Exploring Critical Success Factors of AI-Integrated Digital Twins on Saudi Construction Project Deliverables: A PLS-SEM Approach

Abstract

Artificial intelligence-enhanced digital twins are widely acknowledged as effective instruments for facilitating digital transformation in the building industry. Nonetheless, their implementation remains uneven, with little knowledge regarding the organizational conditions that convert these technologies into enhanced project outcomes. This study investigates the critical success factors (CSFs) that shape the effectiveness of AI-integrated digital twins in Saudi Arabia’s construction industry. A hierarchical structural equation model was developed to capture three dimensions of CSFs, including human-centric, technological, and governance-related, and to evaluate their impact on project deliverables, including time, cost, resource utilization, quality, and risk. Data from a survey of 120 industry professionals were assessed utilizing a PLS-SEM approach, incorporating rigorous measurement and structural assessments. Results indicate that technology and infrastructural factors have the most significant impact on critical success factors, followed by governance and human-related enablers. Consequently, CSFs substantially forecast project outcomes, mediating the influences of all three domains. These findings underscore the importance of investing in data quality, scalable infrastructure, and governance frameworks, complemented by workforce training and incentives, to realize the benefits of AI-enabled digital transformations fully. The study presents a validated paradigm that elucidates how enabling conditions enhance performance improvements, providing practical direction for industry players and policymakers.

1. Introduction

Digital twins (DTs), defined as continuously synchronized virtual representations of physical assets and processes, are increasingly recognized as a cornerstone of construction’s digital transformation. When integrated with artificial intelligence (AI), DTs extend beyond descriptive modeling to deliver predictive and prescriptive capabilities across the project lifecycle [1,2,3,4]. In the construction industry, a Digital Twin (DT) is defined as an evolving virtual version of a physical asset, process, or system that sustains a dynamic connection via real-time data exchange. In contrast to Building Information Modeling (BIM), which provides a static design-focused model, Digital Twins (DTs) DTs extend beyond visualization to enable monitoring, prediction, and optimization throughtout the project lifecycle. A digital twin typically integrates three components: the physical asset, its virtual counterpart, and the constant data flow between them, enabled by sensors, IoT platforms, and cloud-based systems. When augmented with artificial intelligence, digital twins transition from descriptive models to predictive and prescriptive tools capable of predicting project performance, recognizing risks, and recommending corrective actions. This approach enables decision-makers to transition from reactive to proactive management, thereby differentiating digital twins from other technologies and elucidating their function as transformative decision-support systems in construction [4,5].

The academic studies and industry reports offer valuable but fragmented evidence. For example, studies often examine discrete enablers, skills, infrastructure, data governance, and leadership, without integrating them into a coherent framework that links critical success factors (CSFs) to measurable project deliverables. Therefore, to address this fragmentation, the present study advances a higher-order CSFs construct formed by human-centric factors (HCFs), technology/infrastructure factors (TIFs), and governance/standards factors (GSFs), and positions project deliverables as the downstream outcome, as shown in Figure 1. Their integration enables collaboration, structured adoption, and scalable systems, creating the central synergy zone that drives successful digital transformation in AI-integrated digital twin projects. This integrative perspective enables a simultaneous examination of how people, technology, and governance interact to support the realization of DT and AI value in construction projects [6,7].

This paper has three objectives: theoretically, it develops a reflective–hierarchical model of CSFs to explain variation in project deliverables; methodologically, it applies a PLS-SEM approach with robust validity and reliability tests to estimate structural relationships; and practically, it identifies priority levers, training, data quality, scalable infrastructure, and governance, to guide industry and policymakers in scaling AI-enabled digital twins [3,8,9,10]. These aims are operationalized through the following research questions: RQ1: Which CSFs most strongly form the higher-order CSFs construct? RQ2: To what extent do CSFs improve project deliverables (time, cost, resources, quality, and safety-operationalized via risk-management indicators)?, RQ3: Do the effects vary across respondent demographics (e.g., role and experience)?

This paper is structured as follows. Section 2 reviews the literature on AI-enabled digital twins and critical success factors, thus developing the conceptual framework for the study. Section 3 delineates the research model and hypotheses, including the approach encompassing survey design, data collection, and model formulation. Section 4 presents the empirical findings, including measurement quality, structural relationships, and mediation effects. Section 5 examines the implications of these findings for theory and practice, along with the study’s limitations and opportunities for future research. Section 6 concludes by summarizing the main contributions and offering recommendations for policymakers, industry practitioners, and researchers.

2. Literature Review

2.1. AI-Enabled Digital Twins in the Construction Industry

Since 2016, the integration of Digital Twin technology has progressed significantly within the construction industry, reflecting a notable shift toward digital transformation. As a result, various scholars have provided diverse definitions of digital twins in relation to the built environment, highlighting the dynamic nature of the concept. Despite this diversity, there is no commonly accepted, well-defined definition, which highlights ongoing debate and progress in this field [11]. For instance, several academics have defined digital twins as digital representations of tangible assets, such as buildings or infrastructure components [12]. These representations are realized through advanced digital technologies, including sensors, communication networks, and three-dimensional modeling systems [13]. To illustrate the variety of views, Shahzad et al. [11] examine ten distinct definitions of digital twins identified in academic literature. They emphasize that the digital twin methodology enables real-time data integration and supports bidirectional communication, thereby ensuring that the virtual model remains an accurate and dynamic replica of its physical counterpart.

In the domain of digital twins (DT), it core objectives include data acquisition, modeling, and application [14]. Achieving these aims relies on four main technological pillars: the Internet of Things (IoT), Artificial Intelligence (AI), Extended Reality (XR), and cloud computing. These pillars facilitate the collection of real-time data, insightful analysis, and the digital representation of physical assets. Through comprehensive digital twin applications, these technologies improve organizational effectiveness, creativity, and decision-making [15]. It could be said that a digital twin (DT) is a persistently synchronized virtual representation of a physical asset, process, or system. DTs become predictive and prescriptive when AI techniques (e.g., machine learning, optimization, anomaly detection) are integrated into them. This enables the DTs to forecast states, recommend actions, and self-optimize throughout the project lifecycle [16].

In the construction industry, the integration of AI-enabled DT, which is poised to bridge the long-noted gap between the technology’s promise and its uneven adoption by translating rich, real-time data into scheduling [3,9], cost, quality, and risk decisions at operational speed [3]. For instance, research and industry practice show that AI-enabled DTs can reduce time/schedule [16], lower costs [15] improve resource utilization [16], strengthen quality management [16], and enhance risk management [3,4,16]. Within the construction sector, a key benefit of integrating AI-enabled digital twin (DT) technology lies in its capacity to mitigate project delays and reduce resource wastage, thereby streamlining construction workflows and enhancing overall efficiency [16].

