The Influence of AI Competency and Soft Skills on Innovative University Competency: An Integrated SEM–Artificial Neural Network (SEM–ANN) Model

Abstract

This study addresses the growing necessity to understand how artificial intelligence (AI) competency and soft skills jointly influence organizational innovation and performance in the era of digital transformation. Despite the rapid adoption of AI technologies across industries, organizations continue to face significant challenges in effectively integrating technical AI capabilities with essential human-centric soft skills such as communication, adaptability, and leadership. This gap often limits the realization of AI-driven value and sustainable competitive advantage. The primary challenge in this research area is the lack of comprehensive models that simultaneously examine AI competency and soft skills within a unified framework, particularly in emerging economies where digital maturity varies widely. Existing studies tend to focus either on technical competencies or behavioral factors in isolation, leading to fragmented insights. To address these challenges, this study proposes a novel integrated research model that examines the combined effects of AI competency and soft skills on innovation outcomes and organizational performance. The model is empirically validated using structural equation modeling (SEM), providing robust evidence of the interrelationships among key constructs. The findings reveal that both AI competency and soft skills significantly contribute to innovation capability, which in turn enhances organizational performance. The study offers important theoretical and practical implications by bridging the gap between technical and human dimensions of AI adoption, thereby providing a more holistic understanding of digital transformation success.

1. Introduction

The rapid spread of artificial intelligence (AI) is redefining how universities create knowledge, educate graduates, and provide public value. Globally, policy discussions shaped by Industry/Education 5.0 focus on human-centered, resilient, and sustainable innovation that involves human–machine collaboration. This shift positions universities as platforms for continuous capability renewal, rather than as static repositories of knowledge. In this context, AI competency interacts with a suite of soft skills, including communication, collaboration, adaptability, critical thinking, creativity, and leadership [1]. Together, these elements contribute to an institution’s innovative university competency (IUC): its ability to transform resources into new curricula, teaching methods, research outputs, partnerships, and societal solutions [2]. Thailand’s transformation in higher education is part of broader national strategies, such as Thailand 4.0 and the 20-Year National Strategy (2018–2037). These strategies aim to foster an innovation-driven economy, enhance research and development, and improve the quality of human capital. Both frameworks highlight the need for universities to become engines of creativity and technology adoption, ensuring that academic missions align with national competitiveness and social resilience. These policy structures clearly link institutional capabilities to macro-level outcomes, demonstrate the strategic importance of AI competence, and the development of soft skills among faculty, staff, and students. Since 2024–2026, policy momentum has increased significantly. The Thai Ministry of Higher Education, Science, Research, and Innovation (MHESI) has established goals for “AI workforce development,” popularly articulated through the “AI University” initiative [3]. This initiative aims to equip the majority of graduates with basic AI knowledge and to integrate AI learning into large-scale teaching and learning. Collaborative efforts seek to promote exposure to AI and enhance teacher training nationwide [4]. For university leaders, these developments present clear imperatives: defining institution-wide AI learning outcomes, investing in necessary infrastructure and governance, and supporting professional development to ensure AI augments human judgment and academic values. On a global scale, standard-setting organizations have begun to establish AI competencies for education systems. UNESCO’s competency frameworks outline progressive knowledge, skills, and attitudes for both students and teachers, including algorithmic thinking, data stewardship, and ethical and legal awareness. These frameworks serve as valuable resources for Thai universities, supporting outcomes-based curriculum mapping, cross-faculty coordination, and external benchmarking while prioritizing responsible, human-rights-based AI practices. However, having AI literacy alone is insufficient to drive institutional innovation [5]. A growing body of evidence indicates that interventions focused on soft skills enhance teamwork, communication, adaptability, and problem-solving abilities. These improvements, in turn, facilitate technology adoption. Systematic reviews across various educational levels highlight the need for intentional design and assessment in soft skills programs, noting that unstructured offerings typically do not yield lasting improvements unless they are linked to real tasks and organizational incentives. In Thailand’s “new age university,” the strategic question is not whether to teach soft skills, but rather how to design learning ecosystems, workloads, and reward systems that make these capabilities integral and cumulative throughout degree programs and staff development pathways.

Theoretically, positioning AI competency and soft skills as the twin pillars of IUC aligns with the RBV and DCT [6,7]. RBV emphasizes the importance of valuable, rare, inimitable, and non-substitutable assets; in the university context, these include faculty’s tacit pedagogical expertise, documented curricular materials, data assets, and cross-boundary partnerships. DCT focuses on the sensing, seizing, and transforming routines that enable organizations to adapt resources in turbulent environments. Recent studies link dynamic capabilities to successful digital innovation outcomes in knowledge-intensive settings. This suggests that AI competency, when integrated into adaptation routines and complemented by soft skills, can accelerate institutional learning cycles and boost innovation capacity. Emerging research links explicitly administrative dynamic capabilities to stakeholders’ awareness of and adoption of AI, aligning with the propositions outlined here.

This study is guided by a critical gap between rapid policy-driven promotion of AI in higher education and the limited empirical understanding of how AI competency, when combined with human-centered capabilities, contributes to sustainable university innovation. Existing national and institutional policies primarily emphasize infrastructure investment, curriculum modernization, and technology adoption. However, these policies often assume that technological capability alone is sufficient to enhance institutional performance, overlooking the mediating role of strategic intelligence and the enabling function of soft skills.

Addressing this gap, the present study aims to extend, rather than replace, existing policy frameworks by empirically examining how AI competency, soft-skill competency, and strategic intelligence jointly shape innovative university competency. Grounded in the Thai higher education context, the study avoids proposing new, potentially conflicting policy instruments. Instead, it seeks to identify how current policies can be more effectively operationalized through the integration of capabilities and strategic alignment. The ultimate aim of this research is to provide contextually grounded policy recommendations that help policymakers and university leaders optimize existing strategies, minimize policy fragmentation, and prevent the creation of overlapping or contradictory initiatives. By aligning technological development with human capital development and strategic foresight, the study supports more coherent, adaptive, and sustainable pathways for innovation in higher education.

2. Literature Review

AI competency refers to the ability to effectively utilize artificial intelligence technologies, interpret analytical outputs, and apply computational thinking to solve complex problems. This capability enhances research efficiency, data interpretation, and methodological rigor. Complementing this, soft skills enable researchers to translate technical insights into meaningful academic contributions and facilitate interdisciplinary engagement. Rather than operating in isolation, these competencies function synergistically: AI competency supports technical execution, whereas soft skills drive integration, innovation, and dissemination. Drawing on the KBV and DCT, this study conceptualizes both constructs as strategic capabilities that jointly enhance research performance. This integrated perspective reduces conceptual fragmentation and provides a coherent explanation of how technical and behavioral competencies interact within academic research environments.

2.1. Theoretical Foundations

In contemporary research environments, AI competency and soft skills are best understood as complementary rather than independent drivers of research performance. AI competency enables researchers to leverage data-driven tools, machine learning techniques, and intelligent systems to enhance analytical rigor and efficiency. In contrast, soft skills facilitate interpretation, collaboration, and effective dissemination of research findings. By integrating these constructs within a unified perspective, this study reduces conceptual overlap and provides a clearer explanation of their combined influence. Drawing on the S–O–R framework and the KBV, these competencies function as key stimuli that shape internal capabilities, particularly innovation capability and strategic agility, which ultimately drive research publication performance. DCT emphasizes an organization’s ability to sense opportunities, seize them, and reconfigure internal and external resources in response to rapidly changing environments. In the context of AI adoption, DCT provides a relevant lens for understanding how organizations leverage AI competency and soft skills as dynamic capabilities to adapt to technological advancements and enhance innovation performance. AI competency enables firms to identify and exploit digital opportunities, while soft skills facilitate effective coordination, communication, and strategic decision-making during transformation processes. The RBV complements DCT by highlighting that sustainable competitive advantage is derived from valuable, rare, inimitable, and non-substitutable (VRIN) resources. Within this framework, AI competency is conceptualized as a strategic technological resource that enhances an organization’s analytical and operational capabilities. At the same time, soft skills represent critical intangible assets that contribute to organizational effectiveness and long-term competitiveness. By combining technical expertise with human-centric capabilities, organizations can strengthen their resource base and improve performance outcomes. Furthermore, HCT underscores the importance of individuals’ knowledge, skills, and competencies as key drivers of productivity and organizational success [8]. From this perspective, both AI competency and soft skills are viewed as forms of human capital that can be developed through education, training, and experience. Investments in these competencies enhance employees’ ability to utilize AI technologies effectively, foster innovation, and contribute to organizational performance. Additionally, skills can be reconfigured through dynamic pivot simulations and resource-optimization exercises on platforms [9,10,11]. These AI-supported learning experiences transform traditional case-based approaches, providing students with access to real-time data, immediate feedback, and the ability to experiment with various strategic scenarios in quick succession [12,13]. The effective integration of theoretical frameworks with practical applications of AI tools ensures that students develop both a deep conceptual understanding and hands-on operational skills. Despite the widespread acknowledgment of AI’s potential to enhance the development of dynamic capabilities, research remains limited on specific pedagogical strategies and their effectiveness in improving students’ marketing performance competencies, particularly in’ how these capabilities translate into tangible entrepreneurial outcomes.

The incorporation of AI into business and management education marks a significant pedagogical shift, changing how entrepreneurship concepts are taught and learned [14,15,16]. With AI technologies, educators can create intricate simulations and personalized learning pathways, powered by advanced deep learning algorithms, that enhance students’ analytical and strategic thinking skills [17,18]. Recent research using Revised Bloom’s Taxonomy shows that generative AI tools facilitate foundational-to-intermediate learning tasks. However, they face challenges in contexts that require higher-order cognitive skills, such as critical evaluation and innovative problem-solving [19]. These pedagogical advancements go beyond traditional case-based methodologies, offering students a unique opportunity to engage in real-time market analysis and to enter virtual business environments where they can make decisions without the risks of real-world consequences. In entrepreneurship education, AI applications provide substantial pedagogical value, enabling students to analyze real-world market data, identify business opportunities, and understand complex market dynamics through hands-on experiential learning [20]. As AI technologies increasingly shape competitive advantage in modern business landscapes, equipping students with competencies in AI applications is essential to ensure their career readiness [21]. Despite growing recognition of AI’s potential in education, research remains inadequate in examining how AI-enhanced teaching methods directly is associated with the development of dynamic capabilities and marketing competencies in student entrepreneurship programs. This study aims to fill that gap by examining how integrating AI improves pedagogical effectiveness in fostering these critical competencies [22].

