Malaysia’s rapid industrialization has intensified resource depletion and environmental degradation, heightening the need for circular innovation driven by artificial intelligence (AI) and knowledge-based solutions. However, few empirical studies have examined how AI incorporation, circular innovation, and entrepreneurial learning enhance the process of circular product development through effective knowledge transfer in emerging economies. To overcome this gap, this paper develops a moderated mediation model based on the Resource-Based View (RBV), Dynamic Capabilities Theory (DCT), and the Knowledge-Based View (KBV). A cross-sectional survey involving 450 managers of manufacturing and technology-intensive companies in Malaysia was conducted, and the resultant data were processed via Partial Least Squares Structural Equation Modeling (PLS-SEM). The empirical results suggest that Circular Innovation Orientation (CIO), AI-Driven Decision Support (AIDS), and Entrepreneurial Learning Behavior (ELB) have a significant positive impact on circular product development, both directly and indirectly, due to increased Knowledge Transfer Effectiveness (KTE). The quality of digital infrastructure and support for innovation at the institutional level reinforces these indirect effects, highlighting their critical role in translating internal capabilities into tangible circular innovation results. Theoretically, the paper incorporates the RBV, DCT, and the KBV to form a unified circle of innovation and explains how digital and institutional ecosystems moderate the relationship between capabilities and performance. The findings also have practical implications for Malaysian managers and policymakers, who should invest in AI, increase digital preparedness, and strengthen the institutional system. The research contributes to the existing body of work regarding the circular economy (CE) by introducing a context-specific model that can be reproduced in other emerging economies that are engaged in sustainable industrial change.
The blistering increase in industrial production over the past 10 years has increased global concerns regarding the degradation of the environment, the lack of resources, and the formation of waste; this has challenged the traditional linear economic framework, which is based on extraction, production, consumption, and disposal (Bocken et al., 2016; Kirchherr et al., 2018). The circular economy (CE) has become a key strategic answer to these issues, promoting regenerative models, maximizing resource use, increasing the lifecycle of products, and reducing industrial emissions (Agrawal et al., 2025; Brogi & Menichini, 2023). Simultaneously, artificial intelligence (AI) and digital technologies are changing the production ecosystem by simplifying predictive resources, intelligent product-life cycle monitoring, and data-based sustainability (Rashid et al., 2024; Wang & Zhang, 2025; Wilson et al., 2022). However, despite the technological potential, many companies are facing challenges in realizing the benefits of AI in operationalization in achieving the results of a circular product development, especially in emerging economies that are pursuing sustainable industrial transformation (Gnekpe & Plantec, 2023; Naeem et al., 2025).
The discrete role of AI, innovation orientation, and knowledge-based capabilities in promoting sustainability is well-known in existing literature (Bag & Pretorius, 2022; Cheah et al., 2022; Cheng et al., 2024). However, the way these strategic and technological capabilities interact and improve the performance of circular product development, especially through knowledge-transfer processes, remains unexplored. Existing studies have generally assessed these constructs independently, thus ignoring their interrelations in a unified theory (Pascucci et al., 2024; Tutore et al., 2024). Furthermore, there is little empirical evidence for the moderating impact of digital infrastructure and institutional support in determining the performance of firm-level capabilities compared to CE outcomes (Perotti et al., 2024; Reuter, 2016). This inconsistency between theoretical statements and their practical application highlights the importance of developing a combined model that can summarize the organizational, technological, and institutional antecedents of circular innovation.
The case of Malaysia offers an interesting empirical study of circular innovation due to its strategic focus on sustainable industrialization via national policies, including Industry 4.0, the National Circular Economy Roadmap, and the Green Technology Master Plan (Al-khatib et al., 2024; Ali et al., 2023). As an emerging, manufacturing-intensive economy with an increasing level of carbon emissions, resource dependency, and regulatory pressure, Malaysia faces a dual challenge of maintaining economic competitiveness and transitioning to a CE (Chen et al., 2022; Danish & Senjyu, 2023; Montes-Pineda & Garrido-Yserte, 2024). Although there has been a growing policy focus on these issues, there is limited empirical evidence on how Malaysian companies are using AI-based capabilities, knowledge transfer, and institutional support to achieve circular product strategies (Adam et al., 2025; Sun et al., 2025). Recent evidence from emerging economies also shows that digital platforms, innovation networks, and institutional support can jointly shape circular economy adoption through moderated mediation mechanisms (Sadiq et al., 2026). Malaysia, therefore, provides a perfect case study for exploring the interaction between organizational capabilities and external ecosystems and their impact on the development of CE-oriented innovation in emerging economies.
This study hypothesizes and empirically estimates a moderated-mediation model that explains the positive performance of circular product development via Circular Innovation Orientation (CIO), AI-Driven Decision Support (AIDS), and Entrepreneurial Learning Behavior (ELB). Key factors in this process are knowledge-transfer effectiveness, the moderating effects of Digital Infrastructure Quality (DIQ), and Institutional Support for Innovation (ISI). The study is based on the Resource-Based View (RBV), Dynamic Capabilities Theory (DCT), and the Knowledge-Based View (KBV), and combines technological and knowledge-based capabilities into one CE innovation model (Chesbrough & Appleyard, 2007; Mariani et al., 2023). In practice, it provides evidence-based advice for Malaysian policy and corporate executives on how to use AI and digital ecosystems to hasten the CE shift and strengthen competitiveness within the manufacturing industry.
The functioning of the CE in Malaysia depends on the organization and rules that are in place. Bocken et al. (2016) and Reuter (2016) found that DIQ is particularly useful for supporting circular solutions and sharing knowledge. Effective use of circular operations and sharing of knowledge requires clear, trackable, and connected digital networks (Sun et al., 2025; Wilson et al., 2022). Good policies, support from the government, and ties with industries also help establish a culture that encourages businesses to be more environmentally friendly (Perotti et al., 2024; Schaltegger & Wagner, 2017). Malaysia has recently introduced Industry 4.0, green technologies, and eco-innovation policies, further illustrating the role regulatory systems play in supporting circular changes (Al Halbusi et al., 2024).