2.2. Comparative Analysis and Research Gap

Table 1. Comparative Analysis of Models.

Framework	Core Focus	Constructs Considered	Scope and Limitations	Contribution to Outcomes	References
BIM Integration Models	Information management, design coordination	Emphasis on data sharing, interoperability, collaboration	Primarily focused on technological maturity, with limited attention to governance and workforce readiness.	Supports project coordination and design efficiency but less on holistic deliverables	[11,12,15,16]
IoT–Digital Twin Models	Real-time data capture, monitoring, and control	Sensors, IoT platforms, cloud computing	Strong on technological infrastructure, but fragmented on governance and human enablers	Enhance monitoring and automation; limited in addressing organizational adoption factors	[13,17,18,19,20]
AI–DT Application Models	Predictive and prescriptive analytics in construction	Machine learning, simulation, anomaly detection	Focus on technical capability; governance and workforce issues are often overlooked.	Improve forecasting, cost, and risk management, but rarely embedded in a unified CSF framework.	[3,16,21]
Present paper: Higher-Order CSFs Model (AI-enabled DTs)	Integrated success conditions for adoption and performance	Human-centric factors (HCFs), Technology/Infrastructure (TIFs), Governance/Standards (GSFs)	Explicitly unifies human, technological, and governance domains into a hierarchical construct	Validated PLS-SEM model showing CSFs significantly predict project deliverables (time, cost, resources, quality, risk)	This paper

compares existing models in digital construction management with the higher-order CSFs framework developed in this paper. While BIM integration models emphasize information sharing and design coordination, they are largely confined to technological maturity and overlook governance or workforce readiness. IoT–digital twin frameworks advance real-time monitoring through sensors and connected platforms; however, they tend to remain fragmented in addressing organizational adoption factors. Similarly, AI–DT applications demonstrate predictive and prescriptive capabilities, yet they often treat technology in isolation and neglect governance and human-centric enablers. In contrast, the present study introduces an integrated hierarchical CSFs model that unifies human, technological, and governance dimensions into a single construct, empirically validated using PLS-SEM. By demonstrating how these combined factors directly predict project deliverables such as cost, schedule, quality, and risk, the model addresses gaps in prior approaches and provides a more holistic framework for understanding the success of AI-enabled digital twin adoption in construction.

The review of existing frameworks highlights several critical research gaps that this study seeks to address:

▪

Isolated focus in prior models: BIM, IoT–DT, and AI-driven frameworks often emphasize a single dimension (e.g., technology, monitoring, or analytics) without integrating human, technological, and governance factors.

▪

Limited attention to human-centric enablers: Workforce readiness, training, and incentive structures are frequently overlooked, despite their central role in sustaining adoption.

▪

Lack of governance integration: Few studies incorporate standards, leadership commitment, and ethical AI practices as formal determinants of digital twin success.

▪

Absence of validated hierarchical models: Previous work rarely employs higher-order constructs that capture the combined influence of multiple enablers on project deliverables.

▪

Narrow link to outcomes: Existing studies often demonstrate technical potential; however, they stop short of empirically linking success factors to measurable deliverables such as cost, schedule, quality, and risk.

By addressing these gaps, the present research develops and validates a higher-order CSFs model that unifies human, technological, and governance conditions into a single predictive framework. This integrated approach not only clarifies the mechanisms through which AI-enabled digital twins influence project performance but also provides actionable guidance for policymakers, practitioners, and researchers seeking to advance digital transformation in construction.

2.3. Critical Success Factors (CSFs) Taxonomy

To fully harness the AI-powered DT benefits, it is essential to identify and understand the critical success factors that enable the effective adoption and sustainable operation of this technology within the complex and constantly evolving construction industry. According to earlier research, many of these elements were examined independently in relation to a particular scope or in different sectors, e.g., health, education, and manufacturing. However, understanding of these factors in the construction industry sector, specifically for nations with ambitions and the ability to flourish in this field, is rare. For this reason, a thorough revision was carried out; as a result, three primary categories emerged: governance/standards, technology/infrastructure, and human-centric. The fact that such categories are quite uniform across different new technology adoptions, e.g., BIM. However, due to the need for new and varied digital technologies and the rapid growth of their applications, new and sensitive factors have arisen, which will be covered in this study.

2.3.1. Human-Centric Factors (HCFs)

This human-centric dimension encompasses the essential people-related conditions that enable the successful implementation of DT augmented-AI in construction projects. This factor includes training and education programs, incentives, and rewards, as well as the readiness of a skilled workforce, as shown in Table 2. It is significant to note that all the aforementioned factors have been identified and extensively examined in the literature by researchers and theorists to understand the obstacles and, in turn, to determine the enabling factors for the successful adoption of new technology in various sectors.

In terms of the training and education programs factor, according to the literature, it is considered one of the most critical success factors for the effective deployment of AI-enabled DT. According to their findings, professionals need to possess skills in data analytics, building information modeling (BIM), machine learning (ML), and artificial intelligence (AI) [4,11,16,22,23,24]. This is because such training enhances technical proficiency, improves understanding of AI capabilities, and ensures that professionals at all levels possess the digital literacy necessary to communicate effectively with and comprehend AI systems. Previous research has indicated fewer difficulties at higher levels of training and learning, and some consider the lack of adequate training and education a significant challenge to the successful implementation of AI-powered DT, e.g., [3]. This is because insufficient training and education lead to difficulties in using technology effectively, and some academics have even connected it directly to the emergence of the resistance to change phenomenon [25].

In terms of the incentives and rewards factor, there is still widespread agreement in the literature, especially among DT studies, e.g., [11], and AI-powered DT studies, e.g., [3,24,26,27]. This factor plays a major role in the effective adoption of contemporary technology. For instance, numerous research studies and theories, such as the cross-profession theory and the team theory [28], have demonstrated a strong relationship between improved employee performance and the presence of an incentive and reward system in the workplace. This, in turn, led employees to adopt new technology instead of actively avoiding or resisting it [25].