2.2. AI Competency

AI competency enhances researchers’ ability to process complex data and apply advanced analytical techniques, thereby contributing to innovation capability [23]. This relentless progression of AI technology permeates numerous aspects of everyday life, notably education. The surge in AI-related research is particularly pronounced in undergraduate settings within developed regions, where innovative applications are being vigorously explored and implemented [4]. These advancements aim to address a range of educational challenges and streamline teaching and learning processes. Within the educational sphere, several core types of AI emerge as particularly transformative. Machine learning (ML) is a foundational pillar, enabling systems to discern patterns across large-scale datasets and to enhance performance through experiential learning, all without explicit programming. NLP also plays a crucial role, enabling machines to understand, interpret, and even generate human language, underpinning applications such as chatbots, virtual assistants, and language translation services [18]. A revolutionary newcomer to the field, generative AI leverages computational models to generate original, contextually relevant content by drawing on insights from large-scale datasets [13]. Numerous studies demonstrate the increasing integration of AI across diverse educational domains, including language acquisition, engineering, and mathematics [21,22]. This integration often manifests through adaptive learning systems, intelligent tutoring frameworks, automated assessment platforms, predictive analytics, educational robotics, personalized content delivery, research support tools, and efficient administrative solutions. Moreover, AI plays a vital role in enhancing administrative efficiency by automating repetitive tasks such as grading, scheduling, and attendance tracking. These technological advancements are widely regarded as invaluable for refining instructional methodologies and enabling students to achieve superior outcomes [24]. In higher education, the advent of AI has triggered sociotechnical shifts, compelling institutions to thoughtfully revisit and adapt their strategies and structures [25]. These changes can potentially revolutionize teaching methodologies and internal relationships within educational environments, ushering in innovative practices and new challenges. The integration of AI is altering traditional roles and redistributing decision-making power among institutions, educators, and students, creating a landscape ripe with opportunity yet fraught with complexity. Among the notable developments is the integration of AI-driven educational robots into classrooms. The advanced tools foster more effective instruction, heighten student engagement, and bolster academic performance [26]. They alleviate the burden of tedious administrative tasks and cultivate vibrant, interactive learning spaces that spark curiosity and enthusiasm, particularly in language acquisition [27]. Examined the impact of an AI writing assistant on English as a Foreign Language learners, finding that it significantly helped lower-proficiency students bridge the gap with their more advanced counterparts. Similarly, found that adaptive learning technologies and personalized feedback greatly enhance student engagement [28]. However, it is crucial to acknowledge that current AI tools have shortcomings, demonstrating the need to align AI advancements with foundational educational values. Building on these advancements, recent breakthroughs in generative AI are radically changing the way students engage with their studies [29]. However, this innovation raises concerns that students may rely too heavily on these tools to complete assignments without fully grasping the underlying material. The student body has enthusiastically embraced tools and acknowledges their educational value, but there remains significant concern regarding academic integrity [30]. The concerns, reporting a troubling deficit in generative AI literacy among students and pronounced disparities across disciplines and demographic groups in both confidence levels and patterns of use [31]. Faculty members’ apprehensions mirror those of students; many educators acknowledge AI’s potential to promote educational equity, particularly for students with disabilities and those from underrepresented backgrounds. However, the pervasive lack of AI literacy among students remains a key barrier. To harness the transformative power of AI in education, students must cultivate ethical awareness, digital responsibility, and the requisite skills to navigate these technologies confidently and competently. The educators play a pivotal role in this journey by guiding students to leverage AI in ways that foster autonomy, nurture competence, and cultivate a sense of belonging, ensuring that AI catalyzes learning rather than impedes it [32]. Beyond the classroom walls, AI is making significant inroads at the institutional level. The application of neural network-based models to optimize faculty subject allocation, accounting for factors such as academic credentials, teaching experience, and administrative responsibilities [33]. This innovative approach promotes a more structured and outcomes-focused teaching environment. These observations suggest that while integrating AI into education offers numerous opportunities, its successful implementation depends on adopting comprehensive educational reforms. The universities must undertake a systematic revision of teaching models in order to establish a foundation that fully harnesses AI’s potential to enhance the overall educational experience [34].

H1. 

AI competency has a positive and significant effect on strategic intelligence.

H2. 

AI competency has a positive and significant effect on strategic agility.

2.3. Soft Skills

While digital transformation emphasizes technology, the ultimate success of innovation fundamentally depends on human and social dimensions. Soft skills are increasingly recognized as the “human infrastructure” of innovation ecosystems. In higher education, these competencies mediate the relationship between digital or technical capability and organizational innovation by shaping interpersonal trust, knowledge sharing, psychological safety, and learner agency. Research evidence increasingly underscores their importance in university contexts. The highlight the critical role of soft-skills-oriented interventions across curricular settings from the school to the university level [35]. They found that soft-skills interventions embedded within curricula across educational levels led to improved outcomes, including employability, teamwork, creativity, and academic success. A further meta-analysis reinforced these findings, demonstrating the central role of soft skills development in information technology-related higher education programs [36]. Demonstrates that universities integrating structured soft-skills training into their curricula achieve higher innovation capacity and student employability. These skills amplify the value of technical expertise, the cognitive, social, and emotional resources that underpin institutional productivity and long-term innovation potential. Within the lens of HCT, universities serve as both producers and beneficiaries of human capital. By developing soft skills among staff and students, institutions enhance their capabilities, enabling sustained innovation and responsiveness to emerging technological and societal challenges. Additionally, soft skills are key enablers of human-centered innovation, in which innovations are not merely technological artifacts but are co-designed with human users, embedded within social contexts, and subject to ethical scrutiny. For example, integrating soft skills through serious games in higher education fosters collaboration, decision-making, and reflective learning, supporting more inclusive and human-centered innovation processes [37]. Serious game-based interventions targeting soft skills have been shown to significantly enhance students’ perceptions of their capacity to operate effectively in real-world, interdisciplinary professional contexts. In another study, the creativity-enhancement programs in higher education substantially improved students’ problem-solving abilities and innovation-oriented behaviors [38]. Furthermore, soft skills play a critical mediating role in complex capability configurations by translating technical resources into organizational impact through enhanced team dynamics, effective stakeholder engagement, and reflective practice [39]. In the context of Thai higher education, the development of soft skills is especially salient. University graduates are increasingly expected not only to be technically competent but to demonstrate teamwork, communication, and adaptability in dynamic, interdisciplinary, and international work environments. Hence, in this study, the variable Soft Skills is conceptualized as the aggregate of interpersonal, intrapersonal, and adaptive competencies among faculty, staff, and students, enabling human-centered collaboration, creativity, and transformation.

H3. 

Soft-skill competency has a positive and significant effect on Innovation Capability.

H4. 

Soft-skill competency has a positive and significant effect on Strategic Agility.

2.4. Strategic Intelligence

In this study, Strategic Intelligence (SI) is conceptualized as a mediating capability that explains how AI competency and soft-skill competency are transformed into innovative university competency. Drawing on Dynamic Capabilities Theory (DCT), SI reflects the institution’s ability to sense environmental changes, interpret complex information, and translate insights into strategic actions. Rather than directly influencing outcomes, SI serves as an intermediate mechanism that aligns technological and human capabilities with organizational innovation objectives [40]. Specifically, AI competency enhances the availability and quality of data-driven insights, while soft skills enable effective communication, collaboration, and interpretation of these insights. Strategic intelligence integrates these inputs and converts them into strategic decisions that support innovation. Therefore, SI does not alter the strength of relationships between variables (as in moderation) but rather explains the process through which these relationships occur, consistent with a mediating role. When applied effectively, Strategic Intelligence enables organizations to stay ahead of the curve by anticipating change and accurately forecasting the future [41]. It serves as a cornerstone in management, providing timely and relevant information that enhances decision-making and strengthens risk management strategies [42]. To cultivate Strategic Intelligence, organizations must diligently gather data from a wide array of sources and employ sophisticated analysis techniques. This approach ensures that the information used for decision-making and strategic planning is not only high-quality but also actionable [14,35]. Analytical thinking is a vital skill in this process, as it enables individuals to identify significant trends and patterns, fostering informed decisions and the development of strategies aligned with organizational objectives [40,41]. Environmental scanning is a crucial tool for identifying opportunities and risks across both external and internal environments. This vigilance enables organizations to adapt effectively to change, enhancing their resilience and agility [43]. This method involves actively collecting data from multiple sources, conducting rigorous trend analysis, and presenting findings in a format that is easily understood by decision-makers [44]. In terms of risk management, Strategic Intelligence is essential, encompassing the identification, investigation, and strategic response to risks that could hinder organizational goals. By utilizing Strategic Intelligence, organizations can lessen negative impacts while seizing emerging opportunities [45]. Organizations also need to establish robust frameworks for analyzing, interpreting, and presenting data to support high-quality decisions aligned with long-term objectives [46]. As a result, Strategic Intelligence becomes a crucial ally, guiding organizations in understanding and responding swiftly to a dynamic competitive landscape, anticipating future scenarios, and enhancing strategic decision-making. In the context of higher education transformation, Strategic Intelligence (SI) is increasingly recognized as a critical meta-capability that enables universities to interpret, anticipate, and adapt to dynamic environmental conditions. It represents the synthesis of analytical, contextual, and visionary intelligence. Strategic intelligence enables leaders and academic institutions to convert fragmented information into actionable strategic decisions [47]. In university settings, SI underpins the institution’s ability to align AI adoption and human capability development with long-term strategic goals. For example, leaders with high SI can more effectively identify emerging opportunities in AI integration, anticipate ethical risks, and allocate resources for skill development. Strategic intelligence also moderates the relationship between AI competency and innovative university competency by ensuring that technological capabilities are strategically directed toward innovation performance. Without SI, universities risk fragmented AI initiatives that fail to translate into institutional innovation. Empirical studies confirm that organizations with higher strategic intelligence exhibit stronger absorptive capacity and dynamic capability, which, in turn, improve innovation performance and resilience. Thus, in the context of Thailand’s “new age universities,” SI ensures that investments in AI systems and soft-skill development are integrated with broader innovation strategies. Furthermore, from the perspective of DCT, strategic intelligence operates as a meta-level sensing–seizing–transforming mechanism. Hence, within this study’s conceptual framework, Strategic Intelligence is treated as a moderating variable that strengthens the causal relationships between AI Competency and Innovative University Competency, and between Soft Skills and Innovative University Competency. When SI is high, the university can align AI and human capital initiatives with its strategic vision, accelerating innovation outcomes; when SI is low, these competencies may remain underutilized or directionless.

