A SEM–ANN analysis to examine impact of artificial intelligence technologies on sustainable performance of SMEs

Abstract

This study investigates the impact of Artificial Intelligence (AI) adoption on the sustainable performance of small and medium-sized enterprises (SMEs). Employing a hybrid quantitative approach, this research combines Partial Least Squares Structural Equation Modeling (PLS-SEM) and Artificial Neural Networks (ANN) to examine the influence of various organizational, technological, and external factors on AI adoption. Key factors considered include top management support, employee capability, customer pressure, complexity, vendor support, and relative advantage. Data collected from 305 SMEs across multiple sectors were analyzed. The results reveal that all the proposed factors significantly and positively affect AI adoption, with top management support, employee capability, and relative advantage being the most influential predictors. Additionally, the adoption of AI technologies substantially enhances the economic, social, and environmental performance of SMEs, reflecting improvements in operational efficiency, cost reduction, and social value creation. The ANN results confirm the robustness of the SEM findings, highlighting the critical role of AI in driving sustainability outcomes. Furthermore, the study emphasizes the positive mediation effects of AI adoption on organizational performance, indicating that AI adoption serves as a key enabler in achieving both short-term operational gains and long-term sustainability objectives. This research contributes to the understanding of AI’s transformative role in enhancing the sustainable performance of SMEs in developing economies, offering strategic insights for both policymakers and business leaders.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-86464-3.

Introduction

In the era of the Fourth Industrial Revolution, the rapid evolution of digital technologies dramatically reshaped the landscape of global business operations. Advances in various fields such as, machine learning, and data accumulation, and computer science have facilitated the application of advanced technologies in businesses¹. Among these technologies, Artificial Intelligence (AI) stands out as a transformative force, offering unparalleled opportunities for innovation, efficiency, and competitive advantage. Bunte et al.¹¹ report that industries such as healthcare, banking and finance, transportation, education, and customer service are experiencing significant impacts due to AI integration^2,3. However, in healthcare, AI is enhancing diagnostics and personalized treatments⁴; in banking and finance, it optimizes fraud detection and customer service automation⁵; in transportation, it facilitates autonomous systems and route optimization^6,7 ; and in education^8,9, AI improves personalized learning experiences and administrative efficiency. The integration of AI across these sectors is driving increased efficiency, cost reductions, and improved decision-making processes¹⁰. The term “artificial intelligence” lacks a universally accepted definition due to its many forms and applications¹¹. Nevertheless, its transformative role across these industries underscores its growing importance in shaping future business and societal models¹¹.

AI is defined as the study and creation of computational models and systems that exhibit cognitive abilities similar to those of humans¹². According to Bag et al.¹³, AI technology can enhance productivity, reduce costs, improve product quality, and enhance customer service. The impact of AI technology adoption on company performance, as measured by cost reduction and improved forecasting, has been the subject of multiple studies. These studies demonstrate improvements in business operations¹⁴, increased output through the automation of repetitive tasks (Acemoglu and Restrepo¹⁴¹), enhanced product innovation¹⁵, and firm growth^13,16,17. AI holds significant potential for improving company performance¹⁸, but organizations must overcome substantial hurdles to fully invest in AI¹⁹.

AI can address complex problems by combining computational resources and performing tasks similar to those performed by humans^20,21. Given these capabilities, AI has the potential to exacerbate interrelated sustainability issues^22–24. Sustainable business practices require companies to consider social, economic, and environmental impacts alongside current demands. Effective sustainability efforts are crucial for reducing costs, increasing profits, and opening new socially beneficial opportunities^25,26. Given that SMEs’ financial and non-financial performance is influenced by the adoption of AI techniques, the impact of AI is a critical area of research^27,28.

Small and medium-sized enterprises (SMEs) are often considered a country’s greatest asset, particularly in less developed economies. SMEs represent approximately 90% of all enterprises and 50% of all employment globally. In emerging nations, SMEs contribute an estimated 40% of the official gross domestic product (GDP). In XXX, there are about 3.3 million SMEs, making up more than 30% of the country’s GDP and 25% of its exports²⁹. SMEs employ more than 80% of the workforce and have a substantial presence in the retail, agriculture, manufacturing, wholesale, trade, and service sectors. The Tribune (2021) reports that there are one million SMEs in Sindh, accounting for 30% of the country’s GDP, with approximately 30% of these SMEs operational in Upper Sindh.

Despite their potential, SMEs often face challenges in fully capitalizing on AI technologies due to limited awareness, scarce financial and human resources, a shortage of skilled workforce, and concerns over data privacy and security^30,31. While large corporations have made substantial advancements in AI integration, SMEs still face significant barriers. Consequently, this study aims to examine the current status of AI technology implementation among Pakistani SMEs.

Research contributions

This research makes several key contributions. First, it builds a research model that explores the readiness of SMEs to adopt AI and its impact on business performance in developing nations like XXX. Previous research on AI adoption has primarily used “structural equation modeling (SEM)” to examine linear relationships between components. However, this approach has limitations. To address this, this study employs a two-staged approach, combining SEM with “artificial neural networks (ANN)” to provide a more comprehensive analysis.

Additionally, while earlier research has viewed performance unidimensional, this study considers both operational and economic performance. This method improves knowledge of AI adoption and business performance.

Research objectives

This research study on AI adoption by SMEs for sustainable business using a two-staged hybrid SEM-ANN approach, this study addresses a significant research gap. The objectives of this study are to investigate the following research inquiries:

1. How does the adoption of AI impact the sustainable business performance of SMEs?

2. How can the interactions between SEM and ANN enhance the understanding of AI adoption and business performance?

3. What are the challenges and opportunities presented by AI adoption for SMEs’ sustainability goals?

Literature review

Technology–Organization–Environment (TOE) framework

The “Technology-Organization-Environment (TOE)” framework, introduced by Tornatzky et al.³², and the Diffusion of Innovation (DOI) theory by Rogers³³ are two prominent theories used to examine the adoption and diffusion of innovations within organizations³⁴. Unlike other theories such as the “Technology Acceptance Model (TAM)”³⁵, the “Theory of Planned Behavior (TPB)”³⁶, and the “Unified Theory of Acceptance and Use of Technology (UTAUT)”³⁷, which primarily focus on individual decision-making, the TOE framework provides a broader organizational perspective. Due to its robust application in examining organizational adoption of various forms of innovation, the TOE framework is the primary theoretical lens for this study^34,38.

Its theoretical base and empirical validation make the TOE framework a useful paradigm for analyzing technology uptake³⁹. The framework posits that three sets of factors—“technology, organization, and environment”—collectively influence an organization’s ability to adopt new technologies³⁸. Each of these contexts is critical in understanding the comprehensive nature of technology adoption.

First, Technology Context: This refers to both the internal and external technologies relevant to the organization. It encompasses the existing technologies within the firm as well as the new technologies available in the market. This context includes the perceived benefits, complexity, and compatibility of technology. Innovations that are perceived to be advantageous, less complex, and more compatible with existing systems are more likely to be adopted^38,40. Second, Organization Context: This refers to the characteristics of the organization itself, including its size, scope, managerial structure, and available resources. Factors such as top management support, organizational readiness, and resource availability are critical in determining the adoption of new technologies⁴¹. For instance, organizations with supportive leadership and adequate resources are more likely to embrace innovative technologies^41,42. And last, Environment Context: This refers to the external environment in which the organization operates, including industry characteristics, market structure, and regulatory environment. Competitive pressure, customer expectations, and government regulations can significantly influence the adoption of new technologies. Organizations operating in highly competitive industries are more likely to adopt new technologies to maintain their competitive edge⁴³.

The TOE framework has been widely applied in various studies to examine the adoption of different technologies. For instance, Zhu et al.⁴⁴ applied the TOE framework to study e-business adoption and found that technological readiness, firm size, and competitive pressure were significant predictors of adoption. Similarly, Baker³⁸ used the TOE framework to investigate cloud computing adoption and highlighted the importance of perceived benefits and organizational readiness.

The TOE framework is useful for studying SMEs’ AI adoption factors^44,45. AI technologies offer numerous benefits, including enhanced decision-making, improved operational efficiency, and the potential for innovation^46,47. However, SMEs face unique challenges in adopting these technologies, including limited financial resources, lack of skilled personnel, and concerns over data security^48,49.

Several TOE studies have examined AI adoption⁵⁰. observed that technology maturity, senior management backing, and competitive pressure affected manufacturing business AI adoption. Stenberg⁵¹ evaluated healthcare AI adoption and stressed perceived advantages, organizational preparedness, and regulatory context. SMEs are vital to the economy, especially in emerging nations, therefore understanding AI adoption drivers is crucial. Limited knowledge, budgetary restrictions, and a trained personnel deficit plague SMEs²⁹. The TOE framework may help identify AI adoption elements and their pros and cons.

Diffusion of innovation (DOI) theory

The Diffusion of Innovation (DOI) theory, proposed by Rogers³³, complements the TOE framework by focusing on how innovations are communicated and adopted over time among members of a social system. The DOI theory identifies five key attributes of innovations that influence their adoption: relative advantage, compatibility, complexity, trialability, and observability. Innovations perceived to have greater relative advantages, compatibility, trialability, and observability, and lower complexity are more likely to be adopted.

In the context of AI adoption by SMEs, the DOI theory provides an understanding of the characteristics of AI technologies that influence their adoption. AI technologies that offer significant relative advantages, such as improved decision-making and operational efficiency, are more likely to be adopted by SMEs. Similarly, AI technologies that are compatible with existing systems and less complex to implement are more likely to be adopted³³.

The integration of the TOE framework and DOI theory provides a deep approach to understanding the factors influencing AI adoption by SMEs. While the TOE framework provides a broad organizational perspective, the DOI theory offers the characteristics of AI technologies that influence their adoption. This integrated approach allows for a more advanced understanding of the factors influencing adoption of AI and the potential benefits and challenges associated with it.

By applying this integrated approach, this study aims to examine the factors influencing AI adoption by SMEs in XXX and its impact on sustainable business performance.

Hypotheses development

Technical factors

Implementation cost and AI adoption by SMEs

Implementation cost is a significant factor influencing the adoption of new technologies, including AI, especially for SMEs. SMEs typically operate with limited financial resources, making cost considerations paramount when deciding whether to adopt new technologies⁵². The expenses associated with AI implementation include both financial and human resources required for the process, encompassing hardware, software, training, and ongoing maintenance costs^51,53.

Studies have consistently highlighted the significant impact of implementation costs on technology adoption. Djatikusumo⁵⁴ pointed out that SMEs often struggle with obtaining the necessary funding due to poorly managed organizations, resulting in inefficient IT systems. The financial constraints of SMEs limit their ability to invest in advanced technologies, which often come with high initial costs⁵⁵, who noted that the lack of financial resources or capital is a significant barrier to the development and utilization of digital channels by SMEs.

The higher cost of adopting AI technologies presents a substantial challenge for SMEs. AI implementation is not just about purchasing and integrating new software; it also involves substantial expenditures on infrastructure, training employees to effectively use the technology, and continuous system upgrades and maintenance. Chatterjee and Kar⁵⁶ emphasized that cost-effectiveness is a crucial determinant of technology adoption, suggesting that high implementation costs can deter organizations from embracing new technologies.

In the context of XXX, SMEs have reported financial constraints as a primary reason for their reluctance to adopt AI technologies⁵⁷. The cost factor is particularly pronounced in developing economies, where access to capital is limited, and financial markets are not as developed⁵⁸. This financial limitation makes SMEs more vulnerable to cost-related challenges, making them hesitant to invest in high-cost technologies like AI⁵⁹.

Furthermore, the sensitivity to errors and unforeseen events associated with AI technologies can increase the perceived risk, making SMEs even more cautious. The fear of potential financial losses due to implementation failures or inefficiencies can further dissuade SMEs from adopting AI. As highlighted by⁶⁰, the costs associated with the implementation, including the financial and human resources needed, are significant barriers to technology adoption⁶⁰.

Several studies support the notion that high implementation costs negatively impact technology adoption. For instance, Thong⁶¹ found that financial constraints were a significant barrier to IT adoption among small businesses. Similarly⁶², concluded that cost-related factors were among the most critical barriers to the adoption of various technological innovations by SMEs. The following hypothesis is proposed:

H1(a)

Implementation cost has a negative effect on AI adoption by SMEs.

Relative advantage and AI adoption

Relative advantage, as defined by Rogers⁶³, refers to the degree to which an innovation is perceived as being superior to its predecessors. This superiority can manifest in various forms, including enhanced performance, increased financial returns, improved efficiency, and greater convenience. The concept of relative advantage is pivotal in understanding the adoption of innovations, as it directly influences the perceived value and benefits of the new technology⁶³.

Several studies have highlighted the positive relationship between relative advantage and technology adoption. The technologies offering greater performance and higher financial returns are more likely to be chosen by organizations⁶⁴. This assertion is supported by⁶⁵, who found that the adoption rate of innovation is positively correlated with its relative advantage. In other words, innovations that offer significant benefits over existing technologies are adopted more quickly.

The adoption of AI technologies by SMEs can be significantly influenced by the relative advantages these technologies provide. AI technologies offer numerous benefits, including cost reduction, improved decision-making, and enhanced forecasting capabilities⁶⁶. These advantages are particularly critical for SMEs operating in competitive business environments, where efficiency and effectiveness can determine survival and growth^67,68.

Ramdani and Kawalek⁶⁹ and Grandon and Pearson⁷⁰ also found a favorable correlation between relative advantage and the adoption of innovation. Their research suggests that the perceived benefits of a new technology play a crucial role in its acceptance and implementation. For SMEs, AI technologies can streamline operations, reduce operational costs, and provide superior insights through data analysis, making them highly attractive⁷⁰.

In the fiercely competitive business landscape, the ability to use AI for improved operational efficiency and strategic decision-making is invaluable. AI technologies can automate routine tasks, enhance customer interactions, and provide predictive analytics, which collectively contribute to better performance outcomes. The adoption of AI is significantly influenced by its relative advantage⁷¹. Organizations that perceive AI as offering substantial benefits over traditional methods are more likely to adopt and integrate these technologies into their operations⁸.

The positive effects of relative advantage on technology adoption have been consistently demonstrated in various contexts. For instance, Thong⁶¹ found that perceived relative advantage was a significant determinant of IT adoption among small businesses. Similarly, Cidral et al.⁷² identified relative advantage as a critical factor influencing the adoption of technological innovations by SMEs.

In the context of SMEs, the relative advantages of AI technologies can be particularly pronounced. SMEs in developing countries often face resource constraints and competitive pressures, making the efficiency and cost-saving benefits of AI highly attractive⁷³. The ability to make data-driven decisions and improve forecasting accuracy can provide a significant competitive edge.

H1(b)

Relative advantage has a positive effect on AI adoption.

Complexity and AI adoption

Complexity is a significant barrier to the adoption of new technologies, including AI. Complexity refers to the degree to which an innovation is perceived as difficult to understand and use³³. High complexity can deter potential adopters, as they may find it challenging to comprehend, implement, and integrate the technology into their existing systems.

The relationship between complexity and technology adoption has been extensively studied. Davis et al.³⁵, characterized compatibility, a related concept, as “the degree to which an innovation is viewed as consistent with potential adopters’ existing needs, values, and experience.” Compatibility can be functional or normative. However, Premkumar and Roberts⁷⁴ emphasized that organizations should aim for technological compatibility, which means the innovation can work with current technology, and organizational compatibility, which means it can adhere to established work practices and value systems.

Complexity, on the other hand, is a measure of how challenging an innovation is to comprehend and implement^75,76. highlighted that the difficulty in using new technology is a critical factor influencing its adoption. If potential adopters perceive a technology as hard to use, they are less likely to embrace it. It is suggested that the perceived difficulty of adopting new technology is a good predictor of its adoption success⁷⁷.

The need for simplicity and ease of use for quicker adoption is emphasized in various studies⁷⁸. and⁷⁹. argued that technologies need to be simple and easy to use to be adopted more rapidly. When technologies are perceived as complex, it becomes difficult for organizations to explore and understand them, hindering their acceptance^80,81.

Moreover, overly complicated technical systems negatively impact competency⁸². The diverse applications of AI add to its complexity, making it difficult to study and implement⁷¹. This complexity can be fascinating due to its potential, but it also poses significant challenges for managers who may resist implementing and understanding AI if they perceive it as too complex.

In the context of AI, complexity encompasses various aspects, including the technical intricacies of AI algorithms, the integration with existing systems, and the need for specialized knowledge and skills to operate AI technologies effectively^83,84. These factors contribute to the perceived difficulty and can significantly hinder the adoption of AI by SMEs. Research indicates that high complexity is associated with lower rates of technology adoption. For instance^85,86, found that the perceived complexity of IT systems negatively affected their adoption by small businesses. Similarly^87,88, identified complexity as a critical barrier to the adoption of technological innovations by SMEs.

H1(c)

The adoption of AI is negatively impacted by complexity.

Organizational factors

Top management support and AI adoption

Top management support plays a critical role in the successful adoption and implementation of AI technologies within organizations. The commitment and active involvement of senior leadership are essential for creating a conducive environment, securing necessary resources, and fostering a culture that embraces technological innovations.

Effendi et al.⁸⁹ highlighted that the adoption of new technologies often requires the backing of upper management to ensure the allocation of resources and to establish an environment that supports innovation. This is particularly important in the context of AI technologies, which involve complex deployment processes and various challenges. Fountaine et al.⁹⁰ emphasized the need for business owners and senior managers to actively engage in understanding and researching AI technologies rather than relying solely on technical staff. This involvement helps bridge the gap between technical and managerial perspectives, ensuring that AI initiatives align with the organization’s strategic goals^84,85.

The role of top management extends beyond resource allocation and strategic alignment. It also involves fostering an organizational culture that is receptive to AI adoption. Enholm et al.⁹¹ noted that business culture significantly affects the adoption of AI, and top managers are uniquely positioned to influence and shape this culture. By promoting a culture that values innovation, continuous learning, and openness to change, top management can mitigate resistance and encourage broader acceptance of AI technologies within the organization.

Previous studies have consistently demonstrated a robust, favorable, and statistically significant relationship between the role of managers and the adoption of technology within organizational settings, particularly in SMEs^92–95. found that top management support is a key determinant of successful technology adoption, as it influences organizational readiness and commitment to new initiatives. Similarly, it is highlighted by authors that managerial support positively impacts the adoption of various technologies, including AI, by facilitating the necessary changes in organizational processes and practices^96–98.

The importance of top management support is also underscored by its influence on overcoming barriers to AI adoption. Senior leaders can champion AI initiatives, communicate their strategic importance, and provide the necessary resources and support to address challenges such as cost, complexity, and skill gaps⁹⁸. By demonstrating a clear commitment to AI adoption, top management can inspire confidence and motivate employees to engage with AI technologies actively^99,100.

In the context of SMEs, where resource constraints and organizational inertia can pose significant challenges to technology adoption, the support of top management is even more crucial¹⁰¹. SMEs often operate with limited financial and human resources, making it essential for senior leaders to prioritize AI initiatives and allocate resources strategically^100,101. The following hypothesis is proposed:

H2(a)

Top management support has a positive effect on AI adoption.

Availability of financial resource and AI adoption

The availability of financial resources plays a crucial role in the adoption of innovative technologies by businesses. Resource availability, defined by Maduku et al.¹⁰² as the state of having sufficient resources to implement new technologies, significantly influences an organization’s ability to adopt and integrate novel systems. Implementing and managing new technologies, such as AI, require substantial financial investment for both initial deployment and ongoing operational expenses²⁰.

Small and medium-sized enterprises (SMEs) often face challenges related to financial resource constraints¹⁰³. Most start-ups, in particular, struggle with limited financial resources, which can lead owners to excessively use their personal assets to support business operations¹⁰⁴. This financial strain can hinder their ability to invest in new technologies like AI. In contrast, larger organizations typically have a substantial advantage in securing financial resources compared to SMEs^105,106.

Ramadan and Eleyan¹⁰⁷ found that the capacity to finance technological adoption significantly impacts SMEs’ decision to adopt new technologies. This indicates that access to financial resources is a critical factor influencing the adoption of AI technologies by SMEs. The lack of financial resources poses a significant barrier to AI adoption, as SMEs may struggle to allocate funds necessary for AI implementation, training, and maintenance²⁸.

Research supports the notion that financial resources are essential for technology adoption. For instance, organizations with adequate financial resources can invest in necessary infrastructure, training, and support systems required for effective AI adoption^108,109. This investment is crucial for overcoming initial implementation challenges and ensuring the smooth integration of AI technologies into existing business processes¹⁰⁹.

Furthermore, the availability of financial resources affects an organization’s ability to innovate and compete in the market. Businesses that can allocate sufficient funds for technological advancements are better positioned to adopt the benefits of AI, such as improved efficiency, enhanced decision-making, and competitive advantage. Conversely, financial constraints can limit an organization’s ability to experiment with and adopt new technologies, hindering its potential for growth and innovation. The following hypothesis is proposed:

H2(b)

Availability of Financial Resource has a positive effect on AI adoption.

Employee capability and AI adoption

Employee capability, defined as the skills, knowledge, and expertise possessed by employees within an organization, plays a crucial role in the adoption of AI technologies. According to Ghobakhloo et al.⁶⁰, employees are vital assets for the survival and growth of companies, especially when it comes to implementing technological innovations. Organizations with highly skilled and knowledgeable staff are better equipped to adopt and integrate new technologies effectively.

Maduku et al.¹⁰² emphasize that having a workforce with the necessary qualifications is essential for successfully implementing technological innovations. Skilled personnel can provide the technical expertise and innovation capabilities required to navigate the complexities of AI implementation. This is particularly true in industries that rely heavily on labor-intensive operations, where advancements in tacit skills and employee participation are crucial for technological innovation^3,65.

In contrast, SMEs often face challenges related to human resource deficiencies compared to larger enterprises. Harness et al.¹⁰⁶ highlight that SMEs may lack the human capital necessary to drive innovation and adopt new technologies effectively. This can hinder their ability to compete with larger firms that have more extensive resources and skilled personnel at their disposal.

Research by Tajpour et al.¹¹⁰ supports the notion that employee skills positively influence the adoption and utilization of new technologies by businesses. Organizations that invest in training and development programs to enhance employee capabilities are better positioned to leverage the benefits of AI technologies. Skilled employees can contribute to smoother implementation processes, quicker adaptation to technological changes, and enhanced performance outcomes. Furthermore, the capability of employees to understand and utilize AI technologies effectively is crucial for maximizing their potential benefits¹¹¹. Employees who are knowledgeable about AI concepts, algorithms, and applications can contribute to the development of innovative solutions, improve decision-making processes, and enhance operational efficiency within the organization¹¹². Therefore, based on the existing literature and theoretical base, the following hypothesis is proposed:

H2(c)

Employee Capability has a positive effect on AI adoption.

Environmental factors

Competitor pressure and AI adoption

Competitor pressure refers to the influence exerted on an organization by its rivals within the same industry¹¹³. This external pressure serves as a significant motivating factor for organizations to adopt new technologies, including artificial intelligence (AI). When competitors adopt novel technologies, organizations perceive a competitive advantage that encourages them to follow suit¹¹⁴.

The adoption of AI technologies is driven by the belief that competitors gaining a competitive edge through technological advancements can alter industry dynamics and redefine competitive norms¹¹⁵. Companies under competitive pressure are compelled to innovate and adopt AI to enhance their operational efficiency, improve decision-making processes, and maintain competitiveness in the market¹¹⁶.

Research suggests that organizations that fail to adopt new technologies risk falling behind their competitors and experiencing reduced organizational performance¹¹⁷. In contrast, embracing AI technologies enables companies to avoid competitive disadvantages and position themselves strategically in the market¹¹⁸. Therefore, competitive pressure acts as a catalyst for organizations to explore and integrate AI technologies to gain a foothold in their respective industries.

Several studies have demonstrated a positive relationship between competitive pressure and technology adoption across various sectors^115,116. Organizations are more likely to adopt AI when they perceive their competitors utilizing similar technologies effectively. This competitive dynamic creates an environment where technological adoption becomes a strategic imperative for maintaining market relevance and competitiveness.

In the context of SMEs, competitive pressure plays a pivotal role in influencing decisions related to technology adoption. SMEs often face intense competition from larger firms and must innovate continuously to differentiate themselves and capture market share^93,119. The following hypothesis is proposed:

H3(a)

Competitor Pressure has a positive effect on AI adoption.

Customer pressure and AI adoption

Customer pressure refers to the influence exerted by consumer demand and expectations on businesses to adopt new technologies, such as artificial intelligence (AI). In today’s competitive marketplace, consumer preferences play a pivotal role in driving organizations to innovate and enhance their product offerings¹²⁰.

According to Abed et al.¹²¹ factors like consumer dedication, encouragement, and trust are crucial in motivating businesses to embrace new technologies. Organizations recognize that meeting customer expectations and demands is essential for maintaining customer satisfaction and loyalty¹²². Therefore, businesses are increasingly motivated to adopt advanced technologies, including AI, to improve their products and services and enhance customer experiences^123,124.

Research indicates that businesses perceive customer expectations as a compelling reason to adopt new technologies that can potentially meet or exceed these expectations¹²². As customers become more informed and technologically savvy, they expect businesses to using cutting-edge technologies to deliver superior products and services¹²³.

For SMEs, responding to customer pressure by adopting AI technologies can offer several strategic advantages. By integrating AI into their operations, SMEs can enhance customer relationship management, personalize customer interactions, and offer innovative solutions that meet evolving consumer demands. AI-driven awareness and analytics enable SMEs to make data-driven decisions that enhance operational efficiency and customer satisfaction. Based on the theoretical foundations and empirical evidence, the following hypothesis is proposed:

H3(b)

Customer Pressure has a positive effect on AI adoption.

Vendor support and AI adoption

Vendor support plays a critical role in the adoption of new technologies, including artificial intelligence (AI), by organizations, particularly small and medium-sized enterprises (SMEs). This support encompasses various forms of assistance provided by technology vendors or consultants, such as training, guidance during deployment, maintenance support, and regular updates¹²⁵.

Research indicates that the level of support from vendors significantly influences organizations’ decisions to adopt AI technologies. Organizations perceive vendor support as a crucial factor that mitigates implementation risks and enhances their readiness to innovate¹²⁶. When vendors provide comprehensive support throughout the AI adoption process, including pre-deployment preparation and post-implementation maintenance, organizations are more likely to successfully integrate AI into their operations¹²⁷.

In the context of SMEs, which often face resource constraints and technical challenges, vendor support becomes even more vital. Studies underscore the positive impact of vendor support on the adoption of new technologies by SMEs, facilitating smoother implementation and ensuring optimal utilization of AI capabilities^128–130. Effective vendor support can alleviate SMEs’ concerns about the complexity and costs associated with AI adoption. By providing necessary training, technical expertise, and ongoing assistance, vendors empower SMEs to adopt AI technologies effectively to enhance productivity, improve decision-making processes, and meet competitive demands. The following hypothesis is formulated:

H3(c)

Vendor Support has a positive effect on AI adoption.

Artificial Intelligence adoption and sustainable business performance

Long-term sustainability is a crucial aspect in achieving business success. This is accomplished by integrating financial, environmental, and social considerations into executive practices^131,132. According to. Agarwal et al.¹³³and Boudreau¹⁴, businesses can save money by implementing AI systems, which also enhance their forecasts and operations. Researchers have proposed the implementation of the Triple Bottom Line (TBL) framework, which necessitates substantial modifications to the company’s focus on the aforementioned triple bottom line indicators¹³². By considering the financial, social and environmental aspects, this method enables the implementation of sustainable business practices and the attainment of sustainable performance¹³². Financial success, human welfare, and environmental quality are all critical to the public’s well-being, according to experts^134,135. Nevertheless, experts contend that organizations prioritize the economic aspect over the social and environmental aspects¹³¹. All components play a vital part in the successful performance of a firm^136,137. Research on the organization’s performance often focuses on three essential components: social, environmental, and financial performance. However, there is a lack of studies that comprehensively investigate these three areas, as noted by^131,132. Hence, all three aspects of Sustainable Business Performance have been incorporated into this research study.

AI adoption and economic performance

The adoption of artificial intelligence (AI) technologies holds significant potential for enhancing the economic performance of small and medium-sized enterprises (SMEs). Economic performance is typically assessed by how well and how quickly a company generates revenue and strengthens its market position. Economically viable performance is characterized by a company’s ability to expand its market share, leading to increased revenue and a more favorable return on investment (ROI)¹³⁸.

AI technology enables businesses to gain deeper insights into customer behavior and preferences, leading to more accurate sales forecasts. By analyzing large datasets, AI can identify patterns and trends that human analysts might overlook, allowing companies to tailor their marketing strategies more effectively. This leads to improved customer satisfaction and loyalty, which can drive higher sales volumes and revenue growth. For instance, AI-driven recommendation systems used by e-commerce platforms personalize shopping experiences, resulting in increased sales and customer retention¹³⁹. In various studies it is found that AI adoption can lead to a 38% increase in revenue for SMEs^59,140. Another study revealed that AI adoption can lead to a 20–30% increase in revenue for SMEs²⁸.

One of the most significant advantages of AI adoption is its ability to reduce operational costs and enhance efficiency. A report found that AI adoption can lead to a 10–15% reduction in operational costs for SMEs¹⁴⁰. AI technologies can automate repetitive tasks, streamline workflows, and optimize resource allocation. This reduction in manual labor not only decreases labor costs but also minimizes the likelihood of human error, leading to more efficient and error-free operations. Studies by Acemoglu and Restrepo¹⁴¹ and Bughin¹⁰³ have shown that AI can automate routine processes, leading to significant cost savings. A study found that AI adoption can lead to a 20.5% increase in efficiency for SMEs^142,143. Additionally, Awan et al.¹⁴⁴ and Baabdullah et al.²⁸ highlight that AI’s capability to optimize supply chain management and inventory control further contributes to cost reduction.

AI adoption can significantly enhance a company’s market competitiveness. By adding AI for competitive intelligence, businesses can monitor market trends, analyze competitors’ strategies, and respond more swiftly to market changes. This agility allows SMEs to maintain and even expand their market share, contributing to sustained revenue growth. AI’s predictive analytics capabilities enable businesses to make data-driven decisions that improve ROI. For example, AI can help in pricing strategies by analyzing market conditions and consumer behavior to set optimal prices¹⁴⁵.

AI’s analytical power supports better decision-making and strategic planning. By providing real-time insights and predictive analytics, AI helps business leaders make informed decisions that align with market demands and business goals. Moreover, few studies found that AI adoption can lead to a 25% improvement in decision-making for SMEs^109,146. This strategic advantage is crucial for SMEs aiming to navigate competitive markets and achieve long-term economic success¹⁴⁷. Consequently, the hypothesis posited in the present investigation is as follows:

H4(a)

AI adoption positively affects economic performance of SMEs.

AI adoption and social performance

The adoption of AI technologies can significantly enhance the social performance of SMEs. Social performance refers to an organization’s ability to improve social welfare and ensure the security of its employees and the wider community^138,148.

AI technologies are increasingly being leveraged to address pressing social issues¹⁴⁹. For example, AI can predict and mitigate the impacts of natural disasters by analyzing patterns and providing early warnings, which can save lives and reduce economic losses. In education, AI-driven personalized learning platforms can enhance access to quality education, catering to individual learning needs and improving educational outcomes^3,150,151. In healthcare, AI applications in predictive analytics and personalized medicine can improve patient outcomes by enabling early diagnosis and personalized treatment plans^152,153. AI adoption can lead to a 20% improvement in healthcare outcomes^109,154,155.

The assessment of socially viable performance involves evaluating an organization’s efforts to improve social welfare and security. AI technologies can enhance workplace safety through predictive maintenance and monitoring systems that prevent accidents and ensure a safer working environment¹⁵⁶. Therefore, it is estimated that AI adoption can lead to an 18% decrease in workplace injuries for SMEs^73,157,158. AI can also facilitate more efficient and responsive customer service, leading to higher customer satisfaction and loyalty¹⁵⁹. Additionally, AI-driven recruitment tools can help organizations identify and hire competent personnel, contributing to enhanced workforce quality and retention^160,161.

Organizations that allocate resources toward social responsibility and prioritize customer satisfaction through innovative practices often experience enhanced social performance. AI can play a crucial role in these efforts by providing tools for better community engagement and social impact measurement. AI-powered social listening tools can help companies understand community concerns and respond proactively, thereby fostering trust and improving their social license to operate¹⁶¹. Research by^162,163 highlights that increased social responsibility, facilitated by AI, leads to positive social outcomes and strengthens the company’s reputation.

AI technologies can significantly improve the recruitment and retention processes by identifying the best-fit candidates and predicting employee turnover. AI-driven recruitment platforms analyze vast amounts of data to match candidates with job requirements more accurately, ensuring a better fit and reducing hiring biases^164,165. Moreover, AI can provide personalized development plans and career progression paths for employees, enhancing job satisfaction and retention rates^166,167.

Empirical studies support the positive impact of AI adoption on social performance. Wagner suggests that organizations investing in social responsibility, customer satisfaction, and employee recruitment experience enhanced performance. A study by Zhang et al. found that AI technologies improve organizational capabilities in managing social initiatives, leading to better social outcomes. Furthermore, research by Rana et al.⁸⁸ indicates that AI-driven customer engagement and community support initiatives contribute significantly to the social performance of businesses. Consequently, the hypothesis posited in the present investigation is as follows:

H4(b)

AI adoption positively affects social performance of SMEs.

AI adoption and environmental performance

Adopting artificial intelligence (AI) technologies is pivotal for small and medium-sized enterprises (SMEs) to achieve sustainable commercial performance, particularly in enhancing their environmental performance. Environmental performance (ENP) refers to the evaluation of a company’s impact on the environment resulting from its environmentally friendly practices^168,169.

AI technologies play a significant role in reducing the ecological footprint of businesses¹⁷⁰. AI systems can optimize resource use, thereby minimizing energy consumption and reducing waste¹⁷¹. For instance, AI-powered energy management systems can analyze patterns and optimize energy use in real-time, leading to significant reductions in energy consumption and greenhouse gas emissions^172,173. Similarly, AI can enhance supply chain efficiency, reducing material waste and improving overall resource management¹⁷⁴.

Compliance with environmental regulations is crucial for businesses aiming to improve their environmental performance. AI technologies can assist SMEs in monitoring and ensuring compliance with environmental standards. AI-driven environmental monitoring systems can provide real-time data on emissions, waste, and other environmental indicators, enabling businesses to promptly address any non-compliance issues¹⁷⁵. By ensuring adherence to environmental regulations, companies can avoid penalties and enhance their reputation as environmentally responsible entities.

One of the key indicators of environmental performance is the reduction of waste output and emissions. AI technologies can significantly contribute to these reductions. For example, AI algorithms can optimize production processes to minimize waste and improve efficiency¹⁷⁶. AI can also be used to develop more sustainable product designs, reducing the environmental impact of products throughout their lifecycle. In future, AI algorithms can lead to a 25% increase in productivity among SMEs^177,178. Studies by^131,162 highlight that businesses adopt AI for waste management and emission reduction see notable improvements in their environmental performance. AI technologies can foster the adoption of sustainable practices within organizations. AI-driven analytics can identify areas where sustainable practices can be implemented, such as reducing water usage, improving recycling processes, and enhancing the efficiency of raw material use¹⁷⁹. Additionally, AI can support the development of sustainable business models by providing insights into the environmental impact of different business strategies and suggesting more sustainable alternatives¹⁸⁰.

Empirical studies provide robust evidence supporting the positive impact of AI adoption on environmental performance. They emphasize that firms with environmentally sustainable practices positively impact on the environment both locally and globally. A study by¹⁷¹ found that AI technologies enable companies to achieve better environmental performance by optimizing logistics and reducing carbon footprints. Moreover, research by⁸⁷ indicates that AI-driven innovations in energy management and resource optimization significantly enhance a firm’s environmental performance. The hypothesis posited in the present investigation is as follows:

H4(c)

AI adoption positively affects environmental performance of SMEs.

This research aims to explore how artificial intelligence (AI) is being used by XXXi small and medium-sized firms (SMEs) and how this is affecting performance as a three-dimensional construct (i.e., social, environmental, and economic performance). Figure 1 displays the conceptualized model.

Fig. 1.

Proposed research model.

Methodology

Research design

This study employs a quantitative research design to examine the impact of “artificial intelligence (AI)” adoption on the sustainable business performance of SMEs¹⁸¹. Quantitative research is chosen for its ability to provide precise, objective, and replicable results, making it suitable for analyzing the relationships between variables and testing hypotheses¹⁸². The study utilizes a survey-based approach to gather data, which is then analyzed using advanced statistical methods¹⁸³.

Data collection: procedure and sample

The present study targeted Pakistani SMEs to assess the impact of AI adoption on their sustainable business performance. According to the “Small and Medium Enterprise Development Authority (SMEDA)”, there are over 38 million SMEs in Pakistan, contributing approximately 40% to the national income⁹³. Data were collected from SMEs operating in major cities of Upper Sindh, Pakistan, including Sukkur, Larkana, Shikarpur, Jacobabad, and Khairpur. Due to the lack of a complete list of SMEs in Upper Sindh, a non-probability sampling method was employed, specifically snowball sampling, which is suitable for researchers’ facing challenges in locating suitable participants. Researchers found an initial contact through local Chambers of Commerce and Industries (located in Sukkur, Larkana, Shikarpur, Jacobabad, and Khairpur) associations that deal with SMEs formally or informally. While snowball sampling effectively facilitated access to hard-to-reach SME participants, its limitations regarding generalizability and potential biases, such as social network, selection, information, and volunteer biases, were acknowledged. To mitigate these issues, initial participants (seeds) were chosen to ensure diversity across industries and business scales. Additionally, recruitment efforts extended beyond snowball sampling through social media outreach and SME support networks. Demographics were regularly monitored to maintain balanced representation, and incentives were provided to encourage participation from underrepresented groups. The study transparently reports these limitations to ensure rigor and validity. The selection of Sukkur, Larkana, Shikarpur, Jacobabad, and Khairpur was based on their geographic distribution, economic diversity, and mix of urban-rural settings. These cities represent key SME hubs, with Sukkur serving as an agricultural and trade center, Larkana as an industrial base, and Shikarpur as a trading hub, while Jacobabad and Khairpur exhibit growing SME clusters. This strategic choice ensures a representative dataset to comprehensively analyze AI adoption among SMEs¹⁸⁴.

A total of 400 questionnaires were distributed through face-to-face interactions with owners and managers of SMEs, resulting in 315 completed questionnaires, representing a 78.75% response rate. After screening the final questionnaire 305 found valid filled questionnaire and considered for final data analysis. The survey was conducted over a two-month period (February and March 2024). Participants voluntarily took part in the survey, and anonymity was ensured by excluding questions about their identities. The questionnaire was distributed in sealed envelopes and collected either directly from participants or through representatives of the researchers.

The sample size exceeds the minimum recommended by Reinartz et al.¹⁸⁵, who suggest a sample size of 100 for structural equation modeling using partial least squares (PLS)¹⁸⁵. A pilot study involving 30 SMEs was conducted to ensure the clarity and reliability of the survey instrument, resulting in a Cronbach’s alpha exceeding 0.7, indicating high reliability^186,187.

Measures

The questionnaire for this study was developed based on existing literature, with minor modifications to suit the context of Pakistani SMEs¹⁸⁸. The survey was originally formulated in English and subsequently translated into Urdu and Sindhi, the local languages of Upper Sindh, to ensure participants could easily comprehend the questions. The translation was conducted by professors holding M.Phil. and PhD degrees in Urdu and Sindhi language subjects, respectively, who are currently serving as faculty members at Shah Abdul Latif University in Pakistan. These subject matters and language experts were chosen for their academic qualifications and extensive experience in the respective languages. To ensure the accuracy and validity of the translation, the translators worked closely with the research team, who reviewed and verified the translated survey instrument. Additionally, a back-translation procedure was employed, wherein the translated version was retranslated into English by independent language experts to confirm the fidelity and clarity of the content. The finalized questionnaire items were derived from previous studies and organized into two sections: demographic information and constructs under investigation. A five-point Likert scale, ranging from 1 (“strongly disagree”) to 5 (“strongly agree”), was used to measure each item, providing a balanced range of responses and preventing overemphasis on extremes¹⁸⁹. This scale is commonly used in business management researches^{3,8,12,190–193}. The survey items were adapted from¹⁹¹.

Statistical analysis

SEM was used in this study’s data analysis to assess the suggested hypotheses¹⁹⁴. SEM is a statistical method for analyzing measurement and structural models as well as intricate correlations between variables^195–198.

Confirmatory factor analysis (CFA)

Convergent validity and the causal relationships between the updated items and the variables in the measurement model were evaluated using CFA^199,200. CFA tests the measurement model’s goodness of fit by assessing how well observed variables match the latent constructs they are meant to measure^201,202.

Structural model

Following CFA, the structural model was used to examine the relationships between exogenous (independent) and endogenous (dependent) variables^201,202. The researchers utilized SmartPLS 4 software for SEM analysis. SmartPLS 4 includes several enhancements, such as a revamped Graphical User Interface (GUI), side-by-side data comparison, report saving, customizable charts, and support for various data formats^201,202. PLS-SEM, a variance-based method, is suitable for predictive applications and does not require normally distributed data. Additionally, PLS-SEM supports nonparametric multigroup analysis, allowing for group comparisons^203–205.

In conjunction with SEM, the Artificial Neural Network (ANN) approach was employed to determine the importance of external structures in predicting internal constructs²⁰⁶. ANN mimics the human brain’s structure and function, enhancing knowledge through learning processes and aiding researchers in predicting the significance of antecedents¹³⁴. This dual-method approach ensures a comprehensive analysis of the data and robust validation of the research model.

Results

Demographic profiles

Out of the 305 SME respondents, 73.8% were men, and the remainder were women (Table 1). The majority of respondents were between the ages of 31 and 40 (33.4%). Furthermore, most respondents worked in the retail industry (28.5%). The remainder worked in wholesale (11.1%), manufacturing (18.7%), services (16.7%), livestock (4.90%), agricultural (16.8%), and poultry (3.0%). Lastly, 34.8% firms were in Sukkur and 29.2% were also located in Larkana. Whereas 15.7% were in Khairpur and 8.5% were in Jacobabad.

Table 1.

Demographic information.

Demographic characteristics	N (305)	(%)
Gender
Male	225	73.8
Female	80	26.2
Total	305	100
Age
20–30 years	81	26.6
31–40 years	102	33.4
41–50 years	53	17.4
51–60 years	55	18.0
61 years & above	14	4.60
Total	305	100
Location of firms
Sukkur	106	34.8
Larkana	89	29.2
Khairpur	48	15.7
Jacobabad	26	8.5
Shikarpur	36	11.8
Total	305	100
Business type
Manufacturing	57	18.7
Retailing	87	28.5
Wholesaling	37	12.1
Agriculture	49	16.1
Livestock	15	4.9
Poultry	9	3.0
Services	51	16.7
Total	305	100

SEM-based analysis

Evaluation of the measurement model

The study’s analysis was conducted in two stages, involving a measurement model and a structural model, following the guidelines provided by^194,207. The measurement model was validated using tests for convergent and discriminant validity²⁰¹. To establish the measurement model, the thresholds of factor loadings, average variance extracted (AVE), and composite reliability (CR) were considered. Specifically, a factor loading of 0.4 or above, an AVE of at least 0.5, and a CR of 0.7 or higher were required^187,193. During the confirmatory factor analysis (CFA), one item from the Complexity construct” The use of artificial intelligence would be frustrating”—was removed due to its factor loading of 0.35, which fell below the recommended threshold of 0.4. This low loading indicated that the item did not sufficiently represent the latent construct of Complexity and may have affected the overall validity and reliability of the measurement model. Removing this item improved the overall fit and ensured that the remaining indicators effectively captured the intended construct. Table 2 presents the results of the convergent validity test, showing that all remaining loadings, AVE, and CR values exceeded their respective threshold values. This confirms the convergent validity of the measurement model²⁰⁷.

Table 2.

Measurement of model evaluation.

Constructs	Factor loading	α	AVE	CR
Availability of financial resource (AFR)	0.78	0.76	0.66	0.85
	0.88
	0.78
AI adoption (AI)	0.88	0.86	0.78	0.91
	0.88
	0.88
Competitor pressure (Comp)	0.84	0.79	0.88	0.88
	0.80
	0.87
Customer pressure (Custp)	0.80	0.76	0.86	0.86
	0.83
	0.84
Complexity (Complex)	0.88	0.85	0.91	0.91
	0.90
	0.86
Cost (Cost)	0.79	0.93	0.94	0.94
	0.94
	0.90
	0.93
Economic performance (ECOP)	0.81	0.85	0.90	0.90
	0.82
	0.88
	0.82
Employee capability (ECP)	0.83	0.82	0.89	0.89
	0.85
	0.88
Environmental performance (ENVTP)	0.74	0.80	0.87	0.87
	0.84
	0.82
	0.76
Relative advantage (RA)	0.79	0.83	0.89	0.89
	0.82
	0.85
	0.82
Social performance (SOCP)	0.77	0.81	0.87	0.87
	0.82
	0.79
	0.72
Top management support (TMS)	0.85	0.87	0.91	0.91
	0.86
	0.87
	0.81
Vendor support (SP)	0.83	0.86	0.90	0.90
	0.88
	0.85
	0.76

Note: CR = composite reliability; AVE = average variance extracted; α = Cronbach Alpha

To ensure discriminant validity, we followed the recommendation of Franke and Sarstedt²⁰⁸ and used the heterotrait-monotrait (HTMT) ratio^194,209. This leads us to believe that “theoretically different constructs should have HTMT ratio below the threshold of 0.90 for considerable discriminant validity” as seen in Table 3. Because no HTMT value was more than 0.90, the study was able to establish discriminant validity between latent components using the HTMT ratio.

Table 3.

Heterotrait-monotrait ratio (HTMT.₈₅).

	AFR	AI	COMP	CUSTP	Complex	Cost	ECOP	ECP	ENVTP	RA	SOCP	TMS	VS
AFR
AI	0.12
COMP	0.06	0.08
CUSTP	0.15	0.67	0.07
Complex	0.05	0.18	0.67	0.16
Cost	0.07	0.03	0.05	0.08	0.06
ECOP	0.11	0.75	0.1	0.72	0.14	0.04
ECP	0.17	0.75	0.08	0.72	0.14	0.05	0.08
ENVTP	0.12	0.68	0.10	0.74	0.10	0.05	0.78	0.72
RA	0.12	0.68	0.10	0.64	0.04	0.08	0.78	0.71	0.75
SOCP	0.11	0.68	0.08	1.13	0.16	0.08	0.77	0.69	0.84	0.73
TMS	0.11	0.72	0.1	0.64	0.07	0.06	0.75	0.74	0.75	0.72	0.72
VS	0.15	0.11	0.08	0.09	0.16	0.11	0.07	0.10	0.05	0.07	0.10	0.08

To prove discriminant validity, it is said by Fornell and Lacker²¹⁰ that each construct must have a variation in its measures that is greater than the variance in the constructs shared by other constructs. Hence, the measures presented in Table 4 demonstrate sufficient discriminant validity due to the fact that the correlation coefficient for each construct (in both the column and row components) is lower than the average variance extracted (AVE) by the indicators used to assess that construct, as indicated on the diagonal²¹⁰.

Table 4.

Discriminant validity test (Fornell and Larcker criteria).

	AFR	AI	COMP	CUSTP	Complex	Cost	ECOP	ECP	ENVTP	RA	SOCP	TMS	VS
AFR	0.81
AI	-0.1	0.88
COMP	-0.04	0.06	0.84
CUSTP	-0.13	0.54	0.04	0.82
Complex	-0.03	0.16	0.55	0.14	0.88
Cost	0.06	0.03	-0.04	-0.06	0.07	0.89
ECOP	-0.1	0.64	0.07	0.58	0.13	-0.01	0.83
ECP	-0.15	0.63	0.06	0.57	0.12	-0.02	0.9	0.85
ENVTP	-0.09	0.56	0.04	0.58	0.07	0.00	0.65	0.58	0.79
RA	-0.06	0.58	0.08	0.51	0.01	0.07	0.66	0.59	0.61	0.82
SOCP	-0.08	0.57	0.06	0.9	0.13	-0.03	0.64	0.56	0.67	0.6	0.76
TMS	-0.10	0.63	0.08	0.52	0.07	-0.04	0.65	0.62	0.63	0.62	0.6	0.85
VS	0.12	0.11	-0.07	-0.07	-0.14	0.10	0.06	0.09	0.04	0.06	-0.02	0.08	0.83

Structural model analysis

We started by looking for evidence of collinearity in the structural model by analyzing the internal variance inflation factor (VIF)^201,211. It is worth mentioning that the maximum VIF value obtained was 2.201. This figure is lower than the recommended threshold of 5, which was set by²¹² and indicates that there is no collinearity. In addition, we used the coefficient of determination (R2) to evaluate the variables. According to Table 5, all of these evaluations had results that were higher than their specified cutoff points^213,214.

Table 5.

Coefficient of determination of endogenous constructs.

Variables	R 2	Adjusted R2	Remarks	Benchmarks
AI	0.54	0.53	Moderate	0.75 = > Substantial 0.50 = > Moderate 0.25 = > Weak Hair et al.204
ECOP	0.41	0.41	Weak
ENVTP	0.32	0.31	Weak
SOCP	0.33	0.33	Weak

The bootstrapping method in SmartPLS 4 was used to test hypotheses, with default parameters and 5000 subsamples. In order to assess the importance of relationships, the beta (β) values were examined together with their corresponding t-values and p-values. You may get detailed results from bootstrapping in Table 6. In Fig. 2, you can see the bootstrapping results. Building on the standardized path coefficients of the structural model demonstrated in Table 6, the empirical results of effects indicate a significant and positive association between AI and ECOP (β: 0.64, p = 0.00, t = 13.27), ENVTP and AI (β: 0.56, p = 0.00, t = 10.23), AI and SOCP (β: 0.57, p = 0.00, t = 12.85) respectively. While a significant and positive relationship between CUSTP and AI (β: 0.16, p = 0.00, t = 2.18) and Complex and AI (β: 0.12, p = 0.00, t = 2.26) was also found. Lastly, ECP cause a positive and significant impact on AI (β: 0.25, p = 0.00, t = 3.43), RA and AI (β: 0.18, p = 0.00, t = 2.69), TMS and AI (β: 0.27, p = 0.00, t = 2.73) and VS and AI (β: 0.08, p = 0.00, t = 1.99) respectively. Whereas a few relationships were also found negative and insignificant. In this connection, AFR impacts negatively and insignificantly on AI (β: 0.-0.02, p = 0.68, t = 0.410), COMP and AI (β: 0.64, p = 0.00, t = 13.27) and COST and AI (β: 0.03, p = 0.53, t = 0.630) respectively.

Table 6.

Summary of hypotheses testing.

Hypotheses	β	p-value	t-value	Decision
H1(a): Implementation cost has a negative effect on AI adoption by SMEs.	0.03	0.68	0.630	Unsupported
H1(b): Relative advantage has a positive effect on AI adoption.	0.18	0.01	2.69	Supported
H1(c): Complexity has a negative effect on AI adoption.	0.12	0.02	2.26	Supported
H2(a): Top management support has a positive effect on AI adoption.	0.27	0.01	2.73	Supported
H2(b): Availability of Financial Resource has a positive effect on AI adoption.	− 0.02	0.68	0.410	Unsupported
H2(c): Employee Capability has a positive effect on AI adoption.	0.25	0.00	3.43	Supported
H3(a): Competitor Pressure has a positive effect on AI adoption.	− 0.05	0.86	0.410	Unsupported
H3(b): Customer Pressure has a positive effect on AI adoption.	0.16	0.00	2.81	Supported
H3(c): Vendor Support has a positive effect on AI adoption.	0.08	0.05	1.99	Supported
H4(a): AI adoption positively affects economic performance of SMEs.	0.64	0.00	13.27	Supported
H4(b): AI adoption positively affects social performance of SMEs.	0.57	0.00	12.85	Supported
H4(c): AI adoption positively affects environmental performance of SMEs.	0.56	0.00	10.32	Supported

Fig. 2.

Bootstrapping results.

Artificial neural network analysis

An input, hidden, and output layer multi-layer artificial neural network was utilised. The statistical programme SPSS 29 was utilised for this purpose. Each output neurone in the ANN architecture has two hidden layers in order to facilitate deeper learning^215,216. To minimize the possibility of overfitting, we calculated the RMSE using a ten-fold cross-validating procedure²¹⁶. After consulting Leong et al.²¹⁷, we divided the samples in half and used 90% for training and the other half for testing. Both the hidden and output layers made use of the sigmoid activation function, with the automatic generation of the number of hidden layers being left alone. The relative significance of antecedents was determined by using a ratio of 90:10 data for training and testing of prediction precision²¹⁸. This ratio was utilized to gather data for training and testing. The value of the mean square root error (RMSE) was utilized in both the training and testing data sets in order to assess the estimated correctness of the artificial neural network (ANN) model. Table 7 illustrates that, at 1.8102 and 0.5563, respectively, the average RMSE values of the training and testing processes are quite low. As a result, we can attest to the good model fit. The RMSE score is always positive, according to Hyndman and Koehler, and a value of 0 denotes flawless accuracy (almost unheard of real predictive practise).

Table 7.

RMSE values.

Neural network	COMP	CUSTP	ECP	RA	TMS	VS
NN (i)	0.19	0.75	1.00	0.33	0.76	0.18
NN (ii)	0.11	0.40	0.51	0.32	1.00	0.32
NN (iii)	0.24	0.94	1.00	0.85	0.51	0.33
NN (iv)	0.16	0.48	1.00	0.49	0.86	0.28
NN (v)	0.10	0.69	0.66	0.69	1.00	0.30
NN (vi)	0.24	0.69	1.00	0.62	0.78	0.25
NN (vii)	0.10	0.80	0.78	0.53	1.00	0.28
NN (viii)	0.19	0.33	0.70	0.57	1.00	0.17
NN (ix)	0.07	0.49	0.60	0.45	1.00	0.47
NN (x)	0.18	0.39	1.00	0.78	0.57	0.40
Average importance	0.16	0.60	0.82	0.56	0.85	0.30
Normalized importance (%)	18%	39%	100%	78%	57%	40%

Note: COMP = Competitor Pressure, CUSTP= Customer Pressure, ECP = Employee Capability, RA = Relative Advantage, TMS = Top Management Support, VS = Vendor Support

Ranking of predictors

A sensitivity analysis was carried out (Table 8) in order to determine the normalized relevance of these neurons. This was done in order to evaluate the predictive power of each input neuron. This was achieved by dividing their relative importance by the maximum importance and presenting the results as a percentage (Karaca et al., 2019). Based on the findings, it can be shown that COMP (18%), CUSTP (39%), VS (40%), TMS (57%) and RA (78%) emerge as the subsequent most influential predictors following ECP (100%).

Table 8.

Sensitivity analysis.

Training			Testing
N	SSE	RMSE	N	SSE	RMSE	Total Samples
271	68.920	1.9829	34	4.96	0.3819	305
277	61.040	2.1302	28	6.19	0.4702	305
278	71.520	1.9715	27	5.940	0.4690	305
271	66.820	2.0138	34	4.510	0.3642	305
265	60.740	2.0887	40	15.340	0.6193	305
275	68.470	2.0040	30	9.84	0.5727	305
266	56.100	2.1775	39	10.59	0.5211	305
274	60.740	2.1239	31	3.600	0.3408	305
271	50.580	2.3147	34	12.01	0.5943	305
277	82.610	1.8311	28	3.4	0.3485	305
271	68.920	1.9829	34	4.96	0.3819	305
277	61.040	2.1302	28	6.19	0.4702	305
Mean	64.754	2.0638	Mean	7.638	0.4682
SD	8.4962	0.12639	SD	3.8496	0.1006

Discussion

This study utilized a quantitative approach (PLS and ANN) to investigate the effect of adoption of Artificial Intelligence by SMEs on their Sustainable performance. The results indicate that top management support, employee capability, customer pressure, complexity, vendor support and relative advantage exert a positive and significant influence on adoption of artificial intelligence. It is important that artificial intelligence has also shown a positive and significant impact on economic performance, social performance, and environmental performance of SMEs (H4a, H4a b and H4a c). Using a two-fold SEM-ANN analysis, we validated our model and obtained insightful findings.