Besides the factors mentioned above, the skilled workforce readiness remains a cornerstone for successful adoption. This includes technical proficiency, Digital literacy, adaptability, commitment, and risk management skills [4,24,27,29,30,31]. Technical proficiency refers to the ability of practitioners to apply digital twin systems and their facilitating technologies, such as BIM, IoT platforms, and cloud-based collaboration tools. It entails expertise in artificial intelligence (AI) applications, including predictive analytics and machine learning algorithms, as well as the capacity for managing big data and adhering to ethical standards in data use. Furthermore, digital literacy demonstrates the workforce’s capacity to handle and utilize contemporary software, digital systems, and visualization tools. The adaptability and commitment of professionals to learn again, embrace organizational change, and comply with new data governance frameworks. The risk management skills pointer: A competent and digitally literate workforce minimizes implementation delays, reduces the risk of errors, and improves the consistency of decision-making processes. Hence, the readiness of the skilled workforce is such that investments in governance structures and digital technologies are supported by professionals who are adequately prepared, motivated, and in a position to propel the fruitful accomplishment of AI-integrated digital twin systems within construction [11,29].

Overall, human-related factors, such as training and education, incentives and rewards, and the availability of a competent workforce, are essential conditions for the successful adoption of AI-enabled digital twins in construction. Together, these elements ensure that technology investments and governance are supplemented by a skilled, motivated, and adaptable workforce sufficient to support efficient and long-term implementation.

Table 2. Human-centric factors for AI-enabled (DT) implementation in the construction industry.

Code	Factors	References
HCF	Human-centric factors
HCF1	Training and Education (e.g., Upskilling in AI, ML, BIM, and data analytics).	[4,11,16,22,23,24,32]
HCF2	Incentives and Rewards (Motivation for adoption).	[3,11,24,26,27,33]
HCF3	Skilled Workforce Readiness (e.g., BIM proficiency, AI/ML literacy, data analytics competence).	[4,24,27,29,30,31,32]

Table 2. Human-centric factors for AI-enabled (DT) implementation in the construction industry.

Code	Factors	References
HCF	Human-centric factors
HCF1	Training and Education (e.g., Upskilling in AI, ML, BIM, and data analytics).	[4,11,16,22,23,24,32]
HCF2	Incentives and Rewards (Motivation for adoption).	[3,11,24,26,27,33]
HCF3	Skilled Workforce Readiness (e.g., BIM proficiency, AI/ML literacy, data analytics competence).	[4,24,27,29,30,31,32]

2.3.2. Technology and Infrastructure-Centric Factors (TIFs)

Technology and infrastructure-centric factors establish the digital foundation necessary for the effective deployment of AI-enabled Digital Twin integration systems in construction projects. This includes data quality and management, IT Infrastructure Availability, Scalable Infrastructure, Vendor Support, and Cybersecurity, as shown in Table 3.

In terms of data quality and management, this factor encompasses the accuracy, consistency, completeness, and timeliness of project information updates [34]. Without reliable and timely data streams, AI-enabled digital twins cannot generate valid predictions or effectively support informed decision-making [14,15,16,17,18,30,31,32,35,36]. The availability and robustness of IT infrastructure represent critical enablers for the adoption of AI-enabled digital twins, encompassing hardware, software, and network capabilities. A resilient and well-integrated infrastructure ensures seamless data exchange, supports high computational demands, and provides the reliability necessary for real-time modeling, risk control, and management, as well as timely decision support in construction projects [15,18,19,29,37,38].

Along with this, scalable platforms, such as cloud and edge computing, enable real-time analytics and ensure that AI-enabled digital twin integration solutions can handle the increasing amount and velocity of construction project-related data [19,39,40]. Simultaneously, several studies have demonstrated the significance of vendor assistance, e.g., [11,19,20,35,41]. This reduces the likelihood of technical interruptions during project execution by ensuring system compatibility, integration, and ongoing maintenance [42]. In addition, the Cybersecurity and privacy factor is highly emphasized in the literature. This is because construction projects typically involve sensitive financial and operational data, which must be protected from unauthorized access, intrusions, and cyber-attacks. Secure storage systems, encryption methods, and access controls guarantee data integrity whilst covering compliance and contractual requirements [16,19,20,37,38].

In summary, technology and infrastructure-related aspects, such as data quality, IT and scalable infrastructure, vendor support, and cybersecurity, provide the digital backbone for the right adoption of AI-supported digital twins in construction activities. These aspects determine to what extent AI-enabled DTs can consume quality data, run large-scale computing, offer cross-platform compatibility, and uphold security levels. Strong TIFs contribute to the viability of DT-enabled workflows and have a direct influence on enhancing project performance outcomes by increasing speed, scalability, and overall system reliability.

Table 3. Technology and Infrastructure-centric factors for AI-enabled (DT) implementation in the construction industry.

Code	Factors	References
TIF
TIF1	Data Quality and Management (Accuracy, consistency, and real-time updates).	[14,15,16,17,18,30,31,32,35,36]
TIF2	IT Infrastructure Availability (Hardware, software, and network reliability).	[15,18,19,29,37,38]
TIF3	Scalable Infrastructure (Cloud/edge computing for real-time analytics).	[19,39,40,43,44]
TIF4	Vendor Support (Tech partnerships, system compatibility).	[11,19,20,35,41]
TIF5	Cybersecurity: Data Security (Encryption, access controls, GDPR compliance) and Privacy (Secure storage, access control, and breach prevention).	[16,19,20,32,37,38]

Table 3. Technology and Infrastructure-centric factors for AI-enabled (DT) implementation in the construction industry.

Code	Factors	References
TIF
TIF1	Data Quality and Management (Accuracy, consistency, and real-time updates).	[14,15,16,17,18,30,31,32,35,36]
TIF2	IT Infrastructure Availability (Hardware, software, and network reliability).	[15,18,19,29,37,38]
TIF3	Scalable Infrastructure (Cloud/edge computing for real-time analytics).	[19,39,40,43,44]
TIF4	Vendor Support (Tech partnerships, system compatibility).	[11,19,20,35,41]
TIF5	Cybersecurity: Data Security (Encryption, access controls, GDPR compliance) and Privacy (Secure storage, access control, and breach prevention).	[16,19,20,32,37,38]

2.3.3. Governance and Standards-Centric Factors (GSFs)

Literature continues to debate the crucial role that standards and governance characteristics play in the adoption of emerging technologies, particularly those that rely on specific frameworks, systems, and tools. These governance mechanisms offer strong institutional and legal frameworks that facilitate the successful and long-term integration of artificial intelligence and digital twins. As indicated in Table 4, these consist of leadership commitment, ethical use of AI, adherence to established national and international rules, and continuous auditing and data integrity.

