Procurement of Artificial Intelligence Systems in UAE Public Sectors: An Interpretive Structural Modeling of Critical Success Factors

Abstract

This study investigates the critical success factors (CSFs) influencing the procurement of artificial intelligence (AI) systems within the United Arab Emirates (UAE) public sector. While AI holds immense potential to enhance public service delivery, its successful integration hinges on critical factors. This research utilizes Interpretive Structural Modeling (ISM) to analyze the CSFs impacting AI procurement within the UAE public sector. Through ISM, a structural model is developed to highlight the interrelationships between these CSFs and their influence on the procurement process, outlining the key elements for successful AI procurement within the UAE public sector. Based on the literature review and expert validation from the UAE public sector, ten CSFs were identified. This study found that clear needs assessment is the most influential CSF, while the long-term value of AI systems or services is the least influential. This study provides policymakers and public sector leaders with valuable insights, enabling them to formulate effective strategies to optimize the procurement process and establish a strong foundation for AI adoption. Finally, this will lead to an improved and more efficient public service delivery in the UAE.

1. Introduction

Our world is rapidly changing due to artificial intelligence (AI), and this transformation is also evident in the public sector. Government operations, service delivery, and citizen–government interactions could all be completely transformed by this potent technology [1]. By embracing AI, public institutions can become more efficient, effective, and responsive to the needs of population [2]. Public sector organizations are often burdened by repetitive tasks and complex data analysis. AI offers solutions through automation and data processing capabilities. For instance, in Seoul, South Korea, chatbots have successfully addressed over 90% of public inquiries, showcasing their ability to manage basic citizen questions and allowing human resources to concentrate on other intricate issues [3]. Machine learning algorithms can analyze large-scale datasets to find patterns and trends, thereby enhancing resource allocation and decision-making. For example, a study by Deloitte found that AI-powered analytics can improve budget forecasting accuracy by up to 20% [4]. Decisions about policies frequently rely on massive amounts of data. Extracting meaningful insights from such data can be time-consuming and challenging. AI can analyze massive datasets to find patterns, correlations, and possible risks related to suggested policies [5].

While AI offers immense potential, implementing it within the public sector comes with challenges. Concerns around data privacy, security, and bias in AI algorithms need to be addressed [6]. Additionally, a skilled workforce is needed to develop, maintain, and manage AI systems [7]. To ensure responsible AI adoption, public sector institutions need to develop robust frameworks. These frameworks should address issues like data privacy, algorithmic bias, and transparency. It is crucial to build public trust in AI by ensuring its use is ethical, fair, and accountable [8]. Public sectors face several challenges when procuring AI systems, such as a lack of internal expertise, an evolving regulatory landscape, data challenges, procurement processes, budgetary constraints, vendor lock-in, and public trust concerns [9]. By understanding these challenges, public sector organizations can develop more effective strategies for procuring and implementing AI systems, ultimately leading to a smoother transition toward a more efficient and citizen-centric government. In the context of the UAE public sector, there is a notable lack of empirical studies. More empirical research on critical success factors (CSFs) is needed to procure AI systems in the UAE public sector. This gap impairs our understanding of the critical factors required for successful AI integration and affects the achievement of public good results. A comprehensive identification and analysis of these CSFs could help public sector projects achieve greater efficiency, transparency, and social benefits. Addressing this gap is critical to ensuring that AI procurement serves the public good in the UAE. In this context, this study investigates the following research questions:

• What CSFs underpin the procurement of AI systems and services in the UAE’s public sectors?
• What are the leading and most impactful CSFs?

This research delves into AI procurement in the UAE public sector, which is a unique and under-explored area. Procurement of AI systems, in general, is not a well-studied field. The success of AI projects hinges on effective procurement. This research employs ISM, a powerful technique for uncovering the relationships between CSFs. Existing structural models, like those built with Artificial Neural Networks (ANNs) and Structural Equation Modeling (SEM), often rely on visual representations like matrices, graphs, and flowcharts [10]. However, these models do not always clearly explain the underlying relationships between the factors involved. ISM helps identify driving and dependent factors, leading to a more nuanced understanding of how various elements influence successful AI procurement. This research aims to identify the CSFs specifically for the UAE public sector, thus filling a gap in current knowledge. Understanding these factors can help policymakers and practitioners determine how to best approach AI procurement for the public good. By using ISM to analyze CSFs, this research can contribute to the broader theory of technology adoption in public sectors. It can reveal how existing frameworks might need to be adapted for the specific context of AI procurement in the UAE.

The remainder of the study is organized as follows: Section 2 provides an overview of AI use in public sectors, along with a discussion of CSFs for AI systems and services procurement. Section 3 explores the use of ISM to create a model for assessing CSFs for the purchase of AI systems and services. In addition, MICMAC analysis is used to categorize the CSFs based on their power and dependence. Section 4 provides a discussion along with the theoretical and managerial implications of the findings. Section 5 concludes with the limitations of this study and highlights future research directions.

2. Literature Review

2.1. Adoption of AI in Public Sectors

AI systems are increasingly being applied across public and private sectors, as well as in broader society. The global trend of AI adoption in public sectors is rapidly accelerating, driven by the potential for significant improvements in efficiency, service delivery, and decision-making. Governments worldwide are recognizing the transformative power of AI and actively exploring its applications across various domains. A 2018 report by McKinsey Company estimates that an additional $13 trillion could be added to global economic output by AI by 2030, with significant implications for public sector efficiency [11]. A 2023 perspective from Deloitte highlights the growing interest of governments in leveraging AI for citizen services, citing the use of generative and predictive technologies for streamlined and data-driven service delivery [4]. A report by EY India emphasizes the rise of generative AI in public sector operations, noting its role in automating tasks, improving data-driven decision-making, and fostering agile citizen-centric services [12].