Even though the adoption of AI, a focus on innovation, learning behaviors, and the CE results have been investigated in previous research, either individually or in simplified forms, they only explain circular innovation performance in part. In practical terms, companies seldom transform technology or strategic ability into circular product results via linear paths alone; instead, these results are determined by the extent to which these capabilities are translated into common organizational knowledge, as well as the extent to which this knowledge is supported by digital and institutional conditions. The gap in the literature is thus not whether these factors have an effect or not, but how they interact in a process-based explanation of circular innovation. This is particularly important in emerging economies, where the impact of internal firm resources can be significantly impacted by digital unevenness, capability differences, and institutional variability. The strength of the current model is its demonstration of the interaction between internal capability, knowledge processes, and contextual enablers, thus explaining why certain firms have achieved more successful circular product development results than others.
The present study offers several major contributions by addressing these types of relationships. Firstly, it outlines factors that produce circular innovation in emerging economies, especially in Southeast Asia. Secondly, it provides officials and business leaders in Malaysia with advice on the use of AI, digitalization, and organizational learning to support sustainable development objectives. Finally, given that incorporating CE practices is still a tough decision as well as an opportunity, the research results will prove valuable for organizations working in similar institutional and industrial settings at a global level.
Literature reviewTheoretical foundationThe theoretical background of this research is based on three interconnected theories of strategic management, namely, the RBV, DCT, and the KBV. Together, these theoretical prisms provide an all-encompassing explanatory framework of how companies transform internal strengths and external facilitators into high-level Circular Product Development Performance (CPDP), especially in digitally mediated and sustainability-centered settings.
The RBV assumes that competitive advantage can be maintained by firm-specific resources, which are valuable, rare, inimitable, and non-substitutable (Adam et al., 2025; Chesbrough & Appleyard, 2007). Capabilities related to circular innovation and AI-based decision-making are also considered to be strategic resources within the CE context, helping companies to reduce waste, optimize resource utilization, and produce sustainable products (Ali et al., 2023; Soriano-Pinar et al., 2026). However, the RBV alone cannot explain how the firms dynamically change and redesign these capabilities according to environmental pressures and technological shocks (Kumar et al., 2024; Wang & Zhang, 2025).
DCT is a derivative of the RBV that focuses on the ability to perceive opportunities, pursue innovation, and reorganize resources in response to changing market and environmental conditions (Agrawal et al., 2025). AI technologies enable dynamic sensing, thus facilitating predictive analytics and intelligent decision-making throughout the product lifecycle (Porter & Heppelmann, 2014; Rashid et al., 2024). At the same time, ELB helps companies to reorganize internal operations to incorporate circularity into product development (Fusillo et al., 2025). Once again, however, although DCT explains transformation capability, it fails to clarify how these capabilities are converted into performance results.
The final theory, the KBV, fills this gap by suggesting that knowledge is the most strategic, salient asset in the process of creating innovation and maintaining competitive performance (Pascucci et al., 2024). The effectiveness of organizational knowledge transfer in the CE context defines the internalization and application of AI insights, innovation strategies, and learning behaviors to deliver circular product outcomes (Cantu & Tunisini, 2023; Cheng et al., 2024). The KBV also highlights the fact that the integration of knowledge is determined by the technological infrastructure and institutional ecosystems (Meena et al., 2025; Sun et al., 2025).
Although all three theories have their individual merits, their shortcomings require a combined approach. The RBV acknowledges internal strategic resources but does not take adaptability into consideration; DCT explicates the capabilities of transformation without specifying the operationalization of knowledge; and the KBV acknowledges the existence of knowledge as the linkage mechanism but its applications require enabling conditions. Combining these theories, the current study develops a complex model that can explain the impact of organizational capabilities (CIO, AIDS, ELB) on circular performance in terms of transferring knowledge. It also considers whether DIQ and institutional support enhance this impact. This combination provides a conceptually sound approach to using AI and knowledge-based capabilities to hasten the transition to CE, especially for companies in developing countries such as Malaysia (Al Halbusi et al., 2024; Gama & Magistretti, 2023; Naeem et al., 2025).
Applying the paper’s theoretical rationale to these organizational constructs, the RBV clarifies why CIO, AIDS, and ELB should be considered as strategic internal capabilities that can generate value in the development of circular products when they are precious, hard to copy, and integrated into existing routines. DCT builds on this reasoning by explaining that these capabilities are reconfigurable capabilities that allow firms to identify sustainability opportunities, capture the possibilities of circular innovation, and transform their processes in the face of technological and environmental change. In addition, the KBV articulates that the effectiveness of these capabilities in terms of performance is determined by a given firm’s capacity to transfer, combine, and utilize knowledge across functions. In this regard, Knowledge Transfer Effectiveness (KTE) is the fundamental explanatory construct that relates strategic capabilities to CPDP. This further integration is enhanced by the moderated mediation framework, which demonstrates that the conversion of capabilities does not occur in all contexts but becomes more likely in favorable digital and institutional conditions. Within this framework, the RBV describes the process of capability possession, DCT describes the process of capability deployment and adaptation, and the KBV describes the process of translating capabilities into performance based on knowledge processes. This combined view constitutes this study’s theoretical contribution as it links internal strategic resources, reconfiguration in dynamics, and integration of knowledge into a single circular model of innovation. This approach reflects recent studies that have emphasized the importance of AI capabilities, digitalization, and knowledge integration in enhancing sustainable organizational performance and CE practices (Liu et al., 2026; Sun et al., 2026).