With respect to established national and international guidelines, research studies emphasize developing and adhering to standard policies and processes in project implementation, such as ISO 19650 for BIM and ISO 23247 for Digital Twins. The use of such standards enables consistency, interoperability, as well as adherence to regulations across various construction projects [3,15,18,29,45,46]. Equally pertinent is the adoption of ethical and transparent practices in the instance of AI-enabled digital twins in the construction industry. These include bias mitigation, explainability, accountability, and fairness, which are essential for organizational integrity and to develop stakeholder trust, ensuring regulatory compliance, and the responsible use of sensitive project and operational data [3,35,47,48]. Furthermore, continual audit and systematic monitoring of data integrity are crucial for validating digital twin results and enabling the early detection of anomalies or system failures. Combined, these practices reinforce organizational resilience, support proactive risk management, and ensure that AI-enabled digital twin projects fulfill both technical performance requirements and broader needs for transparency, accountability, and trustworthiness [4,18,20,29,49,50]. In addition, leadership commitment at the organizational and governmental levels serves a key enabling role in enhancing the efficacy of governance and standards. Top management’s active involvement and interest are evident in the availability of the required resources, equipment, and incentives, as well as the establishment of strategic implementation priorities. In exchange, this will enhance stakeholder organizational alignment and promote the widespread adoption of AI-powered digital twin projects in construction [4,11,29,51,52,53].

In conclusion, clear governance structures establish defined roles, responsibilities, and workflows, reducing ambiguity and minimizing resistance to organizational change. These measures enable pilot projects to scale into sustainable, value-generating systems that enhance efficiency, accountability, and trust.

Table 4. Governance and Standard-centric factors for AI-enabled (DT) implementation in the construction industry.

Code	Factors	References
GSF	Governance and Standard-centric factors
GSF1	National guidelines/Codes of practice, Adherence to ISO 19650 for BIM, ISO 23247 for DT, local laws, and Project frameworks.	[3,15,18,29,45,46]
GSF2	Ethical use, and ethical AI Use (Bias mitigation, transparency, and accountability).	[3,6,35,47,48,54]
GSF3	Continuous Data Integrity and Auditing (Regular checks on data integrity and system performance).	[4,18,20,29,49,50]
GSF4	Leadership commitment (Top management support, C-level sponsorship).	[4,11,29,51,52]

Table 4. Governance and Standard-centric factors for AI-enabled (DT) implementation in the construction industry.

Code	Factors	References
GSF	Governance and Standard-centric factors
GSF1	National guidelines/Codes of practice, Adherence to ISO 19650 for BIM, ISO 23247 for DT, local laws, and Project frameworks.	[3,15,18,29,45,46]
GSF2	Ethical use, and ethical AI Use (Bias mitigation, transparency, and accountability).	[3,6,35,47,48,54]
GSF3	Continuous Data Integrity and Auditing (Regular checks on data integrity and system performance).	[4,18,20,29,49,50]
GSF4	Leadership commitment (Top management support, C-level sponsorship).	[4,11,29,51,52]

Collectively, these three above dimensions of CSFs provide a comprehensive taxonomy for understanding the required factors needed for the successful adoption of AI-enabled DTs.

2.4. Project Deliverables (PDs)

According to the literature, integrating Artificial Intelligence with Digital Twins (DTs) can significantly enhance the delivery of construction projects. This is achieved by combining a continuously updated virtual model with data-driven prediction and control. In practice, DTs augmented AI compress time/schedule through predictive planning and automated constraint resolution; reduce cost via optimized logistics and rework avoidance; raise resource efficiency through dynamic allocation and utilization tracking; strengthen quality management with continuous assurance and anomaly detection; and enhance risk management by enabling probabilistic simulation, early-warning systems, and scenario-based contingency planning [21,55,56]. In this framework, these outcomes are formalized as PD1–PD5: time/schedule, cost, resources, quality management, and risk management, and are modeled as the downstream outcome construct influenced by critical success factors, which our study explicitly investigates as in

Table 5. Impact of AI-Integrated Digital Twins on Project Deliverables (PD1–PD5).

Code	Impact of Integrating AI with Digital Twins on Project Deliverables	References
PD1	Time/schedule	[11,55,57,58,59,60,61]
PD2	Cost	[14,15,21,55,56,62,63]
PD3	Resources	[3,16,64,65,66]
PD4	Quality Management	[3,12,15,67,68,69]
PD5	Risk Management	[3,4,12,14,70,71,72]

.

2.5. Conceptual Model and Hypotheses

This paper advances a reflective–reflective hierarchical components model in which Critical Success Factors (CSFs) operate as a second-order construct formed by three first-order dimensions, human-centric factors (HCFs), technology/infrastructure factors (TIFs), and governance/standards factors (GSFs), and predicts project deliverables (PDs) (time, cost, resources, quality, risk) as the outcome. Accordingly, the hypotheses posit positive effects of HCFs → CSFs (H1), TIFs → CSFs (H2), and GSF → CSFs (H3), with CSFs → PDs (H4); optional extensions specify indirect effects from HCFs, TIFs, and GSFs to PDs via CSFs (H5–H7) and a relative-effects conjecture that TIFs exert the most substantial contribution to CSFs (H8). The conceptual framework and all hypothesized paths are summarized in Figure 2.

Based on this review, the paper presents a higher-order framework that combines the human, technological, and governance aspects into a single set of critical success factors. This framework positions project deliverables as the most significant outcomes, thereby linking organizational and technical enablers to tangible performance improvements. The following section discusses the research methodology used to evaluate these associations in real life. This includes developing a survey instrument, sampling strategy, and using PLS-SEM model for validation and hypothesis testing.

3. Methodology

3.1. Questionnaire Survey Development and Data Collection

The survey instrument was deductively derived from the literature-informed construct taxonomy, HCFs (training/education, incentives, workforce readiness), TIFs (data quality/management, IT availability, scalability, vendor support, cybersecurity/privacy), and GSFs (standards, ethics, leadership, data integrity/auditing), together with the outcome bundle PD1–PD5. Items were organized on a Likert-type scale, and content validity was established through expert review and pilot administration. In line with the methods plan, psychometric diagnostics were specified and applied (KMO, Bartlett’s sphericity), followed by reliability, convergent and discriminant validity, and collinearity checks to verify the adequacy of the measurement model before structural estimation. A cross-sectional questionnaire was administered to construction professionals under procedures that covered sampling and administration.

3.2. Targeted Participants and Sampling Technique

The target population was defined as construction professionals spanning role categories and experience levels; demographic attributes and DT/AI knowledge were recorded and are reported in the results. A target recruitment approach was employed to distribute the survey in accordance with standard procedures. The sampling frame targeted construction professionals in the Kingdom of Saudi Arabia during 2025, and a final sample of 120 valid responses was achieved.

