knowledge from different sources, and automate complex analytical tasks (Chen et al., 2022; Mikalef & Gupta, 2021). Further, it is embedded in organizational routines and shaped by strategic intent, technological maturity, and cultural readiness for digital transformation. The term refers to not only the technical ability to deploy AI tools but also the organizational readiness to align AI use with decision-making structures and evolving business goals (Ransbotham et al., 2017).
Recent research efforts have noted that AI capacity is a crucial component of digital competence, which is essential in dynamic environments that require both flexibility and evidence-based decision-making (Sahoo et al., 2024). It also captures how firms create value from big data by learning from experience, testing new approaches, and making informed adjustments to their internal processes (Fosso Wamba et al., 2024). Thus, AI capacity is considered a strategic asset that combines technology, people, and organizational systems to enable data-informed practices and support lasting transformation (Al Halbusi et al., 2025; Mikalef & Gupta, 2021).
Organizational creativityOrganizational creativity reflects a firm’s ability to develop innovative and valuable ideas, processes, products, and services by drawing on and combining internal and external knowledge (Ruvio et al., 2014; Shoham et al., 2012). Moreover, creativity constitutes an important driver of innovation to gain a competitive advantage (Hult et al., 2004). It is closely linked to a firm’s ability to absorb and use external knowledge (Tsai, 2001) and leverage human expertise and creativity to develop innovative solutions (Kreitner et al., 2001).
Both cultural factors that foster the production of new ideas (Story et al., 2015) and behaviors that support the practical use of such ideas affect organizational creativity. Furthermore, organizational creativity depends on collaboration among different parts of an organization (Alexiev et al., 2016) and the development of human talent and social networks (Donate et al., 2016). In the context of AI-driven transformation, studies have indicated that AI capacity can significantly enhance creativity by facilitating knowledge integration, supporting the generation of ideas, and enabling the recombination of dynamic resources (Al Halbusi et al., 2025; Fosso Wamba et al., 2024; Neiroukh et al., 2024; Sahoo et al., 2024). From this perspective, the study views organizational creativity as a combination of cultural orientation and adaptive capability that supports innovation and improves organizational performance.
Organizational performanceOrganizational performance in public services denotes the extent to which public agencies achieve their goals, deliver value for the public, and meet citizen expectations (Neiroukh et al., 2024). It includes dimensions such as operational efficiency, goal attainment, service quality, and citizen satisfaction (Fosso Wamba et al., 2024; Marchiori et al., 2022; Park & Lee, 2020). Specifically, public organizations are faced with critical challenges that affect their ability to maintain a high level of performance. Regulatory complexity, financial constraints, and heightened public accountability often limit their capacity to invest in advanced technologies, infrastructure, and human capital, all of which are critical for enhancing service delivery and fostering innovation (Atobishi et al., 2024). In this context, AI capacity offers new opportunities for improving public sector performance by enhancing decision-making, supporting innovation, and enabling more responsive and efficient services (Al Halbusi et al., 2025; Fosso Wamba et al., 2024; Neiroukh et al., 2024). Therefore, assessing the organizational performance of public services requires a multidimensional and contextual approach that reflects the complex objectives of creating value for the public, ensuring citizen satisfaction, and achieving operational excellence.
Development of hypothesesThe RBV forms the theoretical foundation of our hypotheses. In line with this perspective, AI capacity is regarded as a strategically significant organizational resource that enables value creation and supports long-term performance improvements in public sector organizations (Al Halbusi et al., 2025; Mikalef & Gupta, 2021). We proposed a research model that builds on this perspective by exploring the impact of environmental factors on the development of AI capacity and how this capacity, in turn, contributes to key organizational outcomes through related mechanisms, including creativity, AI management, and AI-supported decision-making (Al Halbusi et al., 2025; Fosso Wamba et al., 2024).
Environmental context plays an important role in fostering the development of AI capacity, particularly in the public sector, where institutional pressure and external support shape strategic priorities and resource allocation (Neiroukh et al., 2024). Similar challenges and enabling factors have been documented in public sector digital transformation efforts globally (Al Halbusi et al., 2025; Atobishi et al., 2024; Fosso Wamba et al., 2024). Government policies, funding mechanisms, regulatory flexibility, and collaborative ecosystems are the main components of business environmental support that can facilitate AI adoption and capacity development (Arroyabe et al., 2024; Sahoo et al., 2024). In addition, increasing social expectations and public demand for greater transparency, service quality, and innovation further motivate public agencies to pursue AI-enabled transformation (Corvello, 2025; Schaefer et al., 2021).