Inefficiencies, corruption, and a lack of transparency are common issues in public procurement, resulting in considerable economic losses and eroding public trust [13,14]. The procedure is usually bureaucratic, lengthy, and complicated, making it susceptible to human mistakes and manipulation. Inadequate monitoring mechanisms allow for overpricing, partiality, and resource misallocation, while the sheer volume of data and documentation can overwhelm traditional control measures [15]. Furthermore, limiting access to information can stifle competition, resulting in inferior outcomes for public initiatives.

Automation driven by AI can optimize administrative workflows, enabling public servants to prioritize strategic responsibilities and potentially lower operational expenses [16,17]. AI can analyze vast datasets to identify trends and patterns, enabling evidence-based policy formulation and resource allocation that better address public needs [18]. AI-powered chatbots can provide 24/7 citizen support, personalize service experiences, and automate repetitive tasks, thereby enhancing citizen engagement and satisfaction. AI-powered video analytics can be used for real-time surveillance, traffic management, and crime prediction, potentially enhancing public safety and security [12]. The implementation of AI in the public sector offers significant advantages, including the transformation of service delivery and decision-making processes, and the enhancement of overall citizen welfare. Nevertheless, it is crucial to recognize the challenges linked to AI adoption, including issues of data security, ethical implications, and the need for workforce skills development. Addressing these factors is essential for ensuring responsible and effective implementation.

Public institutions hold vast amounts of sensitive citizen data. Concerns regarding data breaches and privacy violations pose a significant challenge to AI implementation. Algorithms trained on biased data can perpetuate inequalities, and a lack of transparency in data governance can erode public trust [19,20]. Public sector agencies often lack the in-house expertise to develop, implement, and maintain complex AI systems. This skills gap can hinder project selection, vendor evaluation, and effective oversight of AI procurement processes [21,22]. AI algorithms trained on biased datasets can perpetuate existing inequalities and lead to discriminatory outcomes. Addressing algorithmic bias and ensuring inclusive AI development is crucial for responsible AI procurement in public sectors [23,24]. It is also crucial to ensure transparency in AI decision-making processes to foster trust and secure public acceptance of AI-driven solutions. Public agencies need to understand how AI systems arrive at their conclusions, particularly for applications with high-stakes consequences [13,25]. Beyond data privacy and algorithmic bias, broader ethical considerations require careful exploration. These include potential job displacement due to automation, the use of AI for surveillance, and the potential misuse of AI for malicious purposes [26,27]. Outdated IT infrastructure in some public sector organizations can hinder AI implementation by limiting data accessibility and processing capabilities [28]. Existing procurement regulations might not be well-suited for the unique considerations of AI technologies, potentially leading to cumbersome processes and hindering innovation [29].

Table 1. Pros and cons of the various aspects.

Aspects	Pros	Cons
Efficiency	AI is capable of automating repetition and time-consuming activities, and increasing processing speeds, which leads to an increase in productivity and a reduction in operational costs [4,11].	AI can displace jobs, lose human oversight, and give less scope for critical thinking in the decision-making process if it is overemphasized [30].
Decision-Making	AI can also analyze large and complex datasets to support evidence-based decision-making, potentially improving accuracy and consistency [18].	AI algorithms may not only perpetuate but amplify existing data biases, eventually leading to biased results or discrimination towards some groups [19,20,31].
Transparency	AI systems could generate lucid and consistent decision processes and, hence, increase transparency if designed for explainability [32].	Many AI systems are by nature “black box”—that is, impenetrable to scrutiny of their decisions [24,25].
Accountability	AI can standardize decision-making processes, reducing human error and providing a consistent basis for policy application [33].	Assigning responsibility for AI-driven decisions can be challenging, creating gaps in accountability and complicating legal and ethical governance [34].
Cost	AI systems can help minimize long-term operational costs after an initial setup by way of task automation and resource optimization [16,17].	Setting up AI systems could be very costly, especially at the beginning, with all the implicit maintenance and updating that will constantly be required afterward. This will likely be a heavy drain on public sector budgets, especially in the lower-income regions [35].
Data Privacy and Security	AI has the potential to enhance the capability of managing, analyzing, and securing data for better public service delivery and engaging citizens [25,36].	The vast amounts of data required for AI systems give rise to large concerns about the privacy and security of personal data and the potential risks of breaches or misuses [4,37].
Ethical Consideration	AI could help in better identification of and response to societal needs for improved public welfare and service delivery [20].	It raises ethical concerns in relation to surveillance, social control, eroding human rights, and differential impacts [1,19].
Scalability	AI can scale services through its management of huge amounts of tasks or data where human workers would be impractical [28,38].	Scaling AI without careful consideration can lead to widespread systemic issues, such as biased outcomes or unequal access to services [5,39].
Innovation	It is within the power of AI to help accelerate innovation in public services by making it possible for new approaches toward policy design, service delivery, and problem-solving [7,28].	AI-driven innovation sometimes races ahead of regulatory frameworks, opens the way to ethical and legal dilemmas, or plants doubt in the public’s mind [33,40].
Public Trust	If effectively implemented, AI can enhance the quality and efficiency of public services, potentially improving public trust in government [41].	Failures or misuse of AI in the public sector may further result in a loss of trust by citizens, particularly if it is perceived to be unfair, opaque, or prone to errors [42].
Adaptability	It will thus make public services more flexible and responsive because AI systems can adapt to new data and changing circumstances [31,43].	Non-adaptive, inflexible AI systems may miss the appropriateness of dynamic or unexpected situations and therefore make inappropriate or even harmful decisions [23,25].
Long-Term Value	AI can help in optimizing the use of resources, hence avoiding unnecessary wastage and creating long-term sustainability in the public sector [44,45].	It can lead to environmental degradation by consuming artificial intelligence systems with high energy levels, particularly through data centers and computation [3,46].

provides a list of pros and cons of the various aspects of the adoption of AI systems and services in the public sector.

Several frameworks and models exist to guide successful AI adoption in public sectors. However, it is important to acknowledge their limitations, especially in the context of research on AI procurement in the UAE public sector. The framework by the Organization for Economic Co-operation and Development (OECD) outlines five key principles for responsible AI development and deployment, including human-centered design, fairness, transparency, accountability, and safety [47]. While comprehensive, it does not delve specifically into the procurement process. The World Bank’s framework for artificial intelligence in the public sector emphasizes the importance of central digital agencies, data governance, and collaboration for successful AI adoption [48]. However, it may not provide a detailed roadmap for the specific procurement challenges faced by individual countries. These models categorize public sector organizations into stages based on their AI adoption maturity. They offer guidance on building capabilities and infrastructure, but details on the procurement process may be limited [22]. Existing models might not provide detailed guidance on vendor selection, contract negotiation, risk mitigation strategies, unique cultural considerations, existing policies, or infrastructure limitations within the UAE public sector.

2.2. The UAE Public Sector Context

The UAE public sector follows a specific framework for procurement, with some unique considerations for technology acquisition. The Federal Regulation of Conditions of Purchases, Tenders, and Contracts, Financial Order No. 16 of 1975 (the “Public Tenders Law”) establishes the minimum standards for government procurement practices across the UAE. It outlines principles such as transparency, fair competition, and value for money [49]. Each emirate (state) within the UAE might have additional regulations specific to its procurement practices. These regulations might build upon the Federal Public Tenders Law but could also include local variations [50]. The UAE government often prioritizes procurement from local companies to foster domestic industry growth [51]. This might involve requiring companies to establish a local presence before participating in tenders for technology acquisition. Public sector organizations might pre-qualify companies to participate in tenders, ensuring that they meet specific criteria related to experience, technical expertise, and financial stability. This can be particularly relevant for complex technology acquisitions. Direct procurement from international vendors might be restricted, with a requirement to go through authorized local distributors or resellers. The combination of federal regulations and emirate-specific provisions can create a complex procurement landscape for technology acquisition. Understanding the specific requirements applicable to AI projects and the emirate is crucial. While promoting local companies is a priority, ensuring access to cutting-edge technologies also remains important. The UAE government is working to create a balance between these objectives [52].

The UAE government has implemented many AI technologies to further its national objectives. For example, in healthcare, the Dubai Health Authority (DHA) employs IBM Watson, an AI platform, to improve cancer detection and treatment by analyzing large amounts of medical data and recommending individualized treatment strategies [53]. In education, the Ministry of Education uses AI-powered learning systems such as Alef Education [54], which uses machine learning to personalize instructional content to specific student requirements, thereby increasing engagement and outcomes. Dubai’s Smart Dubai Office uses artificial intelligence for traffic management via technologies such as RTA’s Integrated Traffic Platform [55], which forecasts traffic patterns and optimizes flow using real-time data. The judiciary has integrated AI into systems such as the Dubai Courts’ “Remote Litigation” platform, which automates legal procedures and document management [56]. Furthermore, the UAE’s space program employs AI for satellite data analysis, improving missions like the Hope Probe to Mars [57].

The UAE has positioned itself as a regional leader in digital transformation and AI adoption [58]. The government’s ambitious initiatives seek to utilize these technologies for improving public services, fostering economic growth, and establishing the UAE as a leading global innovation hub. In 1982, the establishment of the Public Information Authority marked the initial steps towards computerization in government work [59]. The UAE Centennial 2071 vision, laid out in 2017, is a long-term roadmap for transformative development, with technology playing a central role [60]. The UAE National Digital Transformation Strategy 2025 prioritizes using AI across government services, improving citizen lives, and establishing the UAE as a leader in AI investments [42]. The UAE Artificial Intelligence Strategy 2031 is designed to align with the goals of UAE Centennial 2071, enhance government performance, and position the UAE as a global leader in AI development [39]. The UAE Blockchain Strategy 2021 emphasizes adopting advanced technologies like AI to achieve ambitious national goals [44]. The UAE AI and Blockchain Council brings together representatives from all emirates to ensure coordinated efforts.

The UAE’s supplier credit evaluation and bid template uniformity policy provides considerable benefits while also posing obstacles [61]. On the plus side, stringent credit evaluations reduce risks by ensuring that only financially sound suppliers are hired, which is crucial for the success of large-scale projects. Standardized bid templates increase openness and efficiency, allowing for more equitable comparisons and decreasing the possibility of bias. The use of modern technology, such as AI, in credit assessments improves decision-making. However, the policy’s emphasis on financial indicators may miss smaller, more inventive providers that lack a good financial track record but offer useful solutions. Furthermore, the rigidity of predefined templates might discourage creativity, limiting the opportunity for novel ideas that could benefit initiatives [62].

Despite the notable advancements, there are ongoing challenges. It remains imperative to build a proficient workforce to support AI development, implementation, and maintenance. Ethical concerns, including addressing bias and ensuring responsible AI development, are critical for establishing public trust. These unique characteristics present various key considerations for AI procurement in the UAE public sector.

2.3. Critical Success Factors for AI Procurement

The public sector is rapidly embracing AI to enhance efficiency, improve citizen services, and tackle complex societal challenges. However, successfully harnessing this transformative technology requires careful planning and execution. A crucial aspect of this process is identifying and managing the CSFs for AI procurement. These factors are essential elements that must be addressed to ensure the successful implementation of AI within the distinctive context of government agencies.

This research delves into the landscape of AI procurement for public services. We explored the distinct challenges and opportunities that governments face when acquiring AI solutions. By analyzing existing literature and drawing insights from expert interviews, we identify and categorize the key CSFs for successful AI procurement within the public sector. Successful IT procurement, particularly in the public sector, requires careful consideration of various factors to ensure cost-effectiveness, efficiency, and alignment with organizational goals. The key CSFs identified in existing literature consider various perspectives such as general IT procurement requirements, and public sector-specific, AI-specific, and UAE-specific considerations. To identify CSFs in public procurement articles, search engines from various publishers were used, including Wiley Online Library, IEEE Explorer, Emerald Insight, Inderscience Online, Science Direct (Elsevier), Springer, and Taylor and Francis Group. Keywords used for the search include “crucial success factors”, “drivers”, “motivators”, “determinants”, “influencing variables”, “public procurement”, “multi-criteria decision-making (MCDM)”, and “multi-attribute decision-making (MADM)”. Further, 21 CSFs were shortlisted after reviewing conference proceedings, book chapters, periodicals, journals, and white papers, with input from professionals. The list of CSFs identified from the literature was presented to 16 procurement experts (thirteen from the UAE public sector and three from academia) to assess their importance and applicability. Based on the opinions of the experts, removing redundancy and other similarities, ten CSFs were identified as shown in

Table 2. CSFs for procurement of AI systems or services in the UAE public sector.