3.3. Development of PLS-SEM Model

The PLS-SEM model was conceptualized with a higher-order reflective construct for Critical Success Factors (CSFs) formed by three first-order dimensions (HCFs, TIFs, GSF), and Project Deliverables (PDs) specified as the outcome construct. A hierarchical components model was estimated in SmartPLS using the PLS algorithm with bias-corrected bootstrapping (5000 resamples) to obtain path estimates and significance levels as in Figure 3. Evaluation followed the staged procedure depicted in the study’s “PLS-SEM Components” framework: first-order measurement assessment, second-order model assessment, and path (structural) model assessment. Reporting standards included R² and f² [6,29].

A cross-sectional survey research design was adopted with construction professionals as the target population. The instrument was developed to capture HCF1–3, TIF1–5, GSF1–4, and PD1–5 using a Likert scale; content validity was established through expert review and a pilot administration. Sampling and data-collection procedures were implemented under an ethics/consent protocol, and participants reflected diverse roles, experience levels, and cities. Data were coded and screened for missingness and outliers; common method bias was assessed via Harman’s one-factor test and full collinearity VIFs. The structural model was specified as a higher-order reflective–reflective hierarchy and was estimated using the two-stage approach (first estimating scores for the lower-order constructs and then using those scores to assess the higher-order CSFs). Estimation was conducted in SmartPLS with the PLS algorithm, and 5000 bias-corrected and accelerated bootstrap resamples, applying conventional significance criteria. Measurement quality was evaluated using outer loadings (≥0.70), indicator reliability, CR (≥0.70), AVE (≥0.50), discriminant validity (Fornell–Larcker and HTMT < 0.85/0.90), and VIF (<3.3). Structural results were reported as path coefficients (β, t, p), R², and f² (0.02/0.15/0.35).

Convergent validity was examined via outer loadings and construct reliability/validity criteria (CR ≥ 0.70; AVE ≥ 0.50). Composite reliability and AVE met thresholds across constructs (e.g., CSFs CR = 0.947, AVE = 0.598; TIFs CR = 0.946, AVE = 0.777; GSFs CR = 0.950, AVE = 0.827; HCFs CR = 0.938, AVE = 0.835; PDs CR = 0.963, AVE = 0.839). Discriminant validity was supported by HTMT values below conventional cutoffs (e.g., GSF↔CSFs = 0.842; TIFs↔CSFs = 0.843) and by the Fornell–Larcker criterion, with square roots of AVE on the diagonal (e.g., CSFs = 0.77; GSF = 0.91; HCFs = 0.91; PDs = 0.92; TIFs = 0.88) exceeding inter-construct correlations.

Collinearity was checked using indicator-level VIFs (approx. 2.3–4.9 across items), which are acceptable in many PLS applications and were reported alongside the structural results. The hypothesized paths were estimated and found to be positive: CSFs → PDs (β = 0.457), TIFs → CSFs (β = 0.471), GSF → CSFs (β = 0.414), and HCFs → CSFs (β = 0.276). Mediation via CSFs was evidenced by significant total indirect effects to PDs (e.g., TIFs → PDs = 0.215, GSF → PDs = 0.189, HCFs → PDs = 0.126; all p < 0.001, with 95% BCa confidence intervals reported). R² values were evaluated as per the predefined criteria and are presented with the path model.

4. Results

4.1. Participant’s Demography

The sample reflects a broad cross-section of the construction workforce and generally mid-to-senior experience profiles. By role, architects constitute the largest segment (28.3%), followed by project managers (19.8%), engineers (16.0%), consultants (15.1%), other roles (14.2%), and contractors (6.6%). Experience is concentrated at 11–20 years (36.8%) and >20 years (28.3%), with smaller shares reporting ≤5 years (17.9%) and 6–10 years (17.0%), as shown in Figure 4. Familiarity with digital twins is predominantly medium (42.5%) and high (29.2%), with low (13.2%), very high (9.4%), and very low (5.7%) minorities; familiarity with AI integration into DTs shows a similar pattern, led by medium (52.8%), then very high (19.8%), high (15.1%), low (6.6%), and very low (5.7%).

Collectively, these distributions indicate a respondent pool with diverse functional perspectives, substantial tenure, and moderate, though not universal, exposure to DT-enabled AI integration concepts, providing a suitable basis for the study’s PLS-SEM analysis of critical success factors and project outcomes.

4.2. Measurement Model

Figure 5 reports outer loadings for reflective indicators at both the first order and (via a repeated-indicator specification) the higher order CSFs level. Indicators load strongly on their intended first-order constructs, e.g., GSF1–GSF4 → GSF = 0.84–0.94, HCF1–HCF3 → HCFs = 0.897–0.942, TIF1–TIF2 → TIFs = 0.89–0.90, and PD1–PD5 → PDs = 0.87–0.95, demonstrating convergent validity and adequate indicator reliability (≥0.70). When the same indicators are assigned to the second-order CSFs construct under the hierarchical model, their loadings remain acceptable (GSF items = 0.70–0.871; HCF items = 0.70–0.712; TIF items = 0.802–0.814), indicating that the items chiefly reflect their specific dimensions.

While cohering with the overarching CSFs. The duplication of rows, therefore, reflects the hierarchical measurement approach rather than redundancy, and the consistently high values across constructs support the reliability and validity of the measurement model.

Cronbach’s alpha and the reliability coefficients (ρA, ρC) consistently exceeded the 0.70 benchmark, CSFs (α = 0.938; ρA = 0.940; ρC = 0.947), GSFs (0.930; 0.938; 0.950), HCFs (0.901; 0.902; 0.938), PDs (0.952; 0.966; 0.963), and TIFs (0.928; 0.928; 0.946). This indicates strong internal consistency in the measurement model. Average Variance Extracted also surpassed the 0.50 threshold for convergent validity in all cases, CSFs (0.598), GSF (0.827), HCFs (0.835), PDs (0.839), and TIFs (0.777), with particularly high AVEs for GSF, HCFs, and PDs, signifying that their indicators explain substantial variance in their latent constructs. The comparatively lower (yet adequate) AVE for CSFs is expected for a second-order reflective construct assembled from heterogeneous first-order dimensions. Therefore, the very high reliability for PDs (α ≈ 0.95; ρC ≈ 0.96) underscores excellent consistency, while warranting a check for potential item redundancy in future refinements. Accordingly, the evidence supports a reliable and convergently valid measurement model for subsequent structural analysis, as in

Table 6. CR (Composite Reliability) and AVE (Average Variance Extracted) per construct.