Although the positive influence of the environmental context on technology adoption has been well established, its specific impact on the dynamic development of AI capacity in public sector organizations is underexplored (Neiroukh et al., 2024; Sahoo et al., 2024). The unique institutional and cultural characteristics of public agencies might determine how environmental facilitators influence the creation of AI capacity (Corvello, 2025; Fosso Wamba et al., 2024; Obreja et al., 2024). Therefore, an empirical study was conducted to elucidate this relationship in the context of Vietnam’s public sector, thereby leading to a better understanding of how external pressures and support mechanisms facilitate the development of AI capacity. Considering these insights, the first hypothesis was proposed.
Hypothesis 1 The environmental context has a positive impact on AI capacity.
AI is widely acknowledged as a driver of organizational creativity (Fosso Wamba et al., 2024; Sahoo et al., 2024). Employees can use AI to explore strategic thinking and generate new ideas for performing routine activities (Raisch & Krakowski, 2021). Additionally, data-driven decision support enables managers and staff to recognize patterns and relationships that can stimulate innovative thinking and problem-solving (Corvello, 2025).
While evidence from various sectors has illustrated that AI can stimulate organizational creativity (Amabile, 2020), many of these efforts have a narrow focus and are largely exploratory. As organizations increasingly move toward the wider-scale adoption of AI, its potential to support creative processes in a more structured and consistent way has become more apparent (Fosso Wamba et al., 2024; Sahoo et al., 2024). The role of AI in creativity has attracted growing attention, but few studies have investigated the effect of AI capacity on creativity in the public sector. Considering the distinct structural and functional characteristics of the public sector, one needs to examine whether AI capacity makes a meaningful contribution to creative outcomes in this specific context. Based on this reasoning, the second hypothesis was formulated.
Hypothesis 2 AI capacity has a positive effect on organizational creativity.
Organizational creativity is based on internal and external knowledge as well as organizational culture and behavior (Story et al., 2015), together impacting the creation and implementation of ideas (Alexiev et al., 2016). In public organizations, creativity plays a central role in addressing complex service challenges by enabling the development of innovative solutions that improve quality and generate greater value for the public (Al Halbusi et al., 2025; Fosso Wamba et al., 2024; Sahoo et al., 2024). The contribution of creativity to performance has been widely explored in the private sector, though empirical investigations into how it functions as a performance driver in the public sector are limited (Neiroukh et al., 2024). This study fills that gap by examining the extent to which creative capacity contributes to performance improvement in government organizations. This led to the development of the third hypothesis.
Hypothesis 3 Organizational creativity has a positive effect on organizational performance.
AI capacity is increasingly being considered a strategic driver of organizational performance and can improve operational efficiency, promote more data-driven decision-making, and lead to innovation (Fosso Wamba et al., 2024; Mikalef & Gupta, 2021; Neiroukh et al., 2024). Agencies in the public sector often operate with institutional complexity and rising service demands, and AI offers tools for improving responsiveness and creating greater value for the public (Sahoo et al., 2024). Recent studies have emphasized the significance of building organizational AI capabilities to achieve sustained performance gains, rather than limiting efforts to standalone projects (Almheiri et al., 2024; Mikalef et al., 2023). Research has also revealed that strategic investment in AI can raise productivity and citizen satisfaction (Aljuhmani et al., 2024). Specifically, public organizations with a well-developed AI capacity can enhance productivity, foster innovation, and produce higher value for the public (Fosso Wamba et al., 2024; Neiroukh et al., 2024). However, despite growing interest in this regard, empirical examinations of these relationships in public organizations are few. Based on this reasoning, we proposed the fourth hypothesis.
Hypothesis 4 AI capacity has a positive effect on organizational performance.
AI management consists of the organizational systems, processes, and practices that support the deployment, coordination, and integration of AI technologies with regard to organizational processes (Bag et al., 2022; Łapińska et al., 2021). The effectiveness of these practices plays an important role in converting AI investment into meaningful outcomes, including scalability and strategic alignment (Bag et al., 2022). Organizations with mature AI capabilities are often better equipped to institutionalize AI governance by harnessing advanced technologies, promoting cross-functional collaboration, and embedding AI into core workflows (Metawa et al., 2022). Furthermore, a strong AI capacity can optimize administration, enabling more informed and responsive decision-making (Bag et al., 2021).
While the conceptual relationship between AI capacity and AI management is recognized, minimal empirical research has been conducted on their interaction. Because AI capacity reflects an evolving organizational resource that influences technological adoption and governance, we investigated whether higher capacity leads to more effective management. This question is notably relevant in public sector settings, where robust AI management is essential for maintaining transparency, ensuring accountability, and maximizing value for the public. Considering the above, we proposed the fifth hypothesis.
Hypothesis 5 AI capacity has a positive effect on AI management.