H5. 

Strategic Intelligence (SI) has a positive and significant effect on Innovative University Competency (IUC).

H6. 

Strategic Intelligence (SI) mediates the relationship between AI Competency (AIC) and Innovative University Competency (IUC).

H7. 

Strategic Intelligence (SI) mediates the relationship between Soft-Skill Competency (SSC) and Innovative University Competency (IUC).

The literature reviewed in this study highlights the multidimensional roles of AI competency, soft skills, and strategic intelligence in enhancing innovative capabilities within higher education institutions. Theoretical perspectives such as DCT, RBV, and HCT collectively explain how these competencies function as strategic assets. As shown in

Table 1. Summary of literature on AI competency, soft skills, and strategic intelligence.

Context	Key Characteristics	Theoretical Lens	Advantages	Disadvantages	Refs.
Human Capital Development	Emphasizes knowledge, skills, and competencies as drivers of productivity	HCT	Strong foundation for linking skills to performance	Limited focus on technological integration	[8]
AI-based Learning Systems	AI-driven simulations, real-time feedback, dynamic learning environments	DCT, HCT	Enhances experiential learning and decision-making skills	Requires high computational resources and infrastructure	[9,10,11,12,13]
Entrepreneurship Education	AI integration in business and management education	DCT	Improves strategic thinking and opportunity recognition	Limited empirical validation in real-world outcomes	[14,15,16]
Generative AI in Education	AI supports learning tasks across Bloom’s taxonomy	HCT	Facilitates personalized and scalable learning	Weak performance in higher-order cognitive skills	[17,18,19]
Experiential Learning with AI	Real-time market simulations and entrepreneurial training	DCT, RBV	Bridges theory and practice effectively	Lack of longitudinal impact assessment	[20,21,22]
AI Technologies in Education	ML, NLP, and generative AI applications in teaching and administration	RBV	Enhances efficiency and automation	Over-reliance may reduce critical thinking	[23,24]
Sociotechnical Transformation	AI is reshaping institutional structures and learning roles	DCT	Promotes innovation in pedagogy and engagement	Complexity in implementation and governance	[25,26,27]
AI Tools & Student Learning	Adaptive learning, AI writing assistants, engagement tools	HCT	Improves accessibility and learning outcomes	Raises concerns about academic integrity	[28,29,30]
AI Literacy & Ethics	Focus on ethical awareness and digital responsibility	HCT	Promotes responsible AI use	Low literacy levels hinder effectiveness	[31,32]
Institutional AI Applications	AI for administration and resource optimization	RBV	Improves operational efficiency	Requires systemic institutional reform	[33,34]
Soft Skills Development	Curriculum-integrated soft skills training	HCT	Enhances employability and innovation capacity	Difficult to measure quantitatively	[35,36]
Serious Games & Creativity	Soft skills via simulations and creativity programs	DCT	Supports human-centered innovation	Limited scalability across institutions	[37,38]
Organizational Capability	Soft skills as mediators of technical impact	RBV, HCT	Strengthens team dynamics and collaboration	Context-dependent effectiveness	[39]
Strategic Intelligence	Data-driven decision-making and environmental scanning	DCT	Enhances forecasting and risk management	Requires high-quality data and analytical capability	[40,41,42,43,44,45,46]
Strategic Intelligence in Universities	SI as a meta-capability aligning AI and innovation	DCT	Integrates strategy with technology and human capital	Still underexplored empirically	[47]

, AI competency is widely recognized for enhancing analytical capabilities, enabling real-time decision-making, and supporting experiential learning environments. However, several studies also identify limitations, including over-reliance on AI systems, insufficient support for higher-order cognitive skills, and the need for substantial technological infrastructure.

While prior studies have extensively examined AI adoption, soft skills, and strategic capabilities in isolation, the existing literature reveals several important gaps. First, most research adopts a technology-centric perspective, focusing primarily on AI tools and digital infrastructure, while underestimating the role of human-centric capabilities such as soft skills in enabling effective implementation. Second, studies grounded in HCT emphasize individual competencies but often fail to integrate them with organizational-level strategic capabilities. Third, although DCT highlights the importance of sensing, seizing, and transforming processes, there is limited empirical work operationalizing these mechanisms through measurable constructs in higher education contexts. Moreover, the literature shows a lack of consensus regarding how technological and human capabilities interact to produce innovation outcomes. Existing studies tend to examine these factors independently rather than as part of an integrated capability system. This fragmentation limits the ability to fully explain how universities translate digital transformation initiatives into sustained innovation performance. To address these gaps, this study proposes an integrated framework that combines AI competency, soft-skill competency, and strategic intelligence within a unified model. By positioning strategic intelligence as a mediating mechanism, the study provides a more comprehensive explanation of how technological and human resources are orchestrated to enhance a university’s innovative competency.

2.5. Conceptual Model

Within the S–O–R framework, AI competency and soft skills are conceptualized as external and internal stimuli that influence researchers’ cognitive and behavioral processes, as shown in Figure 1. Instead of examining these constructs independently, this study integrates them as complementary capabilities that jointly shape innovation capability and strategic agility. AI competency enhances analytical and technical capacity, while soft skills facilitate adaptive thinking, collaboration, and knowledge application. Together, these capabilities influence internal organizational mechanisms, which subsequently determine research publication performance outcomes. This integrated approach avoids conceptual redundancy and provides a streamlined theoretical explanation.

3. Research Methodology

3.1. Research Design

This study employs a quantitative, explanatory research design grounded in an SEM framework to examine the causal relationships among AI Competency, Soft-Skill Competency, Strategic Intelligence, and IUC. The research focuses on identifying both direct and indirect effects among these variables within the conceptual model. SEM was selected because it allows for simultaneous testing of multiple latent constructs and their interrelationships, providing comprehensive insights into the complex structural dependencies among institutional competencies. The design follows a deductive approach, beginning with the development of a theoretical model based on three established frameworks: DCT, RBV, and HCT. These theories jointly explain how digital, human, and strategic competencies interact to enhance organizational innovation. The study operationalizes AI and soft-skill competencies as independent constructs that is associated with innovative university competency, both directly and through the mediating and moderating roles of strategic intelligence. A cross-sectional survey was conducted to collect quantitative data from academic and administrative personnel at Thai universities. The model’s statistical validation was performed using CFA to assess construct reliability and validity, followed by Structural Model Analysis to test the hypothesized causal pathways. This approach provides empirical evidence on how universities can leverage technological, human, and strategic resources to improve innovation performance. It offers a data-driven foundation for policy and management strategies that support Thailand’s transition toward a “new-age” higher education system.

3.2. Ethical Approval and Stakeholder Sample

This study focuses on participants from social sciences and science faculties to capture the complementary dimensions of AI competency and soft skills within higher education institutions. These two academic domains represent distinct yet interrelated knowledge paradigms. Social science disciplines primarily emphasize interpersonal communication, adaptability, and critical thinking, whereas science disciplines focus on analytical reasoning, technical expertise, and greater exposure to AI technologies. This contrast provides a robust empirical foundation for examining the interaction between human-centric competencies and technology-driven capabilities, which are central to this study’s conceptual framework. From a theoretical perspective, this selection is consistent with the RBV and Human HCT, as it enables the analysis of heterogeneous human capital configurations across disciplinary contexts. Furthermore, in the context of digital transformation in higher education, the adoption and utilization of AI technologies often vary significantly across faculties. Science faculties typically demonstrate higher levels of AI integration, while social science faculties tend to emphasize soft-skill development. Therefore, including both groups enhances the generalizability and explanatory power of the proposed research model. To ensure representativeness, a stratified sampling technique was employed, whereby respondents were proportionally selected from both faculty groups. Data were collected using a structured questionnaire administered to academic staff and students. Prior to analysis, the dataset was carefully screened to eliminate incomplete, inconsistent, or outlier responses, thereby ensuring data quality and integrity. The study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee for Human Research of Mahasarakham University, Thailand (Approval No. 001-814/2026) on 7 January 2026. All participants were informed about the purpose and procedures of the study, and informed consent was obtained electronically prior to completing the questionnaire. Participants indicated their consent by clicking the agreement statement before proceeding to the survey. All items in the questionnaire were rated on a five-point Likert scale (1 = strongly disagree to 5 = strongly agree). The data collection period extended from 7 January 2026 to 30 January 2026 to ensure clarity, reliability, and contextual relevance of the questions. Based on pilot feedback and expert review, we aim to enhance content validity and ensure comprehensive coverage of academic activities across both science and social science disciplines in the Thai context. The final version of the questionnaire was then approved for data collection. A total of 500 questionnaires were distributed, and 475 valid responses were retained after data screening, yielding a response rate of 95%. This sample size exceeds the recommended threshold for SEM analysis, ensuring robust parameter estimation, reliability, and generalizability. The resulting dataset adequately represents the population of Thai higher education institutions, supporting valid inferences about the associations between AI, soft skills, strategic intelligence competencies, and innovative university competencies. All participants were informed about the purpose and procedures of the study, and informed consent was obtained electronically prior to completing the questionnaire. For the online survey, electronic consent was secured via a mandatory confirmation checkbox before access to the questionnaire was granted. For the paper-based survey, participants signed a written consent form attached to the questionnaire. Consent documentation, including electronic records and signed forms, was securely stored in a password-protected database accessible only to the research team and retained solely for academic and ethical compliance purposes. Written informed consent was selected to ensure transparency, accurate documentation, auditability, and strict adherence to institutional ethical standards governing human-subject research in higher education contexts. In addition, data preprocessing procedures were conducted to improve the robustness of the analysis. All measurement items, based on Likert-scale responses, were normalized using a min–max normalization technique to standardize variable ranges and reduce potential scale-related bias. Reliability and validity assessments were subsequently performed using Cronbach’s alpha and CFA. The results confirm that all constructs satisfied the recommended thresholds for internal consistency, convergent validity, and construct reliability. These methodological procedures ensure that the dataset is suitable for rigorous statistical analysis and supports the validity of the study’s findings.

3.3. Research Instrument and Scales

The study employed a structured questionnaire as the primary research instrument to collect quantitative data on the constructs of AI Competency, Soft-Skill Competency, Strategic Intelligence, and IUC. The questionnaire was designed based on theoretical and empirical foundations from prior validated studies to ensure content validity and measurement reliability. The instrument comprised five main phases. Phase 1 collected demographic information, including gender, age, academic position, work experience, and university type. Phase 2 measured AI competency, adapted from prior validated instruments, covering dimensions of AI literacy, data analytics, ethical AI awareness, and digital innovation proficiency [28,29]. Phase 3 assessed soft-skill competency, adapted from established measurement scales, encompassing communication, teamwork, adaptability, problem-solving, creativity, and leadership [7,35]. Phase 4 evaluated Strategic Intelligence, drawing on [46], and captured foresight, analytical reasoning, and strategic alignment. All measurement items used a five-point Likert scale, ranging from 1 = Strongly Disagree to 5 = Strongly Agree, allowing respondents to express varying degrees of agreement or perceived intensity. This wider range increases measurement sensitivity and response variance, improving the precision of SEM estimates. Before final administration, the questionnaire was reviewed by three experts in higher education management and innovation to ensure face and content validity. A pilot test with 30 respondents yielded Cronbach’s alpha values exceeding 0.85 for all constructs, confirming internal consistency and reliability prior to the main data collection.

3.4. Familiarization with the Data

Data were collected from six faculties at MSU to provide a comprehensive view of how AI, soft-skill, and strategic intelligence competencies are associated with innovative university competencies across different academic domains. The study included three Social Science faculties (Mahasarakham Business School (MBS), Faculty of Education (EDU), and Faculty of Humanities and Social Sciences (HUSO). In parallel, three Science faculties participated (Faculty of Science (SCI), Faculty of Engineering (ENG), and Faculty of Informatics (ICT). Data were gathered through both online and paper-based questionnaires distributed via faculty research offices, academic departments, and internal communication networks. The respondents comprised academic and administrative staff who were actively engaged in teaching, research, or strategic management. A total of 475 valid responses were collected, with balanced participation from both social science and science disciplines. The integration of data from these six faculties enhanced the diversity and generalizability of the findings, ensuring that the model accurately reflects multidisciplinary perspectives on how technological capability, human adaptability, and strategic foresight collectively strengthen innovation within Thailand’s emerging “new-age university” framework.

3.5. Data Analysis/Validity Assessment

Given that this study relies on self-reported and cross-sectional data, the potential for common method bias (CMB) was assessed to ensure the validity of the results. First, Harman’s single-factor test was conducted using an exploratory factor analysis in which all measurement items loaded onto a single factor. The results indicate that the first factor accounts for less than 50% of the total variance, suggesting that common method bias is unlikely to significantly affect the findings. In addition, a full collinearity assessment was performed using variance inflation factor (VIF) values. Following established guidelines, VIF values below 3.3 indicate the absence of common method bias. The results show that all constructs have VIF values within acceptable limits, confirming that multicollinearity and common method bias are not major concerns in this study. These results collectively indicate that common method bias does not pose a significant threat to the validity of the study.

4. Presentation, Analysis, and Discussion of Findings

This section presents and discusses the study’s empirical findings. The analysis is structured into four main components. First, descriptive statistics are provided to summarize the data’s characteristics. Second, the measurement model is evaluated through reliability and validity assessments, including Cronbach’s alpha and CFA. Third, the structural model is examined to test the hypothesized relationships among AI competency, soft skills, strategic intelligence, and innovative university competency. Finally, the results are interpreted in light of the theoretical frameworks, including DCT, RBV, and HCT, to provide meaningful insights and implications.

4.1. Demographic Information

The demographic profile of respondents provides insight into the diversity and representativeness of the sample collected from Maha Sarakham University (MSU). Of the 475 valid responses, participants included both academic and administrative staff from six faculties across the social sciences and sciences. The distribution ensured balance among perspectives on teaching, research, and institutional management. In terms of gender, the sample consisted of 55.6% female and 44.4% male participants, reflecting the general workforce composition in Thai higher education. Regarding age, most respondents were between 31 and 45 years old (52.8%), followed by those aged 46–60 (27.6%) and those under 30 (19.6%), indicating a strong representation of mid-career professionals. In terms of work position, 62% were academic staff (lecturers, assistant professors, and researchers), while 38% were administrative or managerial personnel involved in university governance and innovation support. In terms of experience, 43.2% of respondents reported 6–10 years of service, 31.5% reported more than 10 years, and 25.3% reported fewer than 5 years, indicating a mix of experienced and early-career employees. Participants were drawn from six faculties, ensuring cross-disciplinary representation (as shown in

Table 2. Demographic profile of respondents from the study.

Demographic Variables	Categories	Frequency	Percent (%)
Gender	Male	211	44.4
	Female	264	55.6
Age (years)	Below 30	93	19.6
	31–45	251	52.8
	46–60	131	27.6
Employment Position	Academic Staff (Lecturer/Researcher)	295	62.1
	Administrative/Managerial Staff	180	37.9
Years of Service	Less than 5 years	120	25.3
	6–10 years	205	43.2
	More than 10 years	150	31.5
Faculty Group	Social Science (MBS, EDU, HUSO)	238	50.1
	Science (SCI, ENG, ICT)	237	49.9
Type of University Function	Teaching-oriented	189	39.8
	Research-intensive	208	43.8
	Autonomous/Hybrid	78	16.4

). The demographic distribution underscores a well-balanced dataset suitable for structural equation modeling (SEM). This diversity strengthens the reliability and validity of the findings by integrating perspectives from both technological and human-centered academic environments, which are essential for analyzing the relationships among AI Competency, Soft-Skill Competency, Strategic Intelligence, and Innovative University Competency.

4.2. SEM Analysis

This section outlines the SEM model fit and provides statistical evidence of the model’s adequacy in capturing the proposed relationships among the variables under study.

4.2.1. Comprehensive Interpretation of Model Fit Indices

The Confirmatory Factor Analysis (CFA) results presented in the table demonstrate a robust measurement model, confirming that the observed variables effectively represent their latent constructs [48].

• Chi-square (χ²) and χ²/df Ratio

The ratio of chi-square to degrees of freedom (χ²/df) is a commonly used goodness-of-fit index in structural equation modeling (SEM). It evaluates how well the proposed model fits the observed data while accounting for model complexity. The model yielded a χ² value of 1284.6 with 684 degrees of freedom (df), producing a χ²/df ratio of 1.88, computed using Equation (1).

$${\mathrm{x}}^2/\mathrm{df}={\displaystyle \frac{1284.60}{684}}=1.88$$

(1)

In this study, the computed χ²/df value indicates an acceptable model fit, as values below 3.0 are generally considered indicative of a good fit, and values below 2.0 indicate a very good fit [49]. The result suggests minimal discrepancy between the sample covariance matrix and the model’s estimated covariance structure.

2.

Comparative Fit Index (CFI)

The Comparative Fit Index (CFI) assesses the improvement of the proposed model relative to an independence (null) model in which all variables are assumed to be uncorrelated. The CFI = 0.956 compares the hypothesized model with a null (independence) model defined by Equation (2).

$$\mathrm{CFI}=1-{\displaystyle \frac{({\mathrm{x}}_{\mathrm{model}}^2-{\mathrm{df}}_{\mathrm{model}})}{({\mathrm{x}}_{\mathrm{baseline}}^2-{\mathrm{df}}_{\mathrm{baseline}})}}$$

(2)

where ${\mathrm{x}}_{\mathrm{model}}^2$ is chi-square value of the proposed model, ${\mathrm{df}}_{\mathrm{model}}$ is degrees of freedom of the proposed model, ${\mathrm{x}}_{\mathrm{baseline}}^2$ is chi-square value of the null (independence) model, and ${\mathrm{df}}_{\mathrm{baseline}}$ is degrees of freedom of the null model. Values above 0.95 reflect excellent model fit [50], confirming that the data fit the proposed structure significantly better than a random model.

3.

Tucker–Lewis Index (TLI)

The Tucker–Lewis Index (TLI), also known as the Non-Normed Fit Index (NNFI), evaluates model fit while penalizing model complexity. The TLI = 0.947 adjusts for model complexity using Equation (3).

$$\mathrm{TLI}={\displaystyle \frac{({\mathrm{x}}_{\mathrm{baseline}}^2/{\mathrm{df}}_{\mathrm{baseline}})-({\mathrm{x}}_{\mathrm{model}}^2/{\mathrm{df}}_{\mathrm{model}})}{({\mathrm{x}}_{\mathrm{baseline}}^2/{\mathrm{df}}_{\mathrm{baseline}})-1}}$$

(3)

where all parameters are defined as above. This indicates high parsimony and reliability in representing relationships among constructs [51].

4.

Incremental Fit Index (IFI)

The IFI = 0.957 indicates a similar improvement over the baseline model, reinforcing incremental validity.

5.

Root Mean Square Error of Approximation (RMSEA)

The RMSEA measures how well the model approximates the population covariance matrix per degree of freedom. The RMSEA = 0.045, with a 90% confidence interval of [0.041–0.049], is calculated as Equation (4) [50].

$$\mathrm{RMSEA}=\sqrt{{\displaystyle \frac{({\mathrm{x}}_{\mathrm{model}}^2-{\mathrm{df}}_{\mathrm{model}})}{{\mathrm{df}}_{\mathrm{model}}(\mathrm{N}-1)}}}$$

(4)

where X² is the chi-square statistic, df is the degrees of freedom, and N is the sample size. Values ≤ 0.05 indicate a close fit, demonstrating that the model generalizes well across the population. If RMSEA ≤ 0.08 indicates acceptable fit, RMSEA ≤ 0.05 indicates good fit, and lower values represent better approximation and model parsimony.

6.

Standardized Root Mean Square Residual (SRMR)

The SRMR represents the standardized difference between observed and predicted correlations. The SRMR = 0.047 is derived from Equation (5).

$$\mathrm{SRMR}=\sqrt{{\displaystyle \frac{2}{\mathrm{p}(\mathrm{p}+1)}}{\displaystyle \sum _{\mathrm{i}\le \mathrm{j}}{({\mathrm{r}}_{\mathrm{ij}}-{\widehat{\mathrm{r}}}_{\mathrm{ij}})}^2}}$$

(5)

where ${\mathrm{r}}_{\mathrm{ij}}$ and ${\widehat{\mathrm{r}}}_{\mathrm{ij}}$ denote observed correlation between variables i and j and model-implied correlations. A value below 0.08 indicates very low residual discrepancies.

Table 3. SEM fit indices.

Fit Index	Statistic Value	Criteria	Conclusions
Chi-square (χ2)	1287.462	Non-significant desirable (sensitive to N)	Acceptable
χ2/df	1.95	≤3.00 = Acceptable	Good
CFI	0.953	≥0.90 (acceptable); ≥0.95 (excellent)	Excellent
TLI	0.945	≥0.90 (acceptable); ≥0.95 (excellent)	Good
GFI	0.924	≥0.90	Good
AGFI	0.901	≥0.90	Acceptable
RMSEA	0.046	≤0.08 (adequate); ≤0.05 (close fit)	Good
SRMR	0.051	≤0.08	Acceptable

presents the results of the SEM goodness-of-fit indices, which assess how well the hypothesized four-construct measurement model fits the observed data. The model demonstrates a strong overall fit, meeting or exceeding the recommended thresholds across all statistical indices. Although significant due to the large sample size (n = 475), the Chi-square statistic (χ² = 1287.462, df = 660) remains within acceptable limits when adjusted for model complexity. The χ²/df ratio of 1.95 falls well below the benchmark of 3.0, indicating a parsimonious model that balances simplicity with explanatory accuracy. Incremental fit indices, such as the Comparative Fit Index (CFI = 0.953) and the Tucker–Lewis Index (TLI = 0.945), exceed the 0.90 criterion and approach the 0.95 ideal level, indicating that the hypothesized model fits the data substantially better than the null model. Similarly, the Goodness-of-Fit Index (GFI = 0.924) and the Adjusted Goodness-of-Fit Index (AGFI = 0.901) both exceed the 0.90 threshold, indicating that the model accounts for a substantial proportion of the observed variance and covariances. The Root Mean Square Error of Approximation (RMSEA = 0.046), with values below 0.05, indicates a close and acceptable model fit, while the Standardized Root Mean Square Residual (SRMR = 0.051), below the 0.08 cutoff, suggests minimal residual discrepancies between observed and predicted correlations. Collectively, these results confirm that the CFA model achieves excellent construct validity and model adequacy. The convergence of fit indices indicates that the measurement structure for AIC, SSC, SI, and IUC is statistically sound and theoretically consistent, providing a strong foundation for subsequent SEM analyses to test the hypothesized causal relationships among competencies influencing innovative university performance.

4.2.2. Confirmatory Factor Analysis (CFA)

To verify the validity and reliability of the measurement model, CFA was conducted using maximum likelihood estimation. The model examined the relationships among four latent variables and their respective observed indicators. Each indicator was designed to measure a single latent construct, grounded in theoretical alignment with prior research.

Table 4. Standardized factor loadings and significance of observed variables.

Construct	Observed Variable	Standardized Loading (λ)	t-Value	Conclusion
AI Competency (AIC)	AIC1	0.823	13.42	Significant
	AIC2	0.854	14.06	Significant
	AIC3	0.807	12.78	Significant
	AIC4	0.783	11.96	Significant
	AIC5	0.796	12.11	Significant
Soft-Skill Competency (SSC)	SSC1	0.791	12.67	Significant
	SSC2	0.836	13.29	Significant
	SSC3	0.807	12.74	Significant
	SSC4	0.853	13.56	Significant
	SSC5	0.878	14.22	Significant
Strategic Intelligence (SI)	SI1	0.831	13.84	Significant
	SI2	0.869	14.37	Significant
	SI3	0.804	13.06	Significant
	SI4	0.872	14.48	Significant
	SI5	0.851	13.92	Significant
Innovative University Competency (IUC)	IUC1	0.817	13.22	Significant
	IUC2	0.861	13.97	Significant
	IUC3	0.799	12.81	Significant
	IUC4	0.834	13.48	Significant
	IUC5	0.808	12.89	Significant

presents the standardized factor loadings (λ), corresponding t-values, and significance levels for all observed variables under the five latent constructs: AIC, SSC, SI, and IUC. The results indicate strong empirical support for the measurement model. For AIC, standardized loadings range from 0.783 to 0.854. Specifically, AIC2 exhibits the highest loading (λ = 0.854, t = 14.06), followed by AIC1 (λ = 0.823, t = 13.42) and AIC3 (λ = 0.807, t = 12.78). All t-values exceed the critical threshold of 1.96, confirming statistical significance at p < 0.05. These results demonstrate that all five indicators reliably represent the AI Competency construct. Regarding SSC, factor loadings range from 0.791 to 0.878. SSC5 shows the strongest loading (λ = 0.878, t = 14.22), suggesting it is the most influential indicator within this construct. The consistently high t-values (12.67–14.22) confirm that each observed variable significantly contributes to measuring soft-skill competency. For SI, standardized loadings range from 0.804 to 0.872. SI4 has the highest loading (λ = 0.872, t = 14.48), indicating a particularly strong reflection of strategic intelligence. All five items demonstrate statistically significant relationships with the latent construct. Finally, the IUC shows loadings ranging from 0.799 to 0.861. IUC2 records the highest loading (λ = 0.861, t = 13.97), while IUC3 shows the lowest (λ = 0.799, t = 12.81), though both are still well above acceptable thresholds.

4.2.3. Correlation Coefficient Matrix

The correlation coefficient matrix for observed variables provides a comprehensive overview of the linear associations among all measured indicators representing the four latent constructs: AIC, SSC, SI, and IUC. Each correlation coefficient (r_ij) quantifies the strength and direction of the relationship between two observed variables, with values ranging from −1 to +1. Positive coefficients indicate that as one variable increases, the other tends to grow as well, whereas negative coefficients would suggest an inverse relationship. The Pearson correlation coefficient measures the strength and direction of a linear relationship between two observed variables and is defined in Equation (6) [51].

$${\mathrm{r}}_{\mathrm{ij}}={\displaystyle \frac{{\displaystyle \sum _{\mathrm{k}=1}^{\mathrm{n}}({\mathrm{x}}_{\mathrm{ik}}-{\overline{\mathrm{x}}}_{\mathrm{i}})({\mathrm{x}}_{\mathrm{jk}}-{\overline{\mathrm{x}}}_{\mathrm{j}})}}{\sqrt{{\displaystyle \sum _{\mathrm{k}=1}^{\mathrm{n}}({\mathrm{x}}_{\mathrm{ik}}-{\overline{\mathrm{x}}}_{\mathrm{i}}){({\mathrm{x}}_{\mathrm{jk}}-{\overline{\mathrm{x}}}_{\mathrm{j}})}^2\cdot \sqrt{{\displaystyle \sum _{\mathrm{k}=1}^{\mathrm{n}}({\mathrm{x}}_{\mathrm{jk}}-{\overline{\mathrm{x}}}_{\mathrm{j}}){({\mathrm{x}}_{\mathrm{jk}}-{\overline{\mathrm{x}}}_{\mathrm{j}})}^2}}}}}}, \mathrm{or} {\mathrm{r}}_{\mathrm{ij}}={\displaystyle \frac{\mathrm{Cov}({\mathrm{X}}_{\mathrm{i}},{\mathrm{X}}_{\mathrm{j}})}{{\sigma }_{X_{\mathrm{i}}},{\sigma }_{X_{\mathrm{j}}}}}$$

(6)

where r_ij is the correlation between variable X_i and X_j, ${\overline{\mathrm{x}}}_{\mathrm{i}}$ mean of variable X_i, is the standard deviation of variable X_i, and $\mathrm{Cov}({\mathrm{X}}_{\mathrm{i}},{\mathrm{X}}_{\mathrm{j}})$ is the covariance between variables X_i and X_j. In this analysis, the correlation coefficients among all observed variables across the four latent constructs were calculated with six-digit precision. All coefficients are positive and significant, ranging from 0.478129 to 0.789652, indicating moderate-to-strong linear relationships and confirming that the items consistently represent their intended constructs. Within each construct, high inter-item correlations (AIC1-AIC2 = 0.752145; SSC3-SSC4 = 0.734157; SI1-SI2 = 0.772513; IUC1-IUC2 = 0.784932) demonstrate strong internal coherence and convergent validity. Cross-construct correlations remain moderate (0.50–0.65), for example, AIC2-SSC2 = 0.546298 and SI2-IUC3 = 0.642819, supporting discriminant validity, as no coefficient exceeds 0.85. Figure 2 presents the correlation matrix among the key constructs examined in this study: AI Competency (AIC), Soft Skills (SSC), Strategic Intelligence (SI), and Innovative University Competency (IUC). To improve clarity and interpretability, only the most relevant variables are included, and the visualization focuses on significant relationships. The results indicate that AI competency and soft skills are positively correlated with strategic intelligence and innovative university competency, supporting the proposed theoretical relationships. The strongest associations occur between SI and IUC variables, confirming that strategic foresight and analytical capability translate technological and soft-skill competencies into institutional innovation. Meanwhile, the moderate correlation between AI Competency and Soft Skills demonstrates the complementary relationship between technical proficiency and human adaptability. The overall pattern confirms that all constructs are interrelated yet distinct, ensuring both statistical reliability and theoretical clarity. This balanced correlation structure demonstrates the measurement model’s robustness, validating its suitability for subsequent SEM analyses and reinforcing the conclusion that digital, human, and strategic capabilities jointly underpin innovation performance in universities.

4.2.4. Reliability and Validity Assessment

To ensure the robustness of the measurement model, both reliability and validity tests were conducted in accordance with established SEM-CFA guidelines. Reliability refers to the internal consistency of measurement items, while validity assesses how well the indicators represent their intended latent constructs.

• Reliability Analysis

Reliability was examined using Cronbach’s alpha (α) and Composite Reliability (CR). Cronbach’s alpha (α) values for all constructs exceeded 0.90, far surpassing the 0.70 benchmark, confirming strong internal consistency [51]. CR values ranged from 0.934 to 0.951, reflecting stable inter-item correlations across all observed indicators. The formula for composite reliability is defined as Equation (7).

$$\mathrm{CR}2={\displaystyle \frac{{\left({\displaystyle \sum {\lambda }_{\mathrm{i}}}\right)}^2}{{\left({\displaystyle \sum {\lambda }_{\mathrm{i}}}\right)}^2+\left({\displaystyle \sum {\lambda }_{\mathrm{i}}^2}\right)}}$$

(7)

where λ_i represents the standardized factor loading for each observed variable. A CR ≥ 0.70 indicates satisfactory construct reliability.

2.

Validity Analysis

Convergent validity was verified using Average Variance Extracted (AVE), computed as Equation (8).

$$\mathrm{AVE}={\displaystyle \frac{{\displaystyle \sum {\lambda }_{\mathrm{i}}^2}}{\mathrm{n}}}$$

(8)

where is the squared standardized loading and n is the number of items. The reliability and validity analysis was conducted to assess the internal consistency, convergent validity, and discriminant validity of the four latent constructs: AIC, SSC, SI, and IUC. The findings, presented in

Table 5. Reliability and validity results of constructs.

Construct	Cronbach’s Alpha (α)	CR	AVE	√AVE	Threshold Criteria
AI Competency (AIC)	0.921	0.943	0.671	0.8192	α > 0.700; CR > 0.700; AVE > 0.500; √AVE > r
Soft-Skill Competency (SSC)	0.932	0.947	0.684	0.826	α > 0.700; CR > 0.700; AVE > 0.500; √AVE > r
Strategic Intelligence (SI)	0.938	0.951	0.703	0.838	α > 0.700; CR > 0.700; AVE > 0.500; √AVE > r
Innovative University Competency (IUC)	0.928	0.941	0.660	0.812	α > 0.700; CR > 0.700; AVE > 0.500; √AVE > r

, confirm that all constructs exhibit strong psychometric properties and are statistically suitable for inclusion in the structural model. The Cronbach’s alpha (α) values for all constructs ranged from 0.921 to 0.938, far exceeding the acceptable threshold of 0.70, indicating excellent internal consistency among the observed indicators. The CR values ranged from 0.934 to 0.951, demonstrating high internal reliability and stable item intercorrelations. These results confirm that each construct is consistently measured by its corresponding items. In terms of convergent validity, the AVE values ranged between 0.641 and 0.703, exceeding the recommended minimum of 0.50. This suggests that their respective latent constructs explain more than 64% of the variance in the indicators. The square roots of AVE (√AVE) ranged from 0.800 to 0.839, which are greater than the inter-construct correlation coefficients, satisfying the Fornell–Larcker criterion and confirming discriminant validity. Overall, the high Cronbach’s alpha, CR, and AVE values collectively confirm that the measurement model possesses strong reliability, convergent validity, and discriminant validity. These findings validate that all four constructs, AIC, SSC, SI, and IUC, are empirically sound and theoretically coherent, providing a solid foundation for hypothesis testing on the relationships among competencies influencing innovative university performance.

4.2.5. Fit Indices for the Structural Equation Model (SEM)

The overall goodness-of-fit statistics for the proposed SEM, which examines the interrelationships among AIC, SSC, SI, and IUC, are shown in

Table 6. Fit indices for SEM.

Fit Index	Statistic Value	Acceptable Criteria	Conclusion
Chi-square (χ2)	1347.215	Non-significant desirable (sensitive to sample size)	Acceptable
Chi-square/df (χ2/df)	1.95	<3.00 = Acceptable; <2.00 = Good	Good
CFI	0.953	≥0.90 = Acceptable; ≥0.95 = Excellent	Excellent
TLI	0.945	≥0.90 = Acceptable; ≥0.95 = Excellent	Good
IFI	0.954	≥0.90 = Acceptable; ≥0.95 = Excellent	Excellent
GFI	0.924	≥0.90 = Acceptable	Good
AGFI	0.901	≥0.90 = Acceptable	Acceptable
RMSEA	0.046	≤0.08 = Acceptable; ≤0.05 = Close fit	Good
SRMR	0.051	≤0.08 = Acceptable; ≤0.05 = Excellent	Good
NFI	0.928	≥0.90 = Acceptable	Good
PNFI	0.814	≥0.50 = Acceptable	Good
PCFI	0.832	≥0.50 = Acceptable	Good

. The fit indices collectively indicate that the model exhibits strong and acceptable alignment between the hypothesized structure and the observed data, confirming its empirical and theoretical adequacy. The Chi-square statistic (χ² =1347.215, df = 690) is significant (as expected with large samples) and yields a χ²/df ratio of 1.95, which is below the threshold of 3.0, indicating a parsimonious and well-fitting model. Incremental fit indices, including CFI (0.953), TLI (0.945), and IFI (0.954), exceed the minimum acceptable level of 0.90 and approach the ideal level of 0.95, indicating that the proposed model performs substantially better than the null (independence) model. Similarly, NFI (0.928) and GFI (0.924) indicate acceptable model fit, suggesting that the hypothesized structure accounts for a substantial proportion of the variance and covariance. In terms of absolute fit, the RMSEA value of 0.046 (90% CI: 0.041–0.051) demonstrates a close approximate fit, falling within the “good” range (≤0.05), while the SRMR value of 0.051 supports minimal standardized residual differences between observed and model-implied correlations. The AGFI (0.901) further confirms adequate model adjustment for degrees of freedom, while parsimony indices (PNFI = 0.814; PCFI = 0.832) indicate an optimal balance between simplicity and explanatory power. Overall, these indices collectively support the conclusion that the SEM exhibits excellent fit to the empirical data. The model successfully captures the complex relationships among technological, human, cognitive, and institutional capabilities. Therefore, the structural model is statistically robust, theoretically meaningful, and well suited to test the causal pathways hypothesized in the study, namely, that AI Competency and Soft-Skill Competency, mediated by Strategic Intelligence, affect Innovative University Competency in the context of higher education transformation in Thailand.

4.2.6. Path Analysis: Direct, Indirect, and Total Effects

The standardized direct, indirect, and total effects among the primary constructs: AIC, SSC, SI, and IUC are shown in

Table 7. Path analysis: direct, indirect, and total effects.

Path (From → To)	Direct Effect (β)	Indirect Path (s)	Indirect Effect (β)	Total Effect (β)	Conclusion
AI Competency (AIC) → Innovative University Competency (IUC)	0.321674	AIC → SI→ IUC = 0.612347 × 0.657184	0.402213	0.723887	Supported
Soft-Skill Competency (SSC) → Innovative University Competency (IUC)	0.274851	SSC → SI → IUC = 0.583192 × 0.657184	0.383764	0.658615	Supported
Strategic Intelligence (SI) → Innovative University Competency (IUC)	0.657184	—	—	0.657184	Supported
AI Competency (AIC) → Strategic Intelligence (SI)	0.612347	—	—	0.612347	Supported
Soft-Skill Competency (SSC) → Strategic Intelligence (SI)	0.583192	—	—	0.583192	Supported

. The results reveal that all hypothesized relationships were statistically supported, indicating that each construct significantly contributes to university innovation outcomes. The direct effect of AIC on IUC (β = 0.321674) shows that AI capability alone contributes moderately to innovation. However, when mediated by SI (AIC → SI → IUC), the indirect effect rises to β = 0.402213, resulting in a total effect of β = 0.723887. This pattern demonstrates that AI-related knowledge and applications are significantly more impactful when strategic insight is present to interpret and implement AI-driven initiatives effectively. Similarly, SSC has a direct effect on IUC (β = 0.274851) and an indirect effect via SI (β = 0.383764), yielding a total effect of β = 0.658615. This demonstrates that interpersonal, cognitive, and adaptive skills indirectly enhance innovation through strategic reasoning and decision-making. The mediating role of SI suggests that universities with higher strategic foresight can more effectively transform both AI and soft-skill capabilities into innovation capability. Furthermore, the direct paths AIC → SI (β = 0.612347) and SSC → SI (β = 0.583192) highlight that both technical and human competencies are foundational to cultivating strategic intelligence. Collectively, the results confirm partial mediation, in which SI serves as a crucial mechanism linking competencies to innovation performance.

In the single-mediator indirect effect, the relationship between the independent variable (X) and the dependent variable (Y) operates through a mediating construct (M). The computation follows the product-of-coefficients approach, as defined in Equation (9).

$${\mathrm{Indirect}}_{\mathrm{x}\to \mathrm{M}\to \mathrm{Y}}={\mathsf{\beta }}_{\mathrm{x}\to \mathrm{M}}\times {\mathsf{\beta }}_{\mathrm{x}\to \mathrm{Y}}$$

(9)

For the present study, two primary indirect paths were examined: AIC → SI → IUC (0.612347 × 0.657184 = 0.402213) and SSC → SI → IUC (0.583192 × 0.657184 = 0.383764). These results demonstrate that both AI and soft-skill competencies are strongly and indirectly associated with IUC through strategic intelligence. The magnitudes (β = 0.402213 and β = 0.383764) are relatively high, indicating that SI plays a critical mediating role, transforming technical and human competencies into institutional innovation outcomes. The total effect combines direct and indirect influences and is computed using Equation (10).

$${\mathrm{Total}}_{\mathrm{x}\to \mathrm{Y}}={\mathrm{Direct}}_{\mathrm{x}\to \mathrm{Y}}+{\mathrm{Indirect}}_{\mathrm{x}\to \mathrm{Y}}$$

(10)

Accordingly, AIC → IUC (0.321674 + 0.402213 = 0.723887) and SSC → IUC (0.274851 + 0.383764 =0.658615). The total effects reveal that both AIC and SSC have a substantial overall influence on IUC, with AI competency contributing slightly more (as shown in Figure 3). These outcomes reinforce the idea that universities with strong AI and soft-skill infrastructure achieve higher levels of institutional innovation. Overall, this analysis confirms that Strategic Intelligence functions as a powerful mediating mechanism that amplifies the effects of both AI and soft-skill competencies. Rather than operating in isolation, these factors collectively enhance universities’ ability to innovate, adapt, and lead in the era of digital transformation, aligning well with the Dynamic Capability Theory and the Human Capital Theory frameworks.

This study adopts a hybrid SEM–ANN approach to leverage the complementary strengths of both techniques. Structural Equation Modeling (SEM) is employed to test the hypothesized relationships and validate the theoretical framework, as it is well suited for examining causal paths and latent constructs. However, SEM is limited by its reliance on linear assumptions and its primary focus on explanatory analysis. To address these limitations, ANN analysis is incorporated as a second-stage analytical technique. ANN is a nonlinear, data-driven modeling approach capable of capturing complex relationships among variables without requiring strict distributional assumptions. In this study, an ANN is used to enhance predictive performance and identify the relative importance of input variables influencing innovative university competency.

4.3. Integration of SEM and ANN

While SEM confirms the validity of the theoretical relationships among AI competency, soft skills, strategic intelligence, and innovative university competency, ANN provides additional insights that extend beyond hypothesis testing. Specifically, ANN analysis improves predictive accuracy by modeling nonlinear relationships among variables, which are not fully captured by SEM. The ANN results reveal the relative importance of predictors, indicating which factors contribute most significantly to innovative university competency. This ranking provides a deeper understanding of variable influence beyond statistical significance alone. Furthermore, ANN demonstrates superior predictive capability, as evidenced by lower prediction error metrics, highlighting its effectiveness in forecasting innovation outcomes. These findings suggest that AI competency and strategic intelligence are not only statistically significant predictors (as confirmed by SEM) but also exhibit strong predictive power in nonlinear modeling contexts. Therefore, integrating ANN enhances the study’s robustness by complementing SEM’s explanatory strengths with predictive analytics, offering a more comprehensive understanding of the underlying relationships. This study adopts a hybrid SEM–ANN approach to leverage the complementary strengths of both techniques. Structural Equation Modeling (SEM) is employed to test the hypothesized relationships and validate the theoretical framework, as it is well suited to examining causal paths and latent constructs [52]. While SEM is highly effective in examining linear causal relationships and validating theoretical constructs, it may not adequately capture complex nonlinear interactions among variables. Therefore, an ANN was used to enhance the model’s predictive capability and to evaluate the relative importance of key predictors of IUC. The use of ANN in conjunction with SEM has become increasingly common in social science and information systems research because it allows researchers to address both explanatory and predictive objectives simultaneously. In this study, SEM was first used to validate the measurement model and test the hypothesized structural relationships among AIC, SSC, SI, and IUC. Subsequently, ANN was applied to the significant predictors identified in the SEM stage to capture potential nonlinear patterns and evaluate predictive performance. The ANN analysis employed a feed-forward multilayer perceptron (MLP) trained using the backpropagation learning algorithm. This architecture is widely applied in behavioral and management research because it effectively models nonlinear relationships between predictor variables and outcomes while maintaining computational efficiency. In the present study, the latent construct scores generated by SEM served as input variables for the ANN model, thereby ensuring that measurement error had been minimized through the confirmatory factor analysis stage.

4.3.1. ANN Model Design and Mathematical Formulation

To complement the SEM analysis, an ANN model was developed to enhance predictive accuracy and capture nonlinear relationships among variables. The ANN model was constructed using a multilayer perceptron (MLP) architecture with three main layers: an input layer, a hidden layer, and an output layer. The input layer includes the independent variables, namely AIC, SSC, and SI. These variables serve as predictors of the dependent variable, IUC, which is represented in the output layer. The hidden layer processes the input signals through weighted connections and nonlinear activation functions, enabling the model to capture complex interactions among variables. A sigmoid (or hyperbolic tangent) activation function was applied to introduce nonlinearity into the model. The ANN model was trained using a feedforward–backpropagation algorithm, in which the network iteratively adjusts its weights to minimize prediction error. The dataset was divided into training and testing subsets to ensure model generalizability and avoid overfitting. Mathematically, the ANN process can be expressed as Equation (11).

$$\mathrm{X}=\left({\mathrm{x}}_1,{\mathrm{x}}_2,{\mathrm{x}}_3\right)$$

(11)

where x₁ represents AI Competency (AIC), x₂ represents SSC, and x₃ represents SI. The weighted sum of inputs for hidden neuron j is calculated as Equation (12).

$${\mathrm{Z}}_{\mathrm{j}}={\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{w}}_{\mathrm{ij}}{\mathrm{x}}_{\mathrm{i}}}+{\mathrm{b}}_{\mathrm{j}}$$

(12)

where w_ij is the connection weight between input neuron i and hidden neuron j, b_j represents the bias term, and n is the number of input nodes. The activation of the hidden neuron is obtained using the sigmoid function, defined in Equation (13).

$${\mathrm{h}}_{\mathrm{j}}={\displaystyle \frac{1}{1+{\mathrm{e}}^{-{\mathrm{z}}_{\mathrm{j}}}}}$$

(13)

The output neuron aggregates the hidden-layer outputs as in Equation (14).

$$\mathrm{y}=\mathrm{g}\left({\displaystyle \sum _{\mathrm{j}=1}^{\mathrm{m}}{\mathrm{v}}_{\mathrm{j}}{\mathrm{h}}_{\mathrm{j}}+{\mathrm{b}}_{\mathrm{o}}}\right)$$

(14)

where v_j is the weight connecting the hidden neuron j to the output neuron, b_o is the output bias, and m is the number of hidden neurons. The ANN model in this study was implemented as a feed-forward multilayer perceptron trained with the backpropagation algorithm. The network consisted of three input neurons representing AIC, SSC, and SI; one hidden layer with six neurons; and a single output neuron representing IUC. The sigmoid activation function was applied in the hidden layer, while a linear activation function was used in the output layer. The dataset was divided into 70% training and 30% testing subsets. The network was trained for 200 epochs and evaluated across 10 runs to ensure stable predictive performance, using RMSE as the evaluation metric.

Table 8. Artificial neural network predictive performance results.

Dataset	RMSE	MSE	MAE	R2
Training	0.072	0.0052	0.058	0.912
Testing	0.086	0.0074	0.067	0.894
Average (10 runs)	0.079	0.0063	0.062	0.903

presents the predictive performance of the ANN model using multiple evaluation metrics, including RMSE, MSE, MAE, and the coefficient of determination (R²). As shown in Table 8, the ANN model achieved RMSEs of 0.072 and 0.086 on the training and testing datasets, respectively, indicating strong predictive accuracy and stable generalization. The corresponding MSE and MAE values are relatively small, further confirming that the predicted values closely approximate the observed values of Innovative University Competency. Additionally, the R² values of 0.912 (training) and 0.894 (testing) demonstrate that the ANN model explains a substantial proportion of the variance in the dependent variable. The average performance across ten repeated runs yields an RMSE of 0.079, suggesting that the neural network provides reliable and consistent predictive performance. Overall, these results confirm the robustness of the ANN model in predicting Innovative University Competency based on AI Competency, Soft-Skill Competency, and Strategic Intelligence. Figure 4 illustrates the performance and learning behavior of the ANN model. The prediction–actual scatter plot demonstrates a strong alignment between predicted and observed values, indicating high predictive accuracy. The error convergence curve shows a steady reduction in RMSE over training epochs, confirming the model’s stability. Additionally, the performance metrics and training–testing comparison further verify the robustness and generalization capability of the ANN model.

4.3.2. ANN Predictive Analysis and Variable Importance

The ANN analysis provides additional insights beyond SEM by identifying the relative importance of predictors and improving prediction accuracy. The results indicate that AI Competency and Strategic Intelligence are the most influential predictors of Innovative University Competency, followed by Soft-Skill Competency. These rankings highlight that, while all variables are statistically significant (as confirmed by SEM), their predictive contributions differ when nonlinear relationships are accounted for. In particular, Strategic Intelligence demonstrates a strong predictive role, reinforcing its function as a mediating capability that translates technological and human resources into innovation outcomes. Furthermore, the ANN model demonstrates strong predictive performance, as indicated by low prediction error values. This suggests that ANN is effective in modeling complex relationships and enhances the robustness of the study’s findings. The comparative roles of Structural Equation Modeling (SEM) and Artificial Neural Network (ANN) in this study are summarized in

Table 9. Comparison of SEM and ANN Approaches.

Criteria	SEM (Structural Equation Modeling)	ANN (Artificial Neural Network)
Primary Purpose	Theory testing and hypothesis validation	Prediction and pattern recognition
Model Type	Parametric, linear	Non-parametric, nonlinear
Assumptions	Requires normality, linearity	No strict assumptions
Relationships	Linear causal relationships	Captures complex nonlinear relationships
Output	Path coefficients, significance levels	Prediction accuracy, variable importance
Strength	Explains relationships between constructs	Enhances predictive performance
Limitation	Limited in handling nonlinearity	Less interpretable (black-box nature)
Role in Study	Validates the theoretical framework	Improves the prediction and ranking of factors

. SEM is primarily employed to validate the theoretical framework and examine causal relationships among constructs, whereas ANN is utilized to enhance predictive accuracy and capture nonlinear patterns that may not be adequately addressed by SEM.

4.3.3. Feature Importance Analysis Using SHAP

To further interpret the predictive behavior of the ANN model, this study employed SHapley Additive exPlanations (SHAP) to evaluate the relative importance and contribution of each predictor variable to the model output [53]. SHAP is derived from Shapley values in cooperative game theory, where each feature’s contribution is interpreted as a player’s marginal contribution to the overall prediction. In this study, each predictor variable is treated as a player in predicting IUC. The Shapley value for feature i is defined as Equation (15).

$${\varphi }_{\mathrm{i}}={\displaystyle \sum _{\mathrm{S}\subseteq \mathrm{F}}\frac{\left|\mathrm{S}\right|!\left(\left|\mathrm{F}\right|-\left|\mathrm{S}\right|-1\right)!}{\left|\mathrm{F}\right|!}}\left[\mathrm{f}\left(\mathrm{S}\cup \left\{\mathrm{i}\right\}\right)-\mathrm{f}\left(\mathrm{S}\right)\right]$$

(15)

where F represents the full set of features, S represents a subset of features excluding feature i, f(S) is the model prediction using the feature subset S, f(S ∪ {i}) is the prediction when feature i is added. |F|is the total number of features. The overall prediction of the ANN model can be expressed as the sum of SHAP contributions, as defined in Equation (16).

$$\mathrm{f}\left(\mathrm{x}\right)={\varphi }_0+{\displaystyle \sum _{\mathrm{i}=1}^{\mathrm{M}}{\varphi }_{\mathrm{i}}}$$

(16)

where f(x) is the predicted value of the ANN model, ϕ₀ represents the baseline prediction, ϕ_i represents the SHAP contribution of feature i, and M is the total number of input features. In this study, SHAP analysis was applied to quantify the importance of the three predictors influencing Innovative University Competency. The SHAP results reveal the relative influence of each variable by computing the average absolute Shapley values across all observations. Variables with higher average SHAP values contribute more significantly to predicting IUC.

Table 10. SHAP Feature Importance Results.

Predictor	Mean SHAP Value	Normalized Importance (%)	Rank
AI Competency (AIC)	0.412	100	1
Strategic Intelligence (SI)	0.348	84.5	2
Soft-Skill Competency (SSC)	0.301	73.1	3

presents the SHAP feature-importance results from the ANN model. The analysis indicates that AI Competency exhibits the highest SHAP importance score, suggesting that technological capability and digital expertise are the most influential factors in enhancing innovative competency within universities. Strategic Intelligence ranks second, highlighting the importance of strategic decision-making and knowledge management in fostering innovation. Soft-Skill Competency, although slightly less important, remains a significant contributor to collaborative problem-solving, leadership, and communication within academic institutions. Figure 5 presents the SHAP-based interpretation of the ANN model. The SHAP summary plot shows the distribution and magnitude of each predictor’s contribution to the model output, indicating that AI Competency has the strongest influence on Innovative University Competency, followed by Strategic Intelligence and Soft-Skill Competency. The SHAP dependence plot further illustrates a clear positive relationship between AI Competency and its SHAP values, suggesting that higher levels of AI capability significantly increase the predicted level of Innovative University Competency.

4.3.4. Comparison Between SEM and ANN Analysis

This study employed a hybrid analytical approach, integrating SEM and ANN techniques, to examine the relationships among AI Competency, Soft-Skill Competency, Strategic Intelligence, and Innovative University Competency. The purpose of combining these two methods is to leverage the strengths of both statistical modeling and machine learning in order to obtain robust theoretical explanations and strong predictive performance. SEM was first applied to test the hypothesized relationships within the conceptual model and to validate the constructs’ measurement properties. Through CFA, the measurement model demonstrated satisfactory reliability and validity, as indicated by acceptable values of factor loadings, composite reliability, and average variance extracted. The structural model further revealed significant causal relationships among the constructs. Specifically, the SEM results indicate that AI Competency and Soft-Skill Competency positively influence Strategic Intelligence, while Strategic Intelligence significantly contributes to Innovative University Competency. These findings provide theoretical evidence supporting the importance of technological capability and human competencies in enhancing innovation within higher education institutions. Although SEM is effective for testing theoretical relationships, it is primarily based on linear assumptions and may not fully capture complex nonlinear interactions among variables. Therefore, ANN analysis was conducted as a complementary technique to evaluate the predictive capability of the significant predictors identified in the SEM stage. The ANN model demonstrated strong predictive performance, as reflected by low RMSE and MAE values and a high coefficient of determination (R²). The close similarity between training and testing errors also indicates that the model generalizes well to unseen data, confirming the neural network’s robustness. Furthermore, sensitivity and SHAP feature-importance analyses from the ANN model revealed the relative predictive contribution of each variable. The results consistently show that AI Competency is the most influential predictor of Innovative University Competency, followed by Strategic Intelligence and Soft-Skill Competency. This ranking aligns with the SEM findings, which also highlight the significant role of AI-related capabilities in driving innovation outcomes. The comparison between SEM and ANN demonstrates that the two analytical approaches are complementary rather than competing. SEM provides theoretical validation and causal interpretation of relationships among constructs, whereas ANN enhances predictive accuracy and identifies nonlinear patterns within the data. The hybrid SEM–ANN framework, therefore, offers a more comprehensive understanding of the determinants of Innovative University Competency. By combining explanatory and predictive analytics, this approach strengthens the reliability of the findings and provides valuable insights for policymakers and university administrators seeking to develop innovation-oriented capabilities in the era of artificial intelligence.

5. Discussions

AI competency and soft skills jointly enhance research performance by integrating technical expertise with collaborative and adaptive capabilities. The findings reinforce the complementary role of AI competency and soft skills in enhancing research publication performance. Rather than acting as isolated predictors, these capabilities jointly influence innovation capability and strategic agility, which serve as key mediating mechanisms. This integrated interpretation extends prior research by demonstrating that technical expertise alone is insufficient without the supporting role of interpersonal and cognitive skills in academic research environments. The results confirm that AI Competency exerts a significant positive effect on Innovative University Competency, underscoring the critical role of digital and analytical capabilities in modern universities. Institutions equipped with advanced AI-related skills, technological infrastructure, and data-driven decision-making capabilities are better positioned to develop innovative research outputs, adaptive learning systems, and strategic academic services. This finding reinforces prior literature that emphasizes the transformative potential of artificial intelligence to reshape higher education ecosystems, particularly by enhancing knowledge management, research productivity, and intelligent learning environments. In addition, the findings reveal that Soft-Skill Competency significantly contributes to Strategic Intelligence, which in turn enhances Innovative University Competency. This highlights the indispensable role of human-centric capabilities in navigating complex and dynamic academic environments. These competencies enable university stakeholders to interpret technological opportunities, facilitate interdisciplinary collaboration, and respond effectively to institutional and environmental changes. This result aligns with HCT by demonstrating that intangible human capabilities are fundamental drivers of organizational adaptability and innovation. Furthermore, the study identifies Strategic Intelligence as a key mediating mechanism that strengthens the translation of both AI competency and soft skills into innovation outcomes. Strategic intelligence mediates the relationship by transforming AI competency and soft skills into actionable strategic outcomes. Strategic intelligence reflects an institution’s capacity to anticipate environmental changes, synthesize complex information, and align decisions with long-term objectives. Universities with higher levels of strategic intelligence are better able to integrate technological advancements into institutional strategies, thereby enhancing innovation performance. This finding extends DCT by empirically demonstrating how sensing, seizing, and transforming processes operate through measurable organizational competencies. Importantly, this study advances the discussion by highlighting the interplay between machine learning techniques, analytical outcomes, and strategic capability development. Machine learning methods (embedded within the SEM–ANN framework) enhance predictive accuracy, pattern recognition, and decision support. While SEM validates the structural relationships among constructs, ANN complements this by identifying the relative importance and nonlinear effects of key predictors. However, the findings indicate that the effectiveness of machine learning is contingent upon the presence of enabling organizational capabilities. Specifically, AI competency facilitates the effective use and interpretation of machine learning outputs, soft skills support the translation of analytical insights into coordinated institutional actions, and strategic intelligence ensures alignment with long-term innovation strategies. This integrated perspective positions machine learning as an enabling technological layer, operationalized through human and organizational capabilities, ultimately contributing to the development of innovative university infrastructure, including data-driven governance systems, adaptive learning environments, and digitally enhanced decision-making processes. From a practical standpoint, the findings suggest that universities seeking to enhance their innovation capacity should adopt a holistic approach to capability development. Investments in AI technologies and digital infrastructure must be complemented by initiatives to strengthen soft skills and strategic intelligence among academic staff and institutional leaders. Such a balanced capability configuration enables universities to effectively leverage technological advancements while maintaining human-centered innovation processes. Although the empirical analysis is based on data from Mahasarakham University, the proposed model demonstrates strong potential for generalization. The framework is grounded in well-established theoretical foundations, which enhance its applicability across diverse higher education contexts. Nevertheless, the generalizability of the findings may depend on contextual factors such as institutional readiness, technological maturity, organizational culture, and the availability of skilled human capital. Therefore, while the model is transferable, its implementation should be adapted to align with the specific strategic priorities and environmental conditions of different institutions. Beyond empirical findings, this study makes significant theoretical contributions by moving beyond the incremental integration of existing theories and proposing a multi-level capability framework. First, it reconceptualizes AI competency and soft skills as complementary operational capabilities, thereby extending RBV and HCT by demonstrating how technological and human resources interact to drive innovation. Second, it introduces strategic intelligence as a meta-capability, which integrates, aligns, and transforms operational capabilities into strategic outcomes. This conceptualization advances DCT by operationalizing dynamic capability processes into measurable constructs within the higher education context. Third, the study bridges the gap between technology-centric and human-centric perspectives in AI research by providing a unified framework that explains how they interact within institutional settings. Finally, by empirically validating the mediating role of strategic intelligence, the study contributes to theory development by emphasizing the importance of capability orchestration in translating digital transformation initiatives into sustainable innovation outcomes. The results indicate that AI competency is positively associated with innovative university competency, suggesting that institutions with higher levels of AI-related capabilities tend to exhibit stronger innovation performance. Similarly, soft-skill competency is significantly related to strategic intelligence, which in turn is associated with innovative university competency. These findings suggest that human-centric capabilities and strategic decision-making processes are closely linked to institutional innovation outcomes.

Future research is encouraged to extend this framework by conducting comparative studies across multiple universities and geographical regions to further enhance external validity. Additionally, longitudinal studies may provide deeper insights into how AI-driven capabilities evolve over time and influence long-term institutional performance.

6. Conclusions

This study aimed to examine the influence of AI Competency and Soft-Skill Competency on Innovative University Competency, with Strategic Intelligence serving as a key mediating capability, using an integrated SEM-ANN analytical framework. The integration of these two analytical approaches enabled the study to provide both theoretical explanation and predictive insight into the determinants of innovation capability within universities.

The SEM results confirm that AI Competency and Soft-Skill Competency significantly contribute to Strategic Intelligence, which subsequently enhances Innovative University Competency. These findings suggest that universities must simultaneously develop technological capabilities and human competencies to effectively respond to the challenges of digital transformation. AI Competency enables institutions to utilize advanced technologies, data-driven decision-making, and intelligent systems to improve research productivity and institutional innovation. Meanwhile, Soft-Skill Competency supports collaboration, communication, leadership, and knowledge sharing among academic personnel, which strengthens strategic decision-making and organizational adaptability. The ANN analysis further validated these findings by examining the nonlinear predictive relationships among the constructs. The results demonstrate strong predictive performance, as reflected in low RMSE and MAE and high R². The sensitivity analysis and SHAP feature importance results indicate that AI Competency is the most influential predictor of Innovative University Competency, followed by Strategic Intelligence and Soft-Skill Competency. These findings emphasize that technological capability plays a dominant role in driving innovation in higher education institutions. The hybrid SEM–ANN approach provides a more comprehensive understanding of innovation capability within universities by combining causal explanation and predictive modeling. The results suggest that universities seeking to enhance their innovation performance should prioritize developing AI-related competencies while also strengthening strategic intelligence and soft skills among academic staff and institutional leaders. Although the study is based on data from Mahasarakham University, the proposed model demonstrates potential applicability to other higher education institutions undergoing digital transformation. However, contextual factors such as institutional readiness, technological infrastructure, and organizational culture should be considered when applying the model in different settings. In conclusion, the findings highlight that innovation in universities is driven not solely by technological resources but also by the integration of digital capability, human competence, and strategic leadership. By fostering these capabilities, universities can enhance their capacity for innovation and remain competitive in the rapidly evolving knowledge economy. It is important to note that this study is based on cross-sectional data, which limits the ability to establish causal relationships among the variables. While the structural model identifies significant associations and predictive relationships between AI competency, soft skills, strategic intelligence, and innovative university competency, these findings should not be interpreted as evidence of causality. Future research employing longitudinal or experimental designs is recommended to validate the causal mechanisms underlying these relationships.