S.No.	CSFs	Literature
		1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20
1	Clear Needs Assessment	✓	✓			✓					✓							✓			✓
2	Effective Vendor Selection		✓	✓	✓		✓						✓				✓		✓
3	Strong Contract Negotiation	✓				✓	✓				✓				✓			✓
4	Project and Change Management		✓		✓			✓	✓			✓		✓
5	Compliance with Regulations	✓				✓				✓	✓	✓			✓		✓				✓
6	Stakeholder Management		✓	✓									✓	✓		✓			✓	✓
7	Clear AI Strategy and Alignment	✓			✓		✓							✓	✓		✓				✓
8	Understanding AI Capabilities		✓	✓				✓				✓				✓	✓
9	Data Governance	✓			✓			✓			✓	✓					✓	✓	✓
10	Long-Term Value	✓				✓	✓												✓	✓	✓

1 [63], 2 [46], 3 [64], 4 [65], 5 [66], 6 [40], 7 [67], 8 [68], 9 [69], 10 [70], 11 [71], 12 [72],13 [48], 14 [73], 15 [11], 16 [45], 17 [41], 18 [43], 19 [37], 20 [38].

.

A brief description of each CSF as conveyed to experts is provided below for clarity. By adapting these AI procurement CSFs to the specific context of the UAE public sector, government agencies can ensure responsible, effective, and sustainable AI adoption.

• Clear Needs Assessment: A thorough understanding of the organization’s needs and clearly defined project requirements are essential for selecting the most suitable IT solutions [46,63].
• Effective Vendor Selection: A robust vendor selection process that considers factors like vendor experience, technical capabilities, and cost-competitiveness is critical [40,64,65].
• Strong Contract Negotiation: Well-defined contracts with clear terms, conditions, and performance metrics are crucial for risk mitigation and ensuring successful project execution [64,66].
• Project and Change Management: Effective project management practices and comprehensive change management strategies are essential for smooth IT implementation and user adoption [67,68].
• Compliance with Regulations: Public sector IT procurement adheres to specific regulations and policies governing financial accountability, transparency, and fair competition [69,70].
• Stakeholder Management: It is crucial to involve stakeholders like government officials, end users, and community members throughout the procurement process to ensure their support and the success of the project [71,72].
• Clear AI Strategy: A well-defined AI strategy outlining goals, use cases, and ethical considerations is crucial for guiding successful AI procurement. Aligning the AI solution with existing IT infrastructure and long-term technological vision is also important [48,73].
• Understanding of AI Capabilities: Public agencies need to realistically assess their needs and the capabilities of available AI solutions. Avoiding overreliance on AI and ensuring human oversight is crucial for responsible implementation [11,73].
• Data Governance: Strong data governance frameworks are vital for safeguarding data security, privacy, and quality in AI initiatives. Furthermore, acquiring AI solutions that employ transparent algorithms to enhance accountability and public confidence is critical for the UAE public sector [41,45,70].
• Long-Term Value: The UAE public sector should prioritize procuring AI solutions that offer long-term value and sustainability. This includes considering factors like scalability, maintainability, environmental and social impact, and the availability of ongoing support [37,38,43].

2.4. ISM for Identifying Driving and Dependent CSFs in AI Procurement

While research specifically focusing on ISM and AI procurement is limited, its application in exploring CSFs for complex projects holds significant advantages. AI procurement involves numerous interconnected factors, making it challenging to identify their relative importance and relationships. ISM provides a structured methodology for systematically analyzing pairwise comparisons of these CSFs, leading to a more comprehensive understanding [74]. By continuously refining the ISM model, it becomes possible to pinpoint CSFs that exert substantial influence on other factors, alongside dependent CSFs that are profoundly impacted by these influential factors [74]. This categorization helps prioritize efforts and resources during AI procurement. The expert-driven approach of ISM fosters collaboration and discussion among stakeholders involved in AI procurement. The pairwise comparisons encourage stakeholders to explicitly articulate their understanding of the relationships between CSFs, potentially leading to a more unified and informed decision-making process [75]. The final ISM model, typically presented as a digraph, provides a clear visual representation of the interrelationships among CSFs [76]. This transparency facilitates communication between stakeholders, ensuring everyone involved has a shared understanding of the key drivers and dependencies in the AI procurement process. By leveraging these advantages, ISM can be a valuable tool for identifying, prioritizing, and effectively managing the CSFs crucial for successful AI procurement within complex public sector environments. The ISM methodology involves five key steps: (1) Developing the structural self-interaction matrix (SSIM); (2) Creating the reachability matrix; (3) Partitioning of factors; (4) Classifying factors; and (5) Constructing the ISM.

3. Materials and Methods

3.1. ISM and MICMAC to Analyze CSFs for Procurement of AI

This section explores the key success factors for AI procurement within UAE public sectors. It aims to build a structural model based on contextual relationships, which is then followed by a MICMAC analysis to categorize these factors and assess their driving influences. This subsection outlines the Babu, Bhardwaj, and Agrawal [36] methodology, which is adopted for this study as shown in Figure 1. Additionally, MICMAC analysis is utilized to identify the dependent, linkage, and autonomous factors.

3.2. ISM Model for CSFs for Procuring AI

• Step 1: Identification and definition of the critical success factors for procuring AI systems and services

The CSFs for procurement of AI systems in the public sector are based on existing literature as presented in Section 2.3. Furthermore, these CSFs are validated by experts’ opinions on their importance and applicability in the public sector of the UAE. Previous studies suggest that the ideal number of experts for qualitative research is between 5 and 50 [77]. This study used seven experts, the same number as used in other ISM studies such as Mathivathanan, D., et al. (2021) [10]. These experts are procurement managers, AI experts, purchasing managers, and academia members. In this study, a total of 16 experts were contacted (13 from the UAE public and 3 from academia), as mentioned in Section 2.3. However, only five responses have been received. Additionally, two academicians were also asked for their thoughts on the identified CSFs to support the perspectives of industry specialists. Seven experts in total—five from business and two from academia—were contacted to determine the contextual linkages among the components. All of these specialists were given a brief explanation of these CSFs and their needs beforehand, which was followed by a brainstorming session. During the brainstorming sessions, as required by the ISM methodology, the final opinions were converged to a single input for a particular relationship between two factors. These CSFs, representing the key elements for successful procurement, form the basis for our interpretive structural modeling (ISM) analysis, where the relationships between these factors will be modeled. Ten potential CSFs were identified from the literature, including clear needs assessment (C1), effective vendor selection (C2), strong contract negotiation (C3), project and change management (C4), compliance with regulations (C5), stakeholder management (C6), clear AI strategy (C7), understanding of AI capabilities (C8), data governance (C9), and long-term value (C10). The identified CSFs were integrated into an interpretive knowledge base designed to capture expert opinions.

The process of developing 90 paired relationship-based questions, along with their underlying logic, proved to be a demanding task for the experts participating in this study. Therefore, only a select group of experts with substantial experience in both the procurement process and the procurement of AI systems or services were contacted. Experts’ opinions were used to create a pairwise contextual link between each pair of the CSFs; for example, if CSF1 and CSF2 are related to one another, a YES is entered in the table (

Table 3. The interpretive logic knowledge base.

SN	Paired CSFs	Comparison	Does a Relationship Exist?	A Concise Summary of the Relationship, Where Applicable
1	C1–C2	Needs Assessment will influence Effective Vendor Selection	Yes	By clearly defining needs and requirements up front, C1 helps ensure that C2 selects the most suitable vendor for the public agency’s AI procurement project.
2	C2–C1	Effective Vendor Selection will influence Needs Assessment	No
.	.
.	.
.	.
.	.
.	.
.	.
90	C10–C9	Long-Term Value and Sustainability will influence Data Governance	No

); otherwise, a NO is written. If a relationship is present, the experts were questioned again on the relationship’s direction—that is, whether it is from i to j. A general awareness of the CSFs was provided to all seven experts. The definition of each CSF, as described in Section 2.3, was communicated to all experts before a brainstorming session. Thus, these experts are appropriate respondents for this ISM model.

• Step 2: Contextual relationship

Defining the contextual link between the specified CSFs is the next phase. In this phase, expert judgment is used to identify pairwise contextual links between each of the CSFs. For example, if CSF C1 affects CSF C2, the interpretation of this relationship is considered. The interpretative knowledge base for our study is developed by communicating with experts who provide ‘Yes’ or ‘No’ responses along with the logical reasoning behind them. This process clarifies how CSF C1 affects CSF C2, and so forth. These discussions are captured in the matrix of the knowledge base. The judgments of the experts are used to build the pairwise contextual links between the CSFs. Pairwise relationship analysis between CSFs was conducted through expert interviews and brainstorming sessions. The knowledge base is further structured into a table format, where each row represents a pair of compared CSFs along with their contextual relationship, as provided by the experts.

Since this study includes ten CSFs, the knowledge base consists of a total of 10 × 9 = 90 entries. After consulting with experts on these 90 potential connections, the knowledge base was established and is represented in Table 3. A comparison measure had to garner a positive response from over 50% of respondents to be categorized as ‘Yes’. Then, during brainstorming sessions, experts were asked to converge (total agreement) on a relationship between each pair of CSFs. All of the ‘Yes’ answers, meaning that there was a contextual link present, were analyzed. Experts’ interpretations were utilized to develop consolidated interpretation statements, which were then documented in the explanation column.

• Step 3: Structural Self Interaction Matrix

The understanding of the ‘Yes/No’ connection among the compared CSFs is represented in an “n × n” matrix, with n denoting the total number of CSFs included in the study. Four symbols were employed to depict the relationships among CSFs affecting the procurement of AI services in public sectors. The symbols “V”, “A”, “X”, and “O” are employed in this study to represent the relationships between the factors i and j (see

Table 4. Structural self-interaction matrix (SSIM).

CSFs	C1	C2	C3	C4	C5	C6	C7	C8	C9	C10
Clear Needs Assessment (C1)	-	V	V	V	V	O	V	V	V	V
Effective Vendor Selection (C2)		-	A	A	A	A	O	A	A	V
Strong Contract Negotiation (C3)			-	V	V	O	O	O	O	O
Project and Change Management (C4)				-	V	V	V	A	O	V
Compliance with Regulation (C5)					-	V	V	O	V	V
Stakeholder Management (C6)						-	X	O	O	V
Clear AI Strategy (C7)							-	O	O	V
Understanding of AI Capabilities (C8)								-	A	V
Data Governance (C9)									-	V
Long-term Value (C10)										-

).

• V = denotes that CSF ‘i’ helps to attain CSF ‘j’
• A = denotes that CSF ‘j’ will be attained by CSF ‘i’
• X = denotes that CSFs ‘i’ and ‘j’ will mutually help to attain each other
• O = denotes that CSFs ‘i’ and ‘j’ are not related

• Step 4: Binary representation of pairwise comparisons

The SSIM attained in the previous step is transformed into a binary matrix, which is called the initial reachability matrix. This is done by replacing the “V”, “A”, “X”, and “O” signs with binary variables 1 and 0 by following the mentioned rules. According to the impact of CSF Ci over CSF Cj, the value of either “1” or “0” is entered for each element (i,j) of the matrix. The value of “0” indicates the lack of a relationship, while the value of “1” indicates the presence of an influential relationship between Ci and Cj (Jayalakshmi and Pramod, 2015 [78]). In this instance, we created a 10 × 10 matrix, with 10 × 9 = 90 as the total number of pairwise comparisons. The initial reachability matrix was created based on the information in the knowledge base. The comparisons are shown as a matrix, where all entries contain the binary values “1” or “0”, except the diagonal elements. If the logic knowledge base indicates that there is a link between the compared barriers, value “1” is written in the cells; if not, value “0” is inserted.

Table 5. Initial reachability matrix.

CSFs	C1	C2	C3	C4	C5	C6	C7	C8	C9	C10	Driving Power
Clear Needs Assessment (C1)	1	1	1	1	1	0	1	1	1	1	9
Effective Vendor Selection (C2)	0	1	0	0	0	0	0	0	0	1	2
Strong Contract Negotiation (C3)	0	1	1	1	1	0	0	0	0	0	4
Project and Change Management (C4)	0	1	0	1	1	1	1	0	0	1	6
Compliance with Regulation (C5)	0	1	0	0	1	1	1	0	1	1	6
Stakeholder Management (C6)	0	1	0	0	0	1	1	0	0	1	4
Clear AI Strategy (C7)	0	0	0	0	0	1	1	0	0	1	3
Understanding of AI Capabilities (C8)	0	1	0	1	0	0	0	1	0	1	4
Data Governance (C9)	0	1	0	0	0	0	0	1	1	1	4
Long-term Value (C10)	0	0	0	0	0	0	0	0	0	1	1
Dependence Power	1	8	2	4	4	4	5	3	3	9

below displays the first reachability matrix that was created in this manner. Here, the only direct relationships are indicated with the value “1”. The diagonal values that are indicated in yellow are always taken to be “1”.

• Step 5: Check for transitivity

Using the transitivity rule, we analyze the initial reachability matrix, which is constructed based on the logical interpretation of ‘Yes/No’ connections, to identify potential transitive relationships. Should Cx impact Cy and Cy impact Cz, then Cx impacts Cz. An example of a transitive relation is observed between CSF C1 (clear needs assessment) and CSF C6 (stakeholder management). In the initial reachability matrix, no direct relationship was identified between them. However, ISM suggests that clear needs assessment influences stakeholder management, thus indirectly affecting the procurement process. Purely transitive relationships with substantial elucidations are considered for additional analysis, while others are excluded. Examining the transitive linkages with the advice of experts is the next stage. Experts are consulted to examine each resultant transitive relationship individually, deducing the likelihood of transitivity, and subsequently updating the knowledge base.

Table 6. Final reachability matrix.

CSFs	C1	C2	C3	C4	C5	C6	C7	C8	C9	C10	Driving Power
Clear Needs Assessment (C1)	1	1	1	1	1	1 *	1	1	1	1	10
Effective Vendor Selection (C2)	0	1	0	0	0	0	0	0	0	1	2
Strong Contract Negotiation (C3)	0	1	1	1	1	1 *	1 *	1 *	1 *	1 *	9
Project and Change Management (C4)	0	1	0	1	1	1	1	1 *	1 *	1	8
Compliance with Regulation (C5)	0	1	0	1 *	1	1	1	1 *	1	1	8
Stakeholder Management (C6)	0	1	0	0	0	1	1	0	0	1	4
Clear AI Strategy (C7)	0	1 *	0	0	0	1	1	0	0	1	4
Understanding of AI Capabilities (C8)	0	1	0	1	1 *	1 *	1 *	1	1 *	1	8
Data Governance (C9)	0	1	0	1 *	1 *	1 *	1 *	1	1	1	8
Long-term Value (C10)	0	0	0	0	0	0	0	0	0	1	1
Dependence Power	1	9	2	6	6	8	8	6	6	10

* Transitive relationship.

displays the final reachability matrix.

• Step 6: Level partition

To ascertain the hierarchical placement of CSFs, levels are divided. By analyzing the influence of each CSF as a driver or dependence, we compute the reachability set, antecedent set, and intersection set, as presented in Table 6. The CSFs above their level are not reached by the CSFs at the highest hierarchical level. The final reachability matrix is created from the previous phase, incorporating relationships derived from indirect or transitive relationships, along with entries based on direct pair-wise assessments. The reachability, antecedent, and intersection sets for each CSF are created from the final reachability matrix. The top-level CSF (Level I) is identified when the intersection set and the reachability set coincide. These Level I CSFs are then excluded from an overall set in the subsequent iteration of the table. Until each CSF is given its appropriate level, this process is repeated. Ultimately, following six cycles, each piece is assigned a level, and

Table 7. Level partition of CSFs.

Element (Mi)	Reachability Set R (Mi)	Antecedent Set A (Ni)	Intersection Set R(Mi)∩A(Ni)	Level
1	1	1	1	6
2	2	1, 2, 3, 4, 5, 6, 7, 8, 9	2	2
3	3	1, 3	3	5
4	4, 5, 8, 9	1, 3, 4, 5, 8, 9	4, 5, 8, 9	4
5	4, 5, 8, 9	1, 3, 4, 5, 8, 9	4, 5, 8, 9	4
6	6, 7	1, 2, 3, 4, 5, 6, 7, 8, 9	6, 7	3
7	6, 7	1, 2, 3, 4, 5, 6, 7, 8, 9	6, 7	3
8	4, 5, 8, 9	1, 3, 4, 5, 8, 9	4, 5, 8, 9	4
9	4, 5, 8, 9	1, 3, 4, 5, 8, 9	4, 5, 8, 9	4
10	10	1, 2, 3, 4, 5, 6, 7, 8, 9, 10	10	1

illustrates this iterative process.

• Step 7: Creating digraph

The study graphically organizes the CSFs by their levels and indicates the connections derived from the final reachability matrix with arrows. In our scenario, creating a digraph entails graphically placing each of the ten CSFs according to the levels they attained during the level divisions. Using the connections in the final reachability matrix, the relationships between the CSFs are represented visually as arrows. The digraph created to illustrate the produced hierarchical model is shown in Figure 2.

• Step 8: Final ISM model

The creation of the ISM model is the last phase. The data are included in the acquired diagram. As a result, the completely restructured model is produced and displayed in Figure 3.

• Step 9: Validation of ISM Model

A limitation of the developed ISM model was the low response rate, primarily due to the experts needing extra time to conduct pairwise comparisons and provide detailed explanations for each comparison pair. In our case, expert input was essential for 10 × 9 = 90 comparisons. It required significant effort to establish contextual linkages and provide reasoning for each of the 90 pairs. The recruitment of volunteers for this taxing procedure was also somewhat challenging. As a result, just seven people offered their time to assist in creating the model. The number of links was drastically reduced once the ISM model was developed. There are only 16 significant relationships in the established model. The number of significant linkages has drastically decreased, making it far less time-consuming for any expert to check the links. Therefore, a larger panel of experts—totaling 10—evaluated the developed ISM model this time.

On a Likert scale of 1 to 5, where 5 represents “strongly agree” and 1 represents “strongly disagree”, each expert was asked to score the linkages. If each link in the model achieved an average score of three out of five, it was considered approved; if the average score of all the linkages is higher than three, the model as a whole is accepted.

Table 8. Validation of ISM model.

SN	Relationship	Responses from Experts (E)										Average Response	Accept/Reject
		E1	E2	E3	E4	E5	E6	E7	E8	E9	E10		Accept
1	Clear Needs Assessment will influence Strong Contract Negotiation	4	5	5	4	3	5	4	5	5	3	4.3	Accept
2	Strong Contract Negotiation will influence Project and Change Management	3	4	5	3	4	4	4	3	5	4	3.9	Accept
3	Strong Contract Negotiation will influence Compliance with Regulations	4	4	5	3	5	4	3	3	4	5	4	Accept
4	Strong Contract Negotiation will influence Understanding of AI Capabilities	4	4	4	3	3	3	4	4	3	4	3.6	Accept
5	Strong Contract Negotiation will influence Data Governance	3	3	2	2	4	4	3	3	4	4	3.2	Accept
6	Project and Change Management will influence Stakeholder Management	5	5	5	4	5	5	4	5	3	5	4.6	Accept
7	Project and Change Management will influence Clear AI Strategy	4	5	3	2	5	4	2	5	4	3	3.7	Accept
8	Compliance with Regulations will influence Stakeholder Management	3	4	5	5	3	2	5	4	4	4	3.9	Accept
9	Compliance with Regulations will influence Clear AI Strategy	3	3	3	4	4	2	4	4	4	5	3.6	Accept
10	Understanding of AI Capabilities will influence Stakeholder Management	3	3	4	2	4	4	4	3	3	3	3.3	Accept
11	Understanding of AI Capabilities will influence Clear AI Strategy	5	5	4	5	5	3	3	4	4	5	4.3	Accept
12	Data Governance will influence Stakeholder Management	3	3	4	2	5	3	4	5	3	5	3.7	Accept
13	Data Governance will influence Clear AI Strategy	2	5	4	3	5	4	2	4	5	5	3.9	Accept
14	Stakeholder Management will influence Effective Vendor Selection	3	3	3	5	4	3	4	4	5	3	3.7	Accept
15	Clear AI Strategy will influence Effective Vendor Selection	4	5	5	4	4	3	3	2	5	2	3.7	Accept
16	Effective Vendor Selection will influence Long-Term Value	3	2	5	4	4	4	5	5	3	2	3.7	Accept
		Overall score for model										3.81	Accept

below presents the evaluation of the ISM model.

3.3. MICMAC Analysis

An established technique for analyzing the influence of variables or elements based on relationships is called MICMAC. We use MICMAC to determine the CSFs for procurement of AI systems and services in the UAE public sector. Based on the influence and dependence of each CSF, calculated as the sum of rows and columns in the final reachability matrix displayed in

Table 9. Driving power and dependence power of CSFs.

CSFs	C1	C2	C3	C4	C5	C6	C7	C8	C9	C10
Driving power	10	2	9	8	8	4	4	8	8	1
Dependence power	1	9	5	6	6	8	8	6	6	10

, MICMAC categorizes the CSFs into four quadrants: autonomous, dependent, linkage, and driving, providing a graphical representation.

The summation across row 2, representing CSF C2, indicates its perceived driving power, while the summation across column 2 reflects its dependence. CSFs are graphed based on their driving power versus dependence and categorized into the four quadrants as illustrated in Figure 4. The autonomous components, or CSFs with modest driving and reliance capabilities, are represented by the first quadrant. CSFs which are weak drivers but naturally highly reliant make up the second quadrant. CSFs which have great driving and dependence power make up the third quadrant. The CSFs in this quadrant are referred to as unstable linkage elements because of the way they impact the system as a whole. The driving factors that have high driving but weak dependence capacities are represented by the fourth quadrant.

4. Discussion

A six-level ISM model was constructed where CSF C10, “long-term value”, is positioned at Level 1, indicating that it is the least influential CSF and thus appears at the topmost layer of the digraph. This suggests that this CSF is not critical to the procurement of AI systems and services in the UAE public sector but may be influenced by other more essential factors. Level 2 consists of CSF C2, i.e., effective vendor selection. This factor has less driving power compared to other factors except CSF C10. Level 3 consists of two CSFs, C6 (stakeholder management) and C7 (clear AI strategy). These factors are found to be important and driving factors at levels 1 and 2. Level 4 consists of four CSFs, namely C4 (project and change management), C5 (compliance with regulations), C8 (understanding of AI capabilities), and C9 (data governance). CSF C3 (strong contract negotiation) is placed at level 5, which implies strong driving power. The final and bottom level of this structure is level 6, which consists of CSF C1 (clear needs assessment), found to be the most influential CSF. This is the most critical success factor and provides valuable connections throughout the entire system, linking to other levels. When policymakers are not sure about needs and do not have requirement definitions, it can upset the procurement process of AI systems and services. Adoption of AI services and systems in public sectors is on the rise and AI systems are still evolving, but these are the most critical success factors for procurement of AI systems and services in UAE public sectors. This study provides the insight that once these CSFs are considered, public sector companies can achieve long-term value and sustainability for AI systems and services.

Moreover, the MICMAC analysis produces a diagram that illustrates the driving and dependence powers, highlighting the relative importance and interrelationships among CSFs. The first quadrant includes autonomous CSFs, which are characterized by their minimal influence and low impact. These CSFs, due to their low driving and dependence powers, are found to have minimal connections within the system [79]. In this study, no autonomous variables are present. If autonomous variables were present, they would be considered driving variables and given high priority. In the second quadrant, CSFs depend significantly on other CSFs within the system but have limited individual driving power to influence or disrupt it [79]. The CSFs effective vendor selection (C2), stakeholder management (C6), clear AI strategy (C7), and long-term value (C10) are placed in this quadrant. This indicates that these CSFs are more reactive and reliant on the success of other foundational CSFs. Effective vendor selection (C2) depends on clear needs assessment (C1) and a thorough understanding of AI capabilities (C8). Without these, vendor selection becomes challenging, making it reactive rather than proactive. Stakeholder management (C6) depends on clear communication of needs (C1) and alignment of AI strategy (C7). It ensures support and cooperation but cannot drive the procurement process alone, being shaped by other established strategies. Clear AI strategy (C7) hinges on understanding organizational needs and goals (C1) and comprehending AI capabilities (C8). While crucial for aligning AI initiatives with organizational objectives, it remains reactive to foundational elements of procurement planning. Focusing on long-term value (C10) ensures the longevity and relevance of AI investments. It relies on initial strategic decisions (C7) and effective vendor management (C2), emphasizing future-proofing but lacking immediate driving power. The third quadrant in the MICMAC analysis contains linkage CSFs, which have both high driving and high dependence powers. These factors are critical as they influence and are influenced by many other CSFs, making them key leverage points in the procurement process that require careful management and coordination. In this study, the CSFs project and change management (C4), compliance with regulations (C5), understanding of AI capabilities (C8), and data governance (C9) fall into this category. They influence and are influenced by multiple other factors, highlighting their critical role. Effective management of these CSFs ensures smooth project execution, regulatory compliance, realistic expectations, and robust data governance, requiring a balanced, integrated approach for successful AI procurement in the UAE public sector. The fourth quadrant includes the CSFs clear needs assessment (C1) and strong contract negotiation (C3), which have strong driving power and less dependence power. Therefore, when acquiring AI systems and services, they should be the first and most important CSFs to consider, since they have the potential to be the primary cause of other CSFs.

4.1. Theoretical Contribution

This study offers several theoretical advances to the body of knowledge in the field of procurement in general and in CSFs for the procurement of AI systems and services in the public sector of the United Arab Emirates in particular. First, this is the first study of its type to examine and elucidate the critical success factors for the public sector in the United Arab Emirates to procure AI systems and services. Second, this study uses ISM and MICMAC analysis to create contextual links between the identified CSFs of AI systems and services procurement in public sectors. To date, no research study has used methods like ISM and MICMAC to investigate the interactions between CSFs of AI system procurement. The driving and dependence strengths of each CSF were assessed using data from multiple experts, categorizing them into one of the four groups shown in the MICMAC diagram, based on their levels of influence and dependence. Ultimately, the division of CSFs into levels and their interconnections across various hierarchies in the proposed ISM model would aid researchers in understanding the levels and interactions among these CSFs. By including the methodologies of ISM and MICMAC, this research provides a significant methodological contribution and offers researchers insights into the interrelationships across CSFs and different levels. It is anticipated that some of the important factors produced by the ISM model may be empirically tested in the future by other academics.

4.2. Managerial Implication

The study has significant implications for managers by offering information on CSFs for AI systems and services procurement that managers need to consider to effectively acquire AI systems or services for UAE public sectors. Specifically, this study demonstrates how well-defined requirements and needs assessments are critical to the success of procurement efforts for an AI system or service. To build the firm’s knowledge base and help employees understand the new technology and its potential benefits, managers should take these CSFs into account. They should conduct knowledge transfer sessions with all stakeholders and implement employee training programs. Another important driving CSF is strong contract negotiation and management, which helps mitigate the risk of vendor lock-in and other risks associated with transactions between AI providers and public sector entities. Public sector officials should also note that the long-term value of AI solutions depends on various factors, and appropriately addressing them will ascertain the sustainability of the system and service.

5. Conclusions

The CSFs for successfully procuring AI services and systems in the UAE public sectors are examined in this research. It has been determined that there is a contextual link between CSFs that impacts the procurement of AI. The ISM presents a hierarchical model that illustrates optimal behaviors for each CSF in procuring AI systems and services in public sectors. It also investigates the dynamic interactions and transitive links between the CSFs. This study identifies CSFs that must be addressed to effectively procure AI. Using ISM, a structural model was created that showed each CSF’s respective driving and dependence powers. According to the MICMAC analysis, these CSFs influence the choice to purchase the AI systems, while additional CSFs serve as linking variables in the procedure. A total of ten CSFs were identified from the literature and subsequently validated by interviews with procurement experts from UAE public sectors. The most important CSF was found to be clear needs assessment and requirement definition, which drives all other CSFs. Focus on long-term value and sustainability is recognized as a CSF at the top level of the hierarchy and has minimal influence on procurement decisions. While AI may assist in maximizing resource use, eliminating needless waste, and promoting long-term sustainability in the public sector, technology can also contribute to environmental degradation by using high energy levels in artificial intelligence systems, notably through data centers and processing.

This study contributes to the body of knowledge by introducing a new set of critical success factors, their contextual relationship using ISM methodology, and the driving and dependence power of each CSF. The insights drawn from this study can help managers in the public sector to make well-informed procurement decisions while dealing with the procurement of AI systems and services. A limitation of this study is that a set of CSFs was developed based on the opinions of a limited number of experts. In the future, exploratory and confirmatory factor analysis can be conducted to further validate these findings.