The rapid advancement of digital technologies requires organizations to establish robust AI management systems for the effective monitoring, updating, and governance of AI applications (Chen et al., 2022). AI management encompasses not only technical components, such as data structures, software systems, and technology infrastructure, but also AI-related organizational dimensions, including human resource management, ethical guidelines, and decision-making (Dwivedi et al., 2021). Effective AI management practices enable public agencies to enhance decision-making processes, improve responsiveness, and optimize operational performance (Bughin et al., 2018).
Organizations with a well-established AI management framework are better positioned to translate the technological potential of AI into tangible performance outcomes. Without such AI-specific frameworks, many organizations risk remaining at the stage of isolated experiments or pilot projects (Ransbotham et al., 2018). While the conceptual link between AI management and organizational performance is widely acknowledged, few empirical studies have explored this relationship in the public sector. Public organizations face distinct challenges, including regulatory requirements, public accountability, and complex stakeholder environments. Therefore, we examined whether effective AI management contributes to performance improvement in this context, leading to the sixth hypothesis.
Hypothesis 6 AI management has a positive effect on organizational performance.
Integrating AI into management systems in the public sector has the potential to significantly improve the quality of data-driven decision-making (Omoga, 2023). Effective AI management enhances the availability, accuracy, and timely generation of information, enabling public organizations to make more informed and responsive decisions. Improved decision-making processes contribute to better governance, reduce the likelihood of error, and promote greater transparency and accountability (Eng’airo, 2024). By strengthening and expanding the capacity to govern, integrate, and utilize AI systems, AI management enables public sector organizations to serve communities more effectively and adapt to evolving demands.
Although this relationship is conceptually well supported, empirical evidence of the direct influence of AI management on AI-driven decision-making in public organizations is still limited. Therefore, we investigated whether improved AI management practices systematically enhance the quality and effectiveness of AI-driven decision-making processes in this context, proposing the seventh hypothesis.
Hypothesis 7 AI management has a positive impact on AI-driven decision-making.
AI-driven decision-making enables organizations to process complex data, generate actionable insights, and raise the quality and rate of decision-making. In the public sector, improved decision-making can lead to more effective service delivery, greater operational efficiency, and higher value for the public (Omoga, 2023). By fostering transparency and accuracy in decision-making processes, AI-driven decision-making can strengthen organizational performance outcomes.
Although the potential benefits of AI-driven decision-making are widely acknowledged, little systematic empirical evidence exists regarding its direct impact on organizational performance in public organizations. Given the critical role of decision quality in shaping public sector outcomes, we tested this relationship empirically, developing our eighth hypothesis.
Hypothesis 8 AI-driven decision-making has a positive effect on organizational performance.
Fig. 1 illustrates the conceptual model developed in this study, showing the eight hypotheses that link environmental context, AI capacity, organizational creativity, AI management, AI-driven decision-making, and organizational performance. Building on both the RBV and TOE framework, and supported by recent literature, the model reveals that AI capacity functions as a dynamic organizational resource that enables innovation, enhances decision quality, and drives performance outcomes in the public sector. This framework forms the basis of our empirical analysis.
Research methodologyData collection
The target population comprised public sector agencies across several provinces in Vietnam. Access to these organizations was established via official government directories, administrative records, and professional networks, which provided reliable contact information for eligible agencies. Senior managers were formally approached through institutional invitations, followed by email correspondence to confirm their participation. This procedure was particularly important within the Vietnamese context, where access to public officials requires formal authorization and adherence to administrative protocols.
To ensure a homogenous sample, only individuals in comparable managerial roles, such as members of the board of directors and department heads, were included, as they share similar responsibilities in regard to organizational policies, operations, and the oversight of digital initiatives. A purposive sampling method was employed to deliberately target these respondents, given their knowledge and decision-making authority, ensuring accurate insights into organizational activities and AI-related practices. This approach enhanced the comparability of responses across agencies and reduced potential bias from differences in rank or responsibility.
The survey respondents were required to read an ethical statement before beginning the online questionnaire. Participation was entirely voluntary, and no incentives were offered. The respondents were informed that they could withdraw from the study at any time, ensuring that they faced no risks and that their responses were fully voluntary. The online questionnaire was configured to prevent the submission of incomplete responses, ensuring that all submitted surveys were suitable for analysis.
Between January and March 2025, 289 responses were submitted; of these, 225 were retained after data cleaning, resulting in a final valid response rate of 78.2 percent. With respect to the respondents’ positions, 38.67 percent of the sample comprised members of the board of directors, followed by managers of technical departments (34.22 percent) and managers of information technology departments (27.11 percent). Most respondents had 6 to 10 years (32 percent) or 11 to 15 years (29 percent) of work experience. Regarding their education level, 48 percent had a bachelor’s degree, 48 percent had a master’s degree, and 4 percent had a doctorate.