	Cronbach’s Alpha	Composite Reliability (rho_a)	Composite Reliability (rho_c)	Average Variance Extracted (AVE)
CSFs	0.938	0.94	0.947	0.598
GSF	0.93	0.938	0.95	0.827
HCFs	0.901	0.902	0.938	0.835
PDs	0.952	0.966	0.963	0.839
TIFs	0.928	0.928	0.946	0.777

.

All heterotrait–monotrait ratios fall below the study’s prespecified cut-offs (HTMT < 0.85/0.90), ranging from 0.344 (TIFs↔PDs) to 0.843 (TIFs↔CSFs), with other key pairs such as GSF↔CSFs = 0.842, HCFs↔CSFs = 0.830, and TIFs↔GSF = 0.731 also below the conservative 0.85 threshold. The relatively low values for all PD pairings (0.438–0.479) further support clear separation between the outcome and antecedent constructs. Therefore, the two highest ratios (TIFs↔CSFs and GSF↔CSFs) are close to the boundary, reflecting the hierarchical nature of CSFs, which is formed by TIFs and GSFs, and thus expected conceptual proximity. Cross-evidence from the Fornell–Larcker matrix shows strong inter-construct correlations, particularly between CSFs and TIFs (r ≈ 0.90), underscoring this proximity; nonetheless, by the HTMT criterion, the measurement model satisfies discriminant validity, with the caveat that these specific pairs warrant interpretive caution as in Figure 6.

The Fornell–Larcker matrix reports the square root of AVE (√AVE) on the diagonal and latent correlations off-diagonal; discriminant validity requires each √AVE to exceed its correlations with other constructs. In our results, √AVE values are CSFs = 0.77, GSF = 0.91, HCFs = 0.91, PDs = 0.92, and TIFs = 0.88. PDs and HCFs clearly satisfy the criterion, and GSF meets it narrowly vis-à-vis CSFs (0.91 > 0.89). By contrast, CSFs do not exceed their correlations with GSF (0.89) and TIFs (0.90), and TIFs fall slightly below their correlation with CSFs (0.90), indicating discriminant—validity tension driven by the hierarchical specification in which CSFs are formed by TIFs and GSF. This conceptual proximity is theoretically expected; accordingly, we flag it as a caveat and complement the assessment with HTMT (all < 0.85), which supports construct separability as in

Table 7. Discriminant validity (Fornell–Larcker).

	CSFs	GSF	HCFs	PDs	TIFs
CSFs	0.77
GSF	0.89	0.91
HCFs	0.75	0.54	0.91
PDs	0.46	0.42	0.45	0.92
TIFs	0.9	0.69	0.54	0.33	0.88

.

4.3. Structural Model

Figure 7 summarizes the hierarchical PLS-SEM in which CSFs (second-order) are formed by HCFs, TIFs, and GSF, and predict PDs. The measurement model shows strong reflective loadings: HCFs items = 0.897–0.942, TIFs = 0.839–0.910, GSF = 0.842–0.939, and PDs = 0.868–0.949, indicating solid convergent reliability across constructs. Structurally, TIFs → CSFs (β = 0.471) is the largest antecedent effect, followed by GSF → CSFs (β = 0.414) and HCFs → CSFs (β = 0.276). CSFs → PDs is positive and meaningful (β = 0.457), yielding R² = 0.209 for PDs, i.e., approximately 21% of the variance in project deliverables is explained by the higher-order success conditions. Overall, the diagram illustrates that robust technology/infrastructure and governance foundations are most crucial in establishing a critical success environment, which in turn enhances time, cost, resource, quality, and risk outcomes.

An R² value of 0.21 indicates that the model explains approximately one-fifth of the variance in project deliverables. However, according to established benchmarks, R² values of 0.02 are considered weak, 0.13 moderate, and 0.26 substantial [73,74]. Consequently, our reported value of 0.21 falls within the moderate explanatory power category, which is acceptable for exploratory models in complex, multicausal domains like construction project performance. Importantly, this research particularly concentrated on validating a higher-order CSFs framework as the sole predictor of project deliverables, rather than incorporating a wide range of exogenous variables (e.g., contract type, market volatility, client decision-making). Within this defined scope, the model achieves a meaningful effect size (f² ≈ 0.26), reinforcing the relevance of CSFs as significant contributors to project outcomes. While more broader models incorporating additional external factors might yield higher R² values, this staged approach ensures conceptual clarity and highlights the unique contribution of CSFs. This may, in turn, provide a robust basis for subsequent research to enhance explanatory capacity.

Table 8. Path coefficients (β, t, p, f²).

	Path Coefficients
Critical Success Factors (CSFs) -> Project Deliverables (PDs)	0.457
GSF -> Critical Success Factors (CSFs)	0.414
HCFs -> Critical Success Factors (CSFs)	0.276
TIFs -> Critical Success Factors (CSFs)	0.471

reports the structural path coefficients of the hierarchical model and shows that all hypothesized relationships are positive and substantively meaningful. The strongest antecedent of CSFs is TIFs → CSFs (β = 0.471), followed by GSF → CSFs (β = 0.414) and HCFs → CSFs (β = 0.276), thereby supporting H1–H3 and the relative-effects conjecture that technology/infrastructure exerts the most significant influence. Downstream, CSFs → PDs (β = 0.457) indicate that strengthening the successful environment is associated with sizable improvements in deliverables (time, cost, resources, quality, risk). Given that CSFs are the sole predictor of PDs in the model, the reported R² for PDs = 0.209 implies an approximate f² ≈ of 0.26, a medium effect by conventional thresholds, underscoring the practical relevance of building robust CSFs for AI-integrated digital twin outcomes.

4.4. Indirect Effects (Mediation via CSFs)

The mediation analysis confirms that GSF, HCFs, and TIFs exert significant indirect effects on project deliverables (PDs) through CSFs. Specifically, the indirect path coefficients are TIFs → PDs = 0.215 (t = 5.528, p < 0.001), GSF → PDs = 0.189 (t = 5.480, p < 0.001), and HCFs → PDs = 0.126 (t = 4.757, p < 0.001), with bootstrap confidence intervals excluding zero as in

Table 9. The indirect effects of governance (GSF), human-centric (HCFs), and technology/infrastructure (TIFs) factors on project deliverables (PDs) are mediated by critical success factors (CSFs).

	Original Sample (O)	Sample Mean (M)	Standard Deviation (STDEV)	T Statistics	p Values
GSF -> Project Deliverables (PDs)	0.189	0.192	0.035	5.480	0.000
HCFs -> Project Deliverables (PDs)	0.126	0.129	0.027	4.757	0.000
TIFs -> Project Deliverables (PDs)	0.215	0.219	0.039	5.528	0.000

. These findings demonstrate that the influence of human, technological, and governance factors on deliverables is transmitted primarily through the second-order CSFs construct, supporting the mediation hypotheses (H5–H7). Among these, TIFs show the most substantial indirect effect, reinforcing the argument that robust data, infrastructure, and cybersecurity capacities are the most critical drivers of project outcomes when filtered through the successful environment established by CSFs.

The bootstrap confidence intervals reinforce the significance of the indirect effects of governance, human-centric, and technology/infrastructure factors on project deliverables through CSFs. For GSF → PDs, the interval ranges from 0.122 to 0.258, for HCFs → PDs from 0.077 to 0.182, and for TIFs → PDs from 0.141 to 0.296, with none crossing zero. These results confirm the robustness and stability of the mediation effects, highlighting that the indirect influence of these factors on deliverables is consistently positive across resamples. Accordingly, TIFs again exhibit the most potent effect, followed by GSF and HCFs, aligning with the hypothesized relative impact of critical success factors on project outcomes as in

Table 10. Bootstrap confidence intervals for the indirect effects of GSF, HCFs, and TIFs on project deliverables (PDs) via CSFs.

	Original Sample (O)	Sample Mean (M)	2.5%	97.5%
GSF -> Project Deliverables (PDs)	0.189	0.192	0.122	0.258
HCFs -> Project Deliverables (PDs)	0.126	0.129	0.077	0.182
TIFs -> Project Deliverables (PDs)	0.215	0.219	0.141	0.296

.

4.5. Robustness/Diagnostics (Multicollinearity)

The variance inflation factor (VIF) analysis shows values ranging from 2.286 to 4.96 across all items, which are within acceptable limits for PLS-SEM, indicating no severe multicollinearity issues. Although some indicators, such as GSF1 (4.96), PD2 (4.87), and PD5 (4.87), approach the upper recommended threshold of 5, they remain below the critical level of 10, which is typically considered problematic in structural equation modeling. These results suggest that while specific indicators share a moderate degree of collinearity, the measurement model is not adversely affected, and the constructs retain sufficient discriminant explanatory power for reliable path estimation as in Figure 8.

5. Discussion

5.1. Discussion of the Model’s Results

The structural model supports all main hypotheses: technology/infrastructure (TIFs), governance/standards (GSF), and human-centric factors (HCFs). Each one strengthens the higher-order CSFs, and in turn, CSFs improve project deliverables (PDs). The largest contribution on CSFs arises from TIFs → CSFs (β = 0.471), followed by GSF → CSFs (β = 0.414) and HCFs → CSFs (β = 0.276); downstream, CSFs → PDs (β = 0.457) shows a meaningful association with time, cost, resource, quality, and risk outcomes. Mediation tests additionally demonstrate that TIFs, GSF, and HCFs affect PDs indirectly through CSFs (all indirect effects significant via bootstrapping), thereby underscoring the importance of the success conditions construct as a conduit through which capabilities are translated into performance. The model explains approximately 21% of the variance in PDs (R² ≈ 0.209), which is consistent with a multicausal performance domain in construction projects.

The dominance of TIFs and GSFs over HCFs in the formation of CSFs is theoretically aligned with the data architecture of DT–AI systems: high-quality, timely data and scalable, secure infrastructure are prerequisites for any analytics-enabled twin, while governance (e.g., standards, leadership commitment) institutionalizes adoption and reduces implementation risk. Empirically, the closeness between CSFs and their antecedent dimensions is visible in HTMT ratios (TIFs↔CSFs = 0.843; GSF↔CSFs = 0.842) and in the Fornell–Larcker matrix, where √AVE for CSFs (0.77) is lower that its correlations with TIFs (0.90) and GSF (0.89). Rather than signaling misspecification, this pattern is consistent with the hierarchical design in which CSFs are formed by these domains; it advises cautions in interpretation while endorsing the conceptual nesting of success conditions within the second-order construct.

A significant finding is the comparatively lower contribution of HCFs to CSFs. This outcome can be explained by cultural and organizational dynamics within the Saudi construction industry. First, the industry has historically prioritized technological investment and compliance with governance frameworks over workforce development. This is in line with national digital transformation strategies that stress the importance of infrastructure readiness and regulatory alignment. Second, training and upskilling programs remain emergent, with digital literacy and AI competencies unevenly distributed among practitioners. This causes a lag between technology deployment and workforce adaptation. Third, cultural aspects such as hierarchical organizational structures and reliance on top-down decision-making might decrease the perceived impact of individual or team-level capabilities, consequently reducing the significance of human factors in relation to technology and governance enablers. Collectively, these contextual conditions may illustrate the diminished impact of HCFs relative to TIFs and GSFs in this study. This finding is consistent with expectations for emergent performance outcomes that illustrate the interaction of technological, organizational, and contextual variables. Importantly, assessments of measurement quality (reliability and convergent validity) and multicollinearity diagnostics (VIF ≈ 2.3–4.9) were satisfactory, mitigating concerns that limited explanatory power reflects indicator weakness or redundancy. The results underscore the necessity to broaden the explanatory study in future research.

From a practical standpoint, the evidence highlights the importance of investments in data governance and infrastructure, encompassing accurate, real-time data streams, scalable architectures (cloud or edge), and robust cybersecurity and privacy protections, together with formal governance mechanisms like standards adherence, ethical AI protocols, and leadership support. These provide the foundation for CSFs that can reliably translate the AI-augmented digital twin potential into construction projects improvements in terms of schedule, cost, and quality performance. Human capital levers, such as targeted training, incentive structures, and readiness-building initiatives, are essential, even if their direct contribution to CSFs appears comparatively modest. For policymakers and clients, aligning with international reference standards (e.g., ISO 19650 for BIM, ISO 23247 for digital twins) and developing incentive frameworks can accelerate the diffusion and uptake of these standards.

Theoretically, the paper suggests a hierarchical perspective on the success conditions governing AI augmented DT adoption, while pointing to the value of incorporating additional explanatory layers, such as procurement strategies, stakeholder integration, and environmental volatility, alongside alternative research designs, such as longitudinal, multilevel, and objective deliverable-based approaches. Extensions via heterogeneity analyses (e.g., multi-group analysis by role or experience), and prioritization frameworks (e.g., Importance–Performance Matrix Analysis, IPMA) offer promising avenues to refine understanding of where interventions can yield the most significant performance improvement.

5.2. Research Limitations

This study provides significant insights into the critical success factors of AI-integrated digital twins in the Saudi construction sector; nevertheless, several limitations must be acknowledged.

First, the research employed a cross-sectional survey design, which restricts causal inference and prevents the observation of how critical success factors evolve over time. Consequently, subsequent research should adopt longitudinal or multi-wave research designs to effectively capture the dynamic nature of digital transformation and establish stronger causal linkages between human, technological, and governance conditions and project deliverables.

Second, the study relied solely on self-reported survey data, which introduces perceptual or social desirability biases. The absence of multiple data sources, such as audited project records, cost and scheduled documentation, or quality audit reports, limits the robustness of the findings. Incorporating such objective data in future studies would improve reliability and reduce bias.

Third, the sample was geographically and contextually limited to Saudi Arabia. While this focus provides locally grounded insights, it constrains generalizability. Expanding the sample to include larger scales and diverse international contexts would allow comparative testing and enhance external validity.

Moreover, potential sampling bias exists in this study, since the respondent pool was mainly skewed toward mid- to senior-level professionals, which underrepresents the perspectives of less experienced employees.

Fourth, the model explained around 21% of the variance in project deliverables, a moderate yet incomplete level of explanatory power. This suggests that other variables, such as procurement strategies, supply chain integration, project complexity, regulatory frameworks, and client decision-making, should be incorporated in future models to capture additional sources of variation.

Taken together, these limitations underscore the importance of extending future research along multiple dimensions: (i) triangulating survey data with objective project performance records, (ii) diversifying samples across industries, countries, and project scales, (iii) undertaking longitudinal and comparative international studies, and (iv) incorporating unobserved yet influential variables into the model. These refinements will improve validity, broaden explanatory power, and offer a more comprehensive understanding of the mechanisms underpinning the successful adoption of AI-integrated digital twins. Figure 9 depicts the fishbone diagram, illustrates these methodological, contextual, and explanatory constraints, which underscoring the need for multi-source, multilevel, and cross-context research designs to advance this field.

5.3. Future Research

Because of the above-mentioned limitations, future research on AI-integrated digital twins should adopt longitudinal and multilevel designs to capture the dynamic and evolving nature of digital transformation in construction projects. Whereas this study relied on a cross-sectional survey, tracking adoption processes over time would help establish stronger causal inferences and reveal how critical success factors shift across different project phases. Moreover, incorporating objective measures of project deliverables, such as actual cost and schedule data, quality inspections, and safety records, would complement perceptual responses and thereby mitigate the limitations inherent in self-reported data. Collectively, such methodological refinements would enhance validity and provide richer insights into how digital twin–AI systems translate into tangible performance outcomes. Beyond methodological considerations, future research should broaden the contextual and conceptual boundaries of AI–DT adoption. For example, comparative studies across countries and regions would allow testing the impact of cultural, institutional, and regulatory differences on adoption trajectories, thereby enhancing the generalizability of findings. Furthermore, integrating external factors such as procurement models, supply chain resilience, market volatility, and stakeholder collaboration into analytical frameworks would provide a more holistic view of the determinants of success. At the same time, emerging themes, including ethical AI, sustainability impacts, and integration with other Industry 5.0 technologies, merit systematic exploration.

Therefore, these directions would broaden the explanatory power of future models and generate actionable knowledge for both researchers and practitioners in the construction domain. Figure 10 illustrates future research directions for AI-integrated digital twins, emphasizing expanded research design, longitudinal studies, and the integration of external factors. Moreover, it highlights the need for comparative studies, objective measures, and the exploration of emerging themes. Moreover, future studies should conduct heterogeneity analyses (e.g., multi-group comparisons by age, experience, and job role) to identify subgroup differences in CSFs effects. Such analyses will clarify whether contextual or demographic variations influence adoption dynamics. This approach will enhance the robustness and practical applicability of the findings.

6. Conclusions

The present paper was designed to explore the critical success factors (CSFs) for the adoption of AI-enabled digital twins (DTs) in the Saudi construction industry. This objective was successfully achieved through a survey of 120 construction professionals, representing diverse roles within the Saudi construction industry. The quantitative collected data enabled the development and validation of a hierarchical structural model to explain how human, technological, and governance factors contribute to project performance in AI–DT contexts This paper conceptualized and empirically validated a hierarchical PLS-SEM model in which critical success factors (CSFs), comprising human-centric (HCFs), technology/infrastructure (TIFs), and governance/standards (GSFs) dimensions; to predict project deliverables (PDs) in AI-integrated digital-twin (DT) projects. The measurement model displayed strong reliability and convergent validity, while the structural model results confirmed all primary hypotheses. Among the CSFs dimensions, TIFs and GSF exerted the most substantial contributions to CSFs, followed by HCFs. In turn, CSFs exerted a positive and significant influence on PDs. Mediation analyses further confirmed that HCFs, TIFs, and GSF affect deliverables primarily through the CSF construct. Accordingly, the model accounted for a meaningful share of performance variance, offering theory-driven and empirically grounded insights into how enabling conditions translate into schedule, cost, resource, quality, and risk outcomes in construction projects.

In terms of the implications for knowledge and practice, the study’s findings advance the theoretical understanding of the DT augmented AI integration by providing a replicable framework that captures the hierarchical structure of success factors and their pathways to project outcomes. From a practical perspective, the results prioritize data quality/management, scalable and secure infrastructure, and formal governance mechanisms (including standards, ethical AI principles, and leadership sponsorship as first-order enablers of adoption. These are complemented by targeted workforce upskilling and incentive structures to consolidate digital transformation at the organizational level. For policymakers, alignment with reference standards (e.g., BIM/DT frameworks) and selective incentives can accelerate diffusion across the broader construction ecosystem level.

The paper’s contributions should be considered, considering its limitations. Specifically, the reliance on cross-sectional design, a single-context sampling, and self-report data introduces constraints on causal inference and generalization. In addition, conceptual proximity among hierarchical constructs warrants interpretive caution. These limitations underscore the need for longitudinal or multi-level research designs, the incorporation of objective performance indicators, and the integration of broader contextual variables, such as procurement strategies, supply chain conditions, and stakeholders’ collaboration. Exploring heterogeneity across projects and regions would further sharpen external validity and causal insights.

As a result, this paper provides a replicable instrument and an integrative framework that can guide future scholarly inquiry and managerial action. By clarifying how CSFs influence performance outcomes, it provides actionable guidance for industry, policymakers, and researchers seeking to realize the transformative value of DT-AI value realization in construction.