Literature review and hypothesis developmentThe literature on AI, circular innovation, and organizational capabilities is promising, but not entirely consistent. Existing studies have noted that AI-enabled systems, innovation orientation, and the product development outcomes are directly enhanced by sustainability and product development, but others indicate that these impacts are contingent on organization preparedness, digital maturity, absorptive processes, and the overall institutional context. This contradiction points to the fact that positive effects are not supposed to be universal; rather, the theoretical question of interest involves the circumstances in which the conversion of internal capabilities into circular results is effective and the mechanisms through which this conversion takes place. Consequently, the present study assumes a more conditional viewpoint by considering knowledge transfer as a major explanatory route and the digital and institutional conditions as key boundary factors in determining the strength of these relationships.
Circular innovation orientation and circular product development performanceCIO refers to the strategic dedication of a firm to a shift in production paradigm aimed at producing regenerative value through eco-design, reuse, remanufacturing, and extending the life of products (Bocken et al., 2016; Cantu & Tunisini, 2023). This construct is deeply rooted in the RBV and its focus on unique, valuable, and non-substitutable capabilities (Adam et al., 2025; Chesbrough & Appleyard, 2007). By focusing on circularity in their innovation approaches, firms thus become more resilient by limiting their reliance on virgin materials, reducing exposure to regulatory risks, and deriving value from secondary resource flows (Agrawal et al., 2025). Practical experience has shown that circular orientation enhances the ability to create superior products that are eco-friendly and increase manufacturing efficiency (Brogi & Menichini, 2023; Perotti et al., 2024). However, conflicting evidence suggests that in emerging economies, institutional voids, technological barriers, and the lack of digital capabilities may prevent the implementation of circular innovation (Danish & Senjyu, 2023; Liang et al., 2023). Therefore, the impact of CIO on product outcomes is not unconditional and is context-dependent. For example, in Malaysia, governmental policies are oriented toward green economy goals. Accordingly:
H1: Circular Innovation Orientation has a positive effect on Circular Product Development Performance.
With AI‑based decision-support systems, sustainability decisions made on the basis of data are improved with real-time analytics, predictive maintenance, and material flows optimization (Porter & Heppelmann, 2014; Wilson et al., 2022). Viewing AI through the prism of DCT allows firms to identify opportunities, grasp innovation avenues, and reorganize resources to generate circular product architectures (Wang & Zhang, 2025). It has been shown that AI helps companies to predict material recovery opportunities and streamline processes, enhancing product circularity (Cheng et al., 2024; Rashid et al., 2024). However, other researchers have argued that the transformative effect of AI depends on organizational preparedness and knowledge systems; if these capabilities are not yet well-developed, the circular benefits of AI cannot be achieved (Gama & Magistretti, 2023; Zong & Guan, 2024). Although Malaysia’s manufacturing industry is advancing AI investment as part of its Industry 4.0 policy, there is a lack of empirical evidence regarding its effects on circular outcomes. Consequently:
H2: AI-Driven Decision Support positively influences Circular Product Development Performance.
ELB is a measure of the capacity of organizations to attain new knowledge, internalize this knowledge, and translate it into their operations and product strategies (Zahra et al., 2000). In line with the KBV, the most strategically relevant resource is knowledge; the ability to create a continuous learning environment in firms results in a higher level of absorptive capacity toward circular innovation (Pascucci et al., 2024). Empirical evidence shows that learning-based companies are in a better position to convert knowledge of the environment into novel product designs in accordance with CE principles (Klofsten et al., 2024). Nevertheless, the literature also emphasizes the fact that digital tools and institutional mechanisms should support learning to transform it into performance outcomes (Fusillo et al., 2025). Entrepreneurial learning plays a key role in the process of transforming incremental changes into circular innovation in emerging economies such as Malaysia. Accordingly:
H3: Entrepreneurial Learning Behavior has a positive effect on Circular Product Development Performance.
KTE refers to a firm’s ability to transfer, receive, and utilize pertinent knowledge within its various departments and functions (Adam et al., 2025). According to the KBV, performance is dependent on the effective conversion of knowledge into strategic outcomes (Pascucci et al., 2024). Companies that have a strong CIO are more likely to be capable of promoting cross-functional cooperation, thus facilitating the sharing of design and sustainability knowledge (Perotti et al., 2024). Similarly, AI systems facilitate successful knowledge exchange by creating actionable intelligence (Chen et al., 2022). Entrepreneurial learning develops an entrepreneurial culture of knowledge-sharing, which improves the diffusion of circular practices (Klofsten et al., 2024). Hence:
H4: Circular Innovation Orientation has a positive impact on Knowledge Transfer Effectiveness.
H5: AI-Driven Decision Support has a positive impact on Knowledge Transfer Effectiveness.
H6: Entrepreneurial Learning Behavior has a positive effect on Knowledge Transfer Effectiveness.
KTE is proposed as a major mediating process in which internal capabilities are transformed into real circular performance. Despite the potential of strategic capabilities to drive innovation, companies can only achieve strong performance results when these capabilities are shared and put into practice (Gama & Magistretti, 2023; Pascucci et al., 2024). Without successful knowledge transfer, circular strategies will fail to produce meaningful product development results. Accordingly:
H7: Knowledge Transfer Effectiveness mediates the relationship between Circular Innovation Orientation and Circular Product Development Performance.
H8: Knowledge Transfer Effectiveness mediates the relationship between AI-Driven Decision Support and Circular Product Development Performance.
H9: Knowledge Transfer Effectiveness mediates the relationship between Entrepreneurial Learning Behavior and Circular Product Development Performance.
DIQ helps companies to accommodate data-intensive operations and incorporate AI insights into the work of various departments (Reuter, 2016). From the perspective of DCT, high-quality digital infrastructure helps to increase capacity in terms of reorganizing resources according to real-time knowledge (Wang & Zhang, 2025). With developed digital ecosystems, the positive effects of KTE on circular outcomes are multiplied because knowledge is successfully integrated into the product development process rather than merely created. On the other hand, in settings with low levels of digital maturity, technological fragmentation may impede the transfer of knowledge (Soriano-Pinar et al., 2026). Therefore:
H10: Digital Infrastructure Quality positively moderates the relationship between Knowledge Transfer Effectiveness and Circular Product Development Performance, such that the relationship becomes stronger when Digital Infrastructure Quality is higher.
H11: Digital Infrastructure Quality positively moderates the relationship between Circular Innovation Orientation and Knowledge Transfer Effectiveness, such that the relationship becomes stronger when Digital Infrastructure Quality is higher.
H12: Digital Infrastructure Quality positively moderates the relationship between AI-Driven Decision Support and Knowledge Transfer Effectiveness, such that the relationship becomes stronger when Digital Infrastructure Quality is higher.
H13: Digital Infrastructure Quality positively moderates the relationship between Entrepreneurial Learning Behavior and Knowledge Transfer Effectiveness, such that the relationship becomes stronger when Digital Infrastructure Quality is higher.
ISI refers to regulatory incentives, government funding, and government-to-business collaboration, which together ease circular adoption (Montes-Pineda & Garrido-Yserte, 2024). Institutional Theory states that institutional forces influence the strategic decision-making process by facilitating or restricting innovation (Perotti et al., 2024). In Malaysia, the existence of strong regulatory frameworks like the National Circular Economy Roadmap ensures the creation of a favorable environment with regard to circular innovation. According to Al Halbusi et al. (2024), institutional support contributes to the ability of firms to make effective use of AI, learning behaviors, and innovation capabilities. Consequently:
H14: Institutional Support for Innovation positively moderates the relationship between Knowledge Transfer Effectiveness and Circular Product Development Performance, such that the relationship becomes stronger when Institutional Support for Innovation is higher.
H15: Institutional Support for Innovation positively moderates the relationship between Circular Innovation Orientation and Knowledge Transfer Effectiveness, such that the relationship becomes stronger when Institutional Support for Innovation is higher.
H16: Institutional Support for Innovation positively moderates the relationship between AI-Driven Decision Support and Knowledge Transfer Effectiveness, such that the relationship becomes stronger when Institutional Support for Innovation is higher.
H17: Institutional Support for Innovation positively moderates the relationship between Entrepreneurial Learning Behavior and Knowledge Transfer Effectiveness, such that the relationship becomes stronger when Institutional Support for Innovation is higher.
As circular innovation in emerging economies is multi-layered in nature, this study utilizes an extensive moderated mediation framework. There is no single organizational factor that drives circular product development, and the relationships cannot be sufficiently explained by direct linear relationships. Instead, the focus is on the interplay between internal capabilities, knowledge-conversion mechanisms, and external enabling conditions. CIO, AIDS, and ELB are unique internal capabilities in this study that are associated with strategic intent, technological intelligence, and learning adaptability. Nonetheless, these capabilities cannot produce circular product results on their own unless the associated knowledge and activities are moved efficiently between organizational units. Consequently, this study adopts KTE as the key mediating factor. DIQ and ISI are also included as moderators because these two constructs constitute two levels of contextual enablement. DIQ embodies the technological preparedness that underpins the codification, integration, and real-time sharing of knowledge, and ISI embodies the wider regulatory, policy, and ecosystem environment that legitimizes and enables innovation practices. Accordingly, DIQ is a technological-operational boundary condition, whereas ISI is an institutional-environmental boundary condition. Modeling these constructs simultaneously facilitates a more accurate explanation of the simultaneous occurrence of similar internal capabilities with stronger or weaker circular outcomes based on the surrounding digital and institutional environment. This method is especially applicable to emerging economies, where digital maturity and policy support are not evenly distributed and may determine the success of sustainability-based innovation.
This combination of the RBV, DCT, and the KBV means that the capabilities of firms, combined with knowledge transfer activities and favorable conditions, can provide amplified circular results (Ali et al., 2023; Gama & Magistretti, 2023). Although KTE is an effective mediator between firm capabilities and performance, its efficacy is enhanced in an environment that is characterized by a strong digital infrastructure and institutional support. This is in line with recent studies that have highlighted the significance of capability-environment congruency in achieving circular transition (Naeem et al., 2025). Therefore:
H18: Digital Infrastructure Quality strengthens the indirect effects of Circular Innovation Orientation, AI-Driven Decision Support, and Entrepreneurial Learning Behavior on Circular Product Development Performance through Knowledge Transfer Effectiveness.
H19: Institutional Support for Innovation strengthens the indirect effects of Circular Innovation Orientation, AI-Driven Decision Support and Entrepreneurial Learning Behavior on Circular Product Development Performance through Knowledge Transfer Effectiveness.
The literature indicates that circular innovation cannot be sufficiently described in terms of isolated direct relationships. While existing studies have recognized significant drivers, there is less evidence regarding how these drivers are converted into circular results and whether they are more or less impactful under certain conditions. This restriction is the reason to present the current model, which describes the connection between inner capabilities and circular product development via knowledge transfer and whether digital and institutional conditions support this process. Fig. 1 outlines the theoretical framework, which describes the interactions between a firm’s internal factors, knowledge, and the environment.
Data and methodologyResearch design and rationale
This study employed a quantitative and cross-sectional survey to examine how AI-based capabilities, circular innovation, entrepreneurial learning, and effective knowledge transfer affect the performance of circular product development of Malaysian manufacturing companies. Partial Least Squares Structural Equation Modeling (PLS-SEM) was selected for several methodological reasons. First, PLS-SEM is suitable for theory formulation and exploratory research based on complex mediation models with multiple latent constructs (Hair et al., 2021). Second, in line with the aim of predicting the results of circular products, it is particularly well-suited to non-normal data and prediction-based research (Agrawal et al., 2025; Ali et al., 2023). Third, PLS-SEM enables hierarchical components and the impact of interaction to be modelled, which facilitates a robust assessment of the proposed moderated mediation correlations (Sun et al., 2025).
Population, sampling method, and determination of sample sizeThe target population includes managers and top managers of Malaysian manufacturing and technology-oriented companies registered under the Federation of Malaysian Manufacturers (FMM) and the Malaysia Digital Economy Corporation (MDEC). A purposive sampling strategy was used to ensure that the respondents were well informed regarding organizational innovation capabilities, AI integration, and sustainability strategies. According to the power analysis approach suggested by Cohen (1992), a sample size of 368 is needed to identify medium effect sizes at a 95% level of confidence (power = 0.80). In order to maximize the generalizability and reduce the possible non-responses, 650 questionnaires were sent out, which led to 450 usable responses. This total is above the minimum advised for PLS-SEM (Hair et al., 2021). This sample size is also acceptable according to the 10X rule, considering that there is an upper limit to the number of arrows leading to a latent construct in the model (Kock & Hadaya, 2018).
Instrument development and justification of measurement scaleReflective multi-item scales based on validated previous studies were used to measure all constructs and ensure the content validity of the scale (Cantu & Tunisini, 2023; Pascucci et al., 2024; Rashid et al., 2024). A 5-point Likert scale ranging from strongly disagree to strongly agree was used. This scale minimizes fatigue among respondents, enhances the level of accuracy in responses in managerial research, and offers the best psychometric characteristics for PLS-SEM analysis in cross-sectional research (Dawes, 2008). Previous studies with similar constructs focusing on Malaysia and Southeast Asia have predominantly used 5-point scales (Ali et al., 2023; Sun et al., 2025), which guarantees consistency and comparability in the methods.
The survey tool was validated in two stages. To achieve clarity, face validity, and relevance to the context, the questionnaire was reviewed by three domain experts in academia and industry. Second, a pilot test was carried out on 30 respondents to test reliability. Internal consistency was verified by the fact that the Cronbach’s alpha for all constructs was greater than the acceptable level of 0.70.
Data collection procedureThe information was gathered from March to July 2024 and comprised a hybrid strategy of distributing the emails and using online survey tools in collaboration with industry associations. The respondents were guaranteed anonymity and confidentiality to reduce evaluation fear and enhance the accuracy of responses. Participation was encouraged via follow-ups every two weeks. Of the 650 questionnaires mailed, 536 responses were obtained, corresponding to an 82.4% response rate. After eliminating incomplete and inconsistent responses, which were detected using missing values and straight-lining, 450 valid responses were identified.
Both statistical and procedural remedies were used to reduce the possible impact of common method bias. Procedurally, the respondents were assured of anonymity and confidentiality, which helped to promote honest reporting. The questionnaire also emphasized that there were no right or wrong answers, thus reducing the chances of socially desirable responses. Moreover, the items were well-arranged and in a randomized order to minimize response pattern bias and consistency motifs. The instrument was phrased clearly and concisely to reduce ambiguity and misinterpretation of items.
Harman’s single-factor test was carried out statistically and the first unrotated factor explained 34.1% of the total variance, which is not as high as the threshold typically applied to signal a severe common method bias issue (Podsakoff et al., 2003). Complete collinearity analysis was also conducted, and the variance inflation factor (VIF) values were <3.3; this indicates that common method bias and multicollinearity were not likely to affect the structural estimates (Kock, 2015). These combined procedural controls and statistical checks provide sufficient evidence that common method bias is not a significant threat to the validity of the study results.
Data analysis techniqueSmartPLS 4.0 was applied to the data analysis in two steps. The first was Measurement Model Evaluation, which involved testing the indicator reliability, internal consistency (Cronbach’s alpha and composite reliability (CR)), convergent validity (Average Variance Extracted (AVE)), and discriminant validity (HTMT ratio) as suggested by Henseler et al. (2015). The second was Structural Model Assessment, which involved hypothesis testing of relationships through bootstrapping with 5000 resamples and testing of path coefficients, R2 values, and moderating and mediating effects.
PLS-SEM was primarily chosen due to its ability to deal with non-normal data and research that is focused on prediction; however, it is suitable for estimating models that involve more than two relationships (direct, mediating, and moderating). The current research includes several latent constructs that are measured reflectively, a mediation model based on KTE, and parallel moderation models focusing on DIQ and ISI. In these circumstances, PLS-SEM offers the flexibility of analysis required to address model complexity yet remains applicable to theory extension and investigation of the structural relationship between constructs in new research environments. Furthermore, the study describes variance in CPDP and KTE as opposed to testing a well-established covariance structure. Therefore, PLS-SEM, rather than covariance-based SEM, is an appropriate methodological tool for measuring the proposed structure and its conditional process relationships. This method is well-suited to the exploratory nature of the research, the non-normal distribution of survey data, and the complicated moderated mediation model (Ringle et al., 2015).
ResultsThe data analysis outlined in this section summarizes the main characteristics of the participating sample while also assessing the quality of the scale, assessing reliability, convergent validity, and discriminant validity. As a result, the structural model includes a complete description of the significance and power of the direct, mediating, and moderating effects between the study factors. Following this, the findings are explained using existing theories and literature to provide recommendations for best practices in terms of advancing circular products and sustainable innovation policies in developing countries.
Table 1 shows the findings of the convergent validity assessment for the measurement model, which includes seven latent constructs: AIDS, CIO, CPDP, DIQ, ELB, ISI, and KTE. All of these constructs are measured with several items and their outer loadings are all higher than the recommended 0.70, ranging from 0.843 to 0.921. This demonstrates that the individual items are good indicators of their underlying latent variables. All constructs have Cronbach’s alpha (α) values between 0.879 and 0.932, which means they are highly consistent and reliable. The CR for each scale is greater than 0.70, indicating that the scales are consistent. All AVE values are >0.50 (ranging from 0.738 to 0.809), which means that each construct explains a significant amount of variance in its indicators and supports convergent validity. These combined results prove that the measurement model is reliable and valid, thus forming a strong basis for the next steps in examining the structural relationships in SmartPLS. All in all, the tools used in the study are dependable and accurate, meaning the following structural model testing is trustworthy.
Convergent validity test (n = 450).
Table 2 presents the HTMT ratios for the correlations, which demonstrate whether the constructs in the measurement model are distinct from one another. Discriminant validity confirms that each construct is separate and sufficiently differentiated. Generally, HTMT scores should be 0.85 or less (or 0.90 if more conservative results are preferred) and values lower than this indicate good discriminant validity. The table shows that all HTMT values between AIDS, CIO, CPDP, DIQ, ELB, ISI, and KTE are below the suggested level, with the highest being 0.551 (between KTE and CPDP). These results mean that the constructs are unique and do not overlap significantly in the things they measure. Consequently, the model has discriminant validity and the chosen constructs are suitable for further analysis.
Table 3 shows the results of the Fornell-Larcker criterion, which is a popular method of checking discriminant validity in structural equation modeling. The diagonal values in the table are the square roots of the AVE for each construct, and the off-diagonal values indicate the relationships between the constructs. Discriminant validity exists when the square root of each construct’s AVE is bigger than any of its correlations with other constructs in the same row or column.
Fornell-Larcker criterion.
Note: Diagonal bold face values are √AVE.
In the present study, all the diagonal values (from 0.857 to 0.898) are greater than the off-diagonal correlations in the same row and column. The square root of AVE for KTE is 0.857, while the highest correlation it has with other constructs is 0.496 (CPDP). This shows that each construct is more similar to its own indicators than the indicators from other constructs. Consequently, the constructs are measurable in a way that makes them distinct and suitable for use in the structural analysis.
Table 4 provides the cross-loadings for each indicator (survey item) on all the latent constructs in the model. Cross-loading analysis is important as it assesses both discriminant validity and the reliability of the indicators. According to the accepted rule, every item should be more closely related to its intended (assigned) construct than to any other construct.
Cross loadings.
According to Table 4, every indicator is most strongly related to its own construct. For instance, every AIDS item (AIDS1–AIDS5) loads most strongly with the AIDS construct (values ranging from 0.861 to 0.921) and has significantly weaker loading with the other constructs. This is true for the remaining constructs, including CIO, CPDP, DIQ, ELB, ISI, and KTE. As a result, the indicators correctly capture the intended constructs and the model has sufficient construct validity and discriminant validity. The analysis confirms that the measurement model is strong and supports reliable analysis, with very little chance of cross-construct contamination or unclear item assignment.
Fig. 2 presents the visualized measurement and structural model results from SmartPLS. Within this model, each blue circle represents a latent construct. Each yellow rectangle represents an observed indicator (survey item) for a construct, and the loadings for those indicators are shown on the connecting arrows. The high outer loadings for all items (above the standard of 0.70) prove that each item is a reliable indicator of the intended latent variable, confirming convergent validity. For example, all DIQ indicators have loadings between 0.879 and 0.892, and the other constructs also have strong loadings for their items. These results further demonstrate the strength and reliability of the measurement model.
The structural model shows the latent constructs connected by arrows. The numbers on the arrows represent the estimated path coefficients. Solid lines show direct and significant relationships, and dashed lines (if they exist) suggest that the relationship is not significant or is being moderated. The R² values in the blue circles for both KTE (0.398) and CPDP (0.445) indicate that the model explains 39.8% of the variation in KTE and 44.5% of the variation in CPDP. The values of the path coefficients (for example, the effect from AIDS to KTE or KTE to CPDP) highlight which antecedents have the strongest effects within the model. In short, Fig. 2 demonstrates that the measurement model is reliable and the hypothesized relationships in the structural model are strong.
Tables 5 and 6 present the direct, mediating, moderating, and conditional indirect effects. Most hypothesized relationships were supported; however, the moderating effects of DIQ and ISI on the KTE → CPDP relationship were not significant. The main columns show the original sample estimate (O), sample mean (M), standard deviation (STDEV), t-statistics, and p-values for every structural path.
Path analysis results (n = 450).
Indirect effects.
The results indicate that AIDS, CIO, DIQ, ELB, ISI, and KTE all positively impact CPDP in a statistically significant manner (p < 0.05), with coefficients between 0.115 and 0.293. AIDS, CIO, DIQ, ELB, and ISI also positively affect KTE, meaning that they help improve how knowledge is shared within organizations. In addition, DIQ and ISI have a direct effect on both CPDP and KTE. The results also suggest that high DIQ and institutional support tend to increase the effect of the main predictors on knowledge transfer. Neither DIQ x KTE nor ISI x KTE has a significant effect on CPDP (p > 0.05), indicating that the effects of these factors are not strong at this stage.
The results show that KTE helps explain how organizational and contextual factors positively influence circular product development. Knowledge transfer is once again highlighted as a key factor by the AIDS -> KTE -> CPDP, CIO -> KTE -> CPDP, and ELB -> KTE -> CPDP indirect paths. Most of the paths have high t-statistics and low p-values, which confirms that these relationships are robust and statistically valid. Overall, the findings indicate that both organizational strengths and favorable conditions play a major role in promoting circular innovation and performance by ensuring effective knowledge sharing.
Fig. 3 shows the structural model outcomes from the PLS-SEM analysis, highlighting how the main constructs in the study are connected. The diagram demonstrates how the six antecedent constructs, (CIO, AIDS, ELB, DIQ, ISI, and KTE), are related to CPDP. All the observed indicators chosen for the constructs show good reliability and validity.
According to the model, the links between circular orientation, AI support, and entrepreneurial learning to both knowledge transfer and product development performance are significant, emphasizing the role of organizational strategy, technology, and learning in continuous innovation. It is evident that transferring knowledge from one part of the organization to another is a key way of transforming these capabilities into better performance. The diagram also highlights the role of digital infrastructure and institutional backing, which help increase the influence of these predictors on sharing knowledge. Nonetheless, the moderation effects of DIQ and ISI on the KTE → CPDP relationship are not significant, which means some contextual factors may not equally benefit all relationships. Ultimately, the structural model demonstrates that having an ideal environment and the requisite organizational capabilities helps organizations succeed in working toward innovation and sustainability.
DiscussionThe findings of this paper are strong empirical indicators underpinning the incorporation of the RBV, DCT, and the KBV to explain how firm-level capabilities translate to the CE approach in the Malaysian manufacturing industry. The strong positive relationship between CIO and CPDP supports the RBV hypothesis that a strategic orientation toward sustainable innovation is a useful and inimitable resource that can create a competitive advantage. This observation is consistent with recent empirical studies demonstrating that innovation practices entrenched in circular principles have better environmental and economic impacts (Agrawal et al., 2025; Cantu & Tunisini, 2023). It also responds to demands in the existing literature for an empirical validation of CIO beyond European settings by presenting evidence of an emerging Southeast Asian economy that is shifting from linear production systems to circular manufacturing ecosystems.
The relationship between AIDS and CPDP validates the theoretical assumption of DCT that digital technologies allow companies to identify the opportunities for sustainability, embrace resource-efficient solutions, and dynamically reorganize processes. This finding is aligned with the works of Rashid et al. (2024) and Wilson et al. (2022), who highlight the importance of AI in improving predictive opportunities and facilitating regenerative production. The study contributes to existing knowledge by empirically showing that AI directly influences circular product outcomes; it is possible to conceptualize AI as a tool as well as a strategic decision-making capability with the potential to positively influence sustainability through innovation.
Similarly, the impacts of ELB on CPDP also provide a strong confirmation of the KBV, which considers organizational learning as the core aspect of building innovation capability. This finding is consistent with Pascucci et al. (2024), who argue that companies capable of internalizing and sharing sustainability knowledge enjoy higher success in adopting circular strategies. In the case of Malaysia, where companies face regulatory pressure as well as redesign and transformation requirements within the Industry 4.0 policy frameworks, entrepreneurial learning can help to ensure proactive adaptation and experimentation in the design of circular products, thus supporting the theoretical complementarity of the KBV and DCT.
The intermediate nature of KTE is a decisive addition to the theory, as it demonstrates how strategic capabilities should be operationalized in terms of knowledge processes to be transformed into performance results. In line with the KBV, the mediation outcomes confirm that knowledge is a dynamic resource, the efficacy of which defines the achievement of the benefits of innovation. These findings are consistent with prior studies emphasizing the strategic role of AI capability, digitalization, and knowledge integration in improving sustainability-oriented innovation performance (Liu et al., 2026; Sun et al., 2026). They also address the gaps in the current literature on CE by elucidating that it is not enough for firms to possess circular and AI capabilities; rather, they need to nurture effective knowledge ecosystems in order to create value. AI capability and knowledge integration mechanisms are increasingly recognized as strategic drivers of sustainable innovation and organizational performance (Liu et al., 2026). These results also empirically confirm recent conceptual postulations by Gama and Magistretti (2023) and Sun et al. (2025), which describe knowledge transfer as the missing link between digital and strategic capabilities and sustainable performance.
Moreover, the results show that DIQ and ISI mainly strengthen the upstream conversion of firm capabilities into knowledge transfer effectiveness. However, their moderating effects on the KTE → CPDP relationship were not significant, suggesting that once knowledge has been transferred, internal execution may matter more than external digital or institutional conditions. This means DCT and Institutional Theory are applicable in such contexts. In line with Reuter (2016) and Wang and Zhang (2025), the positive impact of DIQ also suggests that digital maturity increases the absorptive capacity of firms and accelerates AI-driven circular transformation. At the same time, the moderating effect of ISI implies that regulatory systems, state subsidies, and alignment of policy enhance the transformation of knowledge and innovation into quantifiable results. This is especially true in Malaysia, where the introduction of new policies like the National Circular Economy Roadmap provides an excellent opportunity for technological integration and sustainability change.
Lastly, the moderated mediation results present a new theoretical contribution because they show that the indirect influences of CIO, AIDS, and ELB on CPDP via KTE are dependent on the digital and institutional ecosystem. Consequently, rather than merely being a factor of internal competency, circular transition is a co-evolutionary process between companies and their surrounding environment. These findings contribute to the academic discussion by combining the RBV, DCT, and the KBV into a single dynamic model that is particularly relevant to emerging economies transitioning to digital-industrial formations.
A significant aspect of the suggested framework is the non-significant moderating impacts of DIQ and ISI on the association between KTE and CPDP. These results imply that even though digital and institutional conditions are relevant in reinforcing capability development and knowledge transfer during the earlier stages, they may have a less decisive impact when the knowledge has already been successfully mobilized at the firm level. In these circumstances, it may be the case that internal execution issues are more heavily relied upon to convert transferred knowledge into real circular product results, e.g., managerial commitment, interdepartmental coordination, product design routines, implementation discipline, and the operational capacity to transfer knowledge received into development processes. This trend suggests that DIQ and ISI have a greater impact on enabling conditions of upstream capability activation and knowledge exchange rather than acting as universal amplifiers of all downstream performance relationships. In other words, although digital preparedness and institutional assistance enable companies to create and disseminate knowledge in a more efficient manner, it is not inevitable that this knowledge will result in improved circular product development results. This interpretation provides a subtler theoretical understanding of the contextual effects by distinguishing between the conditions that facilitate knowledge mobilization and those that define the success of implementation.
Although the proposed framework is applicable beyond Malaysia, its transferability cannot be assumed and should be interpreted with care. The model can be applied in theory to other emerging economies since the majority of them have encountered similar issues in terms of industrial transformation, uneven digitalization, institutional uncertainty, and sustainability issues. Nevertheless, these relationships might be stronger and more significant in certain contexts based on local policy regimes, digital infrastructure maturity, industry structure, and organizational preparedness. As a result, the model is reproducible due to its conceptual structure but it cannot be expected to produce the same empirical outcomes in different settings. Future research should thus seek to apply the model to other emerging economies to identify the relationships that are more stable and those that are more context-specific. Such cross-context testing would assist in clarifying the boundary conditions in which this capability-based account of circular innovation can be generalized.
In general, this research helps to fill the existing research gaps and adds to the significant theoretical development by presenting the interaction between AI, circular orientation, and learning capabilities in a synergistic manner, provided that there is digital and institutional preparedness in terms of knowledge transfer. Such lessons provide a repeatable structure that can be applied to other emerging economies that are interested in hastening their move toward sustainable manufacturing.
ConclusionOverall, the research contributes by providing a more process-oriented and context-specific description of circular product development in developing economies. Instead of the assumed direct impact of technological or strategic capabilities on the creation of circular outcomes, the results indicate that these outcomes are contingent on knowledge transfer processes and the digital and institutional environment. This offers a more sophisticated perspective on the concept of circular innovation as the result of interplay between internal and external conditions, which enhances the theoretical and practical applicability of the model developed within this paper. The key contribution of the current research is explaining the joint effect of AI capabilities, a focus on circular innovation, and entrepreneurial learning on the ability of Malaysian manufacturing businesses to develop circular products. Based on the RBV, DCT, and the KBV, the results show that intrinsic capabilities contribute to product circularity both directly and indirectly through the effectiveness of knowledge transfer. The analysis also highlights the enabling capacity of high-quality digital infrastructure and support for innovation at the institutional level, which strengthens the capability-performance nexus and helps to shift toward CE practices. These findings confirm that the process of converging to a CE among emerging economies is not a technological replacement but a strategic, knowledge-intensive process that requires coordination of organizational resources with the external institutional ecosystems. In this regard, the research enriches the existing theory by developing a refined moderated mediation model and provides new empirical data on the Malaysian environment, thus confirming its relevance as an experimental testing platform for the industrial sustainability transition.
Theoretical implicationsIn terms of its theoretical contribution, this study not only applies established theories to a new context but also specifies how they interact in a conditional process model of circular innovation. The current literature has generally applied the RBV, DCT, or the KBV individually to elucidate outcomes that can be related to sustainability. The current research contributes to this literature by demonstrating that circular innovation performance can be more effectively viewed as a stratified process where internal capabilities constitute the strategic foundation, knowledge transfer is the conversion process, and digital and institutional circumstances determine the extent to which this process is successful. This moves the debate away from mere capability-performance arguments and provides a more process-based account of how circular product development comes to be in different contextual conditions. The key theoretical progress within this study is the clarification of the sequence, interaction, and boundary conditions in which internal resources are transformed into circular innovation results.
Practical implicationsThe practical implications of the study are not uniform and are circumstance-dependent. The results indicate that AI investments will yield greater circular product development results when firms already have a clear orientation toward circular innovation and a learning culture that can facilitate the interpretation and utilization of AI-generated insights. For companies with a lower level of digital maturity, the priority should not be large-scale AI deployment but rather enhancing the integration of data, digital coordination, and internal knowledge-sharing habits to facilitate successful technological investments. Similarly, when institutional support is scarce, managers may be more reliant on internal coordination, cross-functional learning, and strategic alliances to continue circular initiatives. For policymakers, digital infrastructure and innovation support mechanisms should be structured as facilitating conditions with regard to knowledge-based capability development as opposed to individual policy interventions. More precisely, regulatory incentives, specific support for digitalization, and collaboration between industry and the government are likely to be more effective when they help firms to convert inner capabilities into practical circular product plans. These implications are indicative of the interactive logic of the model since they demonstrate that the value of AI and circular capabilities is dependent on the level of digital preparedness and institutional support in a given firm’s immediate environment.
Limitations and future research prospectsThe current research faces a number of limitations that outline directions for future research. To begin with, the cross-sectional research design limits the possibility of making causal inferences; further research might adopt longitudinal research designs to examine how AI and circular capabilities evolve dynamically over time. Second, the focus on Malaysian firms means the results may not be generalizable. Future research could involve multi-country comparative studies in Southeast Asia to explore regional variations in the regulatory landscape and digital maturity. Third, the use of perceptual measures creates the potential for common-method bias even with statistical controls; accordingly, objective performance metrics should be incorporated into further research. Further studies may be necessary to investigate other moderating variables, including green human resource management practices, cultural factors, or sustainability leadership, to further develop the theoretical model. Lastly, the accelerated development of generative AI and blockchain technologies creates new research possibilities to examine how new digital ecosystems can speed up or transform circular product development.
CRediT authorship contribution statementWai Nga Leong: Writing – original draft, Conceptualization. Kwong Yee Fong: Resources, Methodology. Chia-Yang Lin: Methodology, Conceptualization. Massoud Moslehpour: Writing – review & editing, Writing – original draft, Conceptualization.