Most of the surveyed organizations were small or medium-sized: 53.33 percent employed between 10 and 49 staff members, and 39.56 percent employed between 50 and 249. Only 7.11 percent of the organizations had >249 employees. In terms of AI adoption, 46.22 percent of the management-level respondents reported that their organizations had been using AI for 1 to 2 years, 30.22 percent reported <1 year of use, and 23.56 percent reported >2 years of AI use. Table 1 summarizes the demographic characteristics of the sample.
Demographic characteristics.
Source: Compiled by the authors.
Questionnaires constitute a well-established technique for collecting primary data from a large number of respondents. In this study, we designed a structured and closed-ended questionnaire for data collection, with the measurement details presented in Table 2. The questionnaire was divided into two sections: the first collected demographic information about the respondents’ organizations (see Table 1), and the second included items related to the conceptual framework of the study. All constructs were measured using a five-point Likert scale, from 1 (strongly disagree) to 5 (strongly agree).
Measurement items.
Source: Compiled by the authors.
The measurement items were adapted from validated scales in previous research to ensure content validity. AI capacity is a second-order construct, adapted from Chen et al. (2022), with three components: AI basic (AIB), assessed with five items that measure hardware, software, data resources, and basic technical capacity; AI skills (AIS), measured with five items to reflect staff competencies and organizational readiness for AI use; and AI proclivity (AIP), measured with five items that capture the organization’s strategic orientation and innovation posture toward AI.
The environmental context is also a second-order construct, adapted from Mikalef et al. (2022), and includes four components: perceived government pressure (PGP), with two items; perceived citizen pressure (PCP), with three items; government incentives (GI), with three items; and regulatory support (RS), with four items.
AI management was measured with three items adapted from the study by Chen et al. (2022), reflecting the organization’s systems and processes for managing AI technologies. Consequently, AI-driven decision-making was assessed with four items from the same source, capturing the extent to which AI informs operational and policy decisions.
Organizational creativity was measured with five items adapted from the work of Mikalef and Gupta (2021), evaluating the organization’s support for, and engagement in, creative and innovative practices. Finally, organizational performance was measured with 11 items adapted from Marchiori et al. (2022), thus capturing multiple dimensions of performance, including service quality, operational efficiency, transparency, and citizen satisfaction.
As mentioned earlier, the study employed measurement items from established scales in prior research; these were refined to ensure their validity and suitability for the public sector context. The wording of the items marked as “major modifications” was adjusted to reflect the terminology and institutional features of the Vietnamese public sector while retaining the meaning of the original construct. To ensure methodological rigor, we implemented a structured adaptation process. The items were first translated into Vietnamese and then back-translated into English by bilingual experts to ensure conceptual consistency. The revised wording was subsequently examined by two academic scholars in public management and five public sector managers, who assessed the clarity, contextual fit, and theoretical alignment of each item. A pilot study involving 40 managers from public organizations was then conducted, demonstrating that the modified items were clearly understood and yielded reliable results. Using these steps, the adapted scales displayed both contextual relevance and methodological soundness. Table 2 presents a summary of the measurement items.
Analytical methodologyTo confirm the validity and reliability of the proposed research model, we employed partial least squares–structural equation modeling (PLS-SEM). This method was considered appropriate for this study because it enables the simultaneous estimation of multiple relationships between one or more independent variables and one or more dependent variables (Akter et al., 2017; Hair et al., 2022). Consequently, PLS-SEM supported the comprehensive modeling of complex relationships within the conceptual framework of this study.
PLS-SEM offers several advantages over other structural equation models. First, it provides flexibility with regard to assumptions about multivariate normality; second, it accommodates the use of both reflective and formative measurement models; third, it enables the estimation of complex models with smaller sample sizes; and finally, it functions as an effective predictive tool for theory building (Nair et al., 2018). In addition, PLS-SEM facilitates the calculation of indirect and total effects, enabling the simultaneous assessment of relationships among multiple constructs while minimizing the overall model error (Akter et al., 2017; Astrachan et al., 2014).
ResultsCommon method biasBecause the study employed a questionnaire to collect self-reported data on both exogenous and endogenous variables, common method bias (CMB) was a possibility and might have affected the results. To minimize this risk, we used several procedural remedies before data collection. The respondents were explicitly informed that there were no objectively correct or incorrect answers and were assured that their responses would remain confidential, in accordance with the guidelines of Podsakoff et al. (2003). Additionally, we assessed the full collinearity variance inflation factor (FCVIF), with values below 3.3 indicating the absence of CMB (Kock, 2015).
Validity and reliabilityAs shown in Table 3, the FCVIF is below the threshold of 3.3 for all constructs, indicating the irrelevance of CMB in this study.
Reliability of the constructs and factor loadings of indicators.
Notes: Factor loading (FL), Cronbach’s alpha (Alpha), composite reliability (CR), average variance extracted (AVE), full collinearity variance inflation factor (FCVIF). * Deleted from the model, given that it did not meet the required criteria or enhance the convergent validity of the construct.
Source: Compiled by the authors.
In Table 3, the factor loadings range from 0.648 to 0.918, all exceeding the recommended threshold of 0.50 (Hair et al., 2022), indicating adequate item reliability. Cronbach’s alpha is 0.680 for PGP, which is considered acceptable, and above 0.70 for all other constructs, demonstrating good internal consistency. The composite reliability (CR) of all constructs also exceeds 0.70, further supporting the reliability of the measurement model. Item AIS4 was deleted from the model because it neither meets the required criteria nor enhances the convergent validity of the construct. These results indicate the strong internal consistency and reliability of the measurement model. Moreover, the average variance extracted (AVE) is greater than 0.50 for all constructs, confirming convergent validity.
The model’s fit was assessed using the standardized root mean square residual (SRMR), with a value of 0.097. Although slightly higher than the often-cited cutoff of 0.08, it is below 0.10 and can still be considered acceptable for complex models (Hair et al., 2022). The predictive relevance was evaluated using Stone–Geisser’s Q² values, obtained through blindfolding. Most constructs show predictive relevance (Q² > 0), with high values for PCP (0.524), PGP (0.492), AIS (0.445), AIB (0.419), and GI (0.407). Moderate predictive relevance was observed for AIP (0.323) and RS (0.260), while organizational performance shows a small-to-moderate predictive relevance (0.179). Only AI management (0.051), organizational creativity (0.020), and AI-driven decision-making (0.012) display low values. Overall, the SRMR (0.097) and Q² values indicate that the model exhibits an acceptable but modest fit and predictive relevance, consistent with recommendations for complex PLS-SEM models (Hair et al., 2022).
Further, we assessed the discriminant validity using the Fornell–Larcker criterion and the heterotrait–monotrait ratio (HTMT). In Table 4, the square roots of AVE, located on the diagonal, are higher than the correlations between constructs, satisfying the Fornell–Larcker criterion. In addition, researchers have recommended HTMT evaluations for assessing discriminant validity (Truong & Nguyen, 2024). According to Henseler et al. (2015), an HTMT below 0.90 indicates the establishment of discriminant validity. With the exception of PGP, the HTMT of which slightly exceeds 0.90, the HTMT values are below this threshold. Overall, these results confirm that the first-order constructs in the research model display adequate discriminant validity.
Discriminant validity.
Notes: The square root of the AVEs is in boldface on the diagonal. Artificial intelligence basic (AIB), artificial intelligence-driven decision-making (AIDDM), artificial intelligence management (AIM), artificial intelligence proclivity (AIP), artificial intelligence skills (AIS), government incentives (GI), organizational creativity (OC), organizational performance (OP), perceived citizen pressure (PCP), perceived government pressure (PGP), regulatory support (RS).
Source: Compiled by the authors.
We first evaluated the measurement model using the PLS-SEM algorithm and then assessed the structural model to test the proposed hypotheses. Bootstrapping with 5000 subsamples ensured the robustness of the estimates, in line with established guidelines for PLS-SEM analysis (Hair et al., 2022). Table 5 presents the results of the structural model analysis.
Path analysis.
Notes: Environmental context (EC), AI capacity (AIC), government incentives (GI), organizational creativity (OC), organizational performance (OP), artificial intelligence-driven decision-making (AIDDM).
Source: Compiled by the authors.
To assess the model’s explanatory power, we examined the R² values of the endogenous variables: 0.598 (AI capacity), 0.024 (AI-driven decision-making), 0.071 (AI management), 0.051 (organizational creativity), and 0.335 (organizational performance). In particular, the R² values for AI capacity (0.598) and organizational performance (0.335) exceed the threshold of 0.26 recommended by Cohen (1988), indicating their substantial explanatory power. By contrast, the relatively low R² values for AI management (0.071) and AI-driven decision-making (0.024) may be interpreted as possible evidence of early-stage AI maturity, reflecting limited organizational experience and institutional development in managing and applying AI within the public sector.
Table 5 reveals that the direct effects of organizational creativity (β = 0.257, p < 0.05), AI capacity (β = 0.139, p < 0.05), AI management (β = 0.224, p < 0.05), and AI-driven decision-making (β = 0.237, p < 0.05) on organizational performance are statistically significant. Thus, H3, H4, H6, and H8 are supported. Additionally, the environmental context has a significantly positive effect on AI capacity (β = 0.774, p < 0.001), supporting H1. Furthermore, AI capacity has significantly positive effects on both organizational creativity (β = 0.290, p < 0.05) and AI management (β = 0.561, p < 0.001), thereby supporting H2 and H5. In contrast, the relationship between AI management and AI-driven decision-making is not statistically significant (β = 0.156, p > 0.05), rejecting H7.
Finally, we examined indirect effects to assess the potential mediating role of AI management and organizational creativity in the relationship between AI capacity and organizational performance (see Table 5). Our findings indicate that environmental context, AI management, and organizational creativity have partial mediating effects on this relationship, which contribute to a more nuanced understanding of the influence of AI capacity on organizational performance in the public sector.
DiscussionThis study provides evidence of how environmental conditions shape the development of AI capacity and how this supports creativity, managerial processes, decision-making, and, ultimately, performance in the public sector. The findings reinforce a theoretical approach that combines the RBV with the TOE framework. The RBV highlights the role of internal resources as the foundation of strategic advantages, whereas the TOE framework explains how technological readiness, organizational arrangements, and environmental pressures influence the adoption and effective use of new capabilities (Gupta et al., 2022). These perspectives capture the interaction between resources and context, which is essential for understanding AI adoption in public organizations.
First, the results confirm that the environmental context significantly influences AI capacity (supporting H1). This finding is consistent with previous studies that emphasize the role of institutional pressures and government-led policy directives in driving AI-related transformation, particularly in public sector organizations (Neiroukh et al., 2024; Obreja et al., 2024). In emerging economies such as Vietnam, centrally coordinated initiatives and citizen expectations act as external triggers that encourage digital investment and AI adoption. From an RBV perspective, these external forces can catalyze the formation of internal capacity by creating strategic urgency and prioritizing resources, whereas the TOE viewpoint frames them as environmental conditions that determine how quickly and effectively capacity can be developed (Ayinaddis, 2025).
Second, AI capacity significantly enhances both organizational creativity and AI management (supporting H2 and H5). This dual influence reveals the strategic potential of AI capacity to enable not only operational improvements but also cognitive, cultural, and managerial shifts within organizations. The result aligns with prior findings suggesting that AI capacity facilitates knowledge integration and the exploration of novel service models (Fosso Wamba et al., 2024; Mikalef & Gupta, 2021; Sahoo et al., 2024). Considering the RBV, these effects represent capacity leverage, where tangible and intangible AI-related assets are transformed into value-creating routines, such as innovation and process governance. Moreover, the TOE framework complements this view by highlighting that these organizational outcomes depend on the alignment of technological readiness, internal processes, and external expectations.
Third, the study confirms the positive effects of AI capacity, AI management, AI-driven decision-making, and organizational creativity on organizational performance (supporting H3, H4, H6, and H8). These findings are consistent with prior literature indicating that embedding AI across technical, managerial, and cultural dimensions leads to synergistic benefits in terms of service efficiency, accountability, and user satisfaction (Fosso Wamba et al., 2024). In particular, the strong effect of creativity on performance highlights the importance of fostering an innovation-oriented culture in public agencies. This aligns with the RBV argument that competitive advantage stems not only from acquiring advanced technologies but also from the organization’s ability to configure them effectively in contextually relevant ways. Additionally, the TOE framework emphasizes that such outcomes are conditioned by institutional and technological readiness.
However, the relationship between AI management and AI-driven decision-making is not statistically significant (rejecting H7), which differs from previous findings (Omoga, 2023). A possible explanation is that despite the implementation of AI systems in public sector organizations, they remain in the early stages of technological maturity. Even where such systems exist, they may not yet be trusted in regard to strategic decisions due to a culture of risk aversion and high-level decision-making. This reflects a possible implementation gap, where managerial tools exist but are not fully trusted, implemented, or aligned with the organizational strategy. Similar issues have been noted in other resource-constrained contexts, where AI implementation is hindered by insufficient managerial commitment and readiness, as mentioned by Ayinaddis (2025). Similarly, Lin et al. (2025) stated that the effectiveness of AI depends on not only technical deployment but also institutional trust, leadership alignment, and system integration. Overall, our findings emphasize the need for stronger internal governance, cross-functional alignment, cultural adaptation, and leadership support to fully realize the decision-enhancing value of AI management.
Finally, the mediation analysis reveals that the relationship between AI capacity and organizational performance is partially mediated by both organizational creativity and AI management (supporting the proposed mediation model). This finding emphasizes that the performance benefits of AI capacity do not arise in isolation but are channeled through enabling systems and behavioral capabilities. This is consistent with the study by Madzík et al. (2024), which demonstrates the central role of organizational learning and leadership support in reaping the gains from digital transformation. In the public sector, decision-making frequently involves risk-averse attitudes and is often bureaucratically constrained; therefore, these mediating mechanisms are essential for translating potential into tangible outcomes. This confirms the RBV principle that value creation depends on an organization’s ability to deploy and integrate its resources through complementary routines and processes. Meanwhile, the TOE framework emphasizes the influence of external and organizational conditions in enabling the development and persistence of these mechanisms.
Conclusions, contributions, and limitationsConclusionsThis study advances the understanding of how public sector organizations develop and leverage AI capacity to enhance organizational performance. Grounded in both the RBV and TOE framework, the study demonstrates that through government pressure, citizen expectations, and institutional support, the environmental context strongly influences the development of AI capacity in public sector organizations. In turn, AI capacity not only enables operational improvements but also supports organizational creativity and structured AI management practices, which together contribute to performance outcomes. The results indicate that these effects are not purely technological but are mediated through cultural, behavioral, and managerial processes in the organization. Moreover, our finding that organizational creativity strongly influences performance highlights the importance of fostering an innovation-oriented culture in public administration to complement digital transformation initiatives.
At the same time, the study provides nuanced insights into the current limitations of AI integration in the public sector. The lack of a significant relationship between AI management and AI-driven decision-making suggests that despite the implementation of AI systems, their full potential for supporting strategic decisions has yet to be realized. This demonstrates a persistent gap between AI adoption and organizational readiness for AI-enabled decision-making, reflecting the need for stronger internal alignment and capacity development. More broadly, our findings reinforce the RBV perspective that sustainable performance gains from AI depend not only on technology acquisition but also on an organization’s ability to integrate, align, and mobilize complementary capacities, such as creativity and effective management processes. A comprehensive capacity-building approach is, therefore, essential, and the TOE perspective further clarifies that these internal efforts display an impact only when supported by favorable external conditions that create an enabling environment for digital transformation.
Theoretical contributionsThis study contributes to the theoretical advancement of the RBV by demonstrating how AI capacity operates as a dynamic organizational resource in the public sector. By grounding the analysis in the RBV, we extended the theory beyond its traditional application in private-sector settings to demonstrate its explanatory power in public administration, where institutional complexity, regulatory constraints, and citizen expectations create unique conditions for resource deployment. The findings confirm that the environmental context functions as a second-order construct shaped by four first-order dimensions: pressure by the government, pressure from citizens, government incentives, and regulatory support (Mikalef et al., 2022). Thus, environmental context exerts a decisive influence on the development of AI capacity, supporting recent calls to revisit the RBV in dynamic and institutionally complex environments (Helfat et al., 2023; Krakowski et al., 2023; Neiroukh et al., 2024). Incorporating insights from the TOE framework further clarifies how technological readiness, organizational structures, and external pressures jointly influence the transformation of resources into capacity (Ayinaddis, 2025; Gupta et al., 2022).
Next, the study contributes to a more nuanced conceptualization of AI capacity as a second-order construct comprising foundational (tangible) resources; inclination (intangible) resources; and skills, or human resources (Chen et al., 2022). This multidimensional view is consistent with prior research on dynamic capabilities and digital transformation (Mikalef & Gupta, 2021; Neiroukh et al., 2024), indicating that building AI capacity requires more than technical infrastructure and also depends on managerial orientation and organizational culture. By empirically validating this structure, the study refines the theoretical positioning of AI capacity, presenting it as a dynamic bundle of resources that enables not only efficiency gains but also innovation and adaptability. While earlier research has demonstrated these relationships mainly in business and industrial contexts (Al Halbusi et al., 2025; Sahoo et al., 2024), the present study extends these insights to the underexplored domain of the public sector.
Furthermore, our findings on the mediating roles of organizational creativity and AI management enrich the RBV perspective by showing how resources are transformed into outcomes via complementary mechanisms. While earlier studies have highlighted absorptive capacity and organizational learning as central mediators of the resource–performance link (Do et al., 2022; Madzík et al., 2024), this study demonstrates that creativity and managerial processes are equally critical for leveraging AI-related resources in public organizations. This perspective extends the emerging literature on how organizational routines, cultural readiness, and contextual conditions shape the effectiveness of technological resources (Lin et al., 2025; Obreja et al., 2024).
Moreover, the unsupported relationship between AI management and AI-driven decision-making indicates that AI capacity alone is insufficient for fully ensuring decision-making improvement without supporting factors such as clear governance processes and a data-driven culture (Eng’airo, 2024;Wamba et al., 2015). This result refines the RBV perspective by illustrating that the translation of AI capacity into performance gains requires not only the accumulation of resources but also the effective orchestration of complementary organizational routines. Viewed through the TOE framework, this finding highlights that institutional support, organizational readiness, and cultural alignment are essential for ensuring that managerial practices are effectively embedded in decision-making processes.
Finally, integrating the RBV with the TOE framework represents a significant theoretical contribution. The RBV highlights the role of internal resources, while the TOE framework explains how external conditions, such as regulatory support, citizen expectations, and government incentives, shape their development and impact (Ayinaddis, 2025; Gupta et al., 2022). Combining these perspectives enables a more comprehensive framework for understanding how internal resources and external environments interact to generate public value. This integration extends the applicability of the RBV to the public sector and highlights the need to situate resource-based arguments within broader institutional and environmental contexts.
Practical implicationsOur findings have several practical implications for public sector managers and policymakers seeking to enhance organizational performance through the development of AI capacity.
First, the results address the critical role of the environmental context in driving AI capacity. Policymakers should recognize that government incentives, regulatory frameworks, and citizen expectations are powerful levers for fostering AI adoption and capacity building in public organizations. To accelerate digital transformation, government agencies should actively construct an enabling environment through supportive policies, clear regulations, and well-designed incentive structures that encourage public sector organizations to invest in AI capacity.
Second, public sector managers should adopt a strategic and holistic approach to developing AI capacity, ensuring that it extends beyond technological acquisition to encompass organizational creativity and AI management practices. The study reveals that AI capacity can drive innovation and performance when supported by a culture that encourages creativity and robust governance structures. Therefore, managers should prioritize initiatives that build both technical and human capacities, foster cross-functional collaboration, and embed innovation into organizational routines.
Third, our findings highlight the importance of investing in organizational creativity as a mechanism for achieving performance gains from AI capacity. Public organizations should cultivate an innovation-friendly environment by encouraging experimentation, supporting the generation of ideas, and recognizing creative contributions. Building such a culture will enable public agencies to translate AI capacity into meaningful improvements in service delivery and the creation of greater public value.
Fourth, the unsupported relationship between AI management and AI-driven decision-making highlights the need to further examine decision-making processes. Public sector leaders should focus on strengthening data governance, ensuring that AI-generated insights are integrated into strategic and operational decisions. This might require enhanced training for decision-makers, greater alignment between AI systems and organizational priorities, and stronger emphasis on developing a data-driven culture at all levels of the organization.
Overall, these implications suggest that fully attaining the performance benefits of AI in the public sector requires a multifaceted approach that combines technological development with cultural, managerial, and governance initiatives. By addressing these complementary dimensions, public sector organizations can leverage AI capacity more effectively to achieve strategic objectives and deliver greater value for the public.
Limitations and future researchWhile this study provides valuable insights into the influence of AI capacity on organizational performance in the public sector, several limitations should be acknowledged. The cross-sectional design limited our ability to establish causal relationships between the variables; as in previous studies investigating resources and performance at a single point in time (Do et al., 2022; Madzík et al., 2024), our findings cannot capture how relationships evolve with organizational progress. Future research could adopt longitudinal approaches to explore how the development and impact of AI capacity evolve over time as public organizations progress in their digital transformation.
The focus on public organizations in the Vietnamese context constitutes another limitation. Although Vietnam presents a relevant case of an emerging economy that is advancing digital government initiatives, institutional and cultural factors might shape the operation of AI capacity in different national or regional settings. Comparative studies across different countries and diverse administrative systems would help assess the generalizability of our findings.
In addition, while this study examined AI capacity in relation to organizational creativity, AI management, and AI-driven decision-making, other potentially significant factors were not included in the model. Future research could incorporate additional variables, such as leadership style, organizational learning, and cross-sector collaboration, to more comprehensively elucidate the mechanisms through which AI capacity drives performance.
Furthermore, the reliance on self-reported data from single respondents may have introduced bias. Although the study followed established practices for adapting and validating measurement scales, as highlighted in recent research (Belhadi et al., 2024; Neiroukh et al., 2024), future work could strengthen the robustness of the process by employing multi-method designs that triangulate surveys with archival data or include multiple informants within each organization.
Finally, the study addresses public sector organizations in a general manner rather than in specific domains. Investigating AI capacity in particular sectors, such as health care, education, and public safety, could yield more granular insights into sector-specific dynamics and challenges. Such research efforts would further inform the development of targeted AI strategies in different areas of public administration.
Declaration of generative AI use and AI assistanceThe authors used ChatGPT only for grammar and clarity improvements. All aspects of the research, including study design, data analysis, and interpretation, were conducted entirely by the authors, who take full responsibility for the content. We also employed the services of two independent professional English editing companies to proofread the first version and the final version of the manuscript, respectively.
CRediT authorship contribution statementTruc Thanh Phan: Writing – review & editing, Writing – original draft, Methodology, Formal analysis, Data curation, Conceptualization. Phuong Van Nguyen: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Software, Resources, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Conceptualization. Vincenzo Corvello: Writing – review & editing, Writing – original draft, Validation, Supervision, Methodology, Investigation. Thao Thi Phuong Pham: Writing – original draft, Resources, Formal analysis, Data curation, Conceptualization.
This research is funded by Vietnam National University Ho Chi Minh City (VNU-HCM) under grant number NCM2025-28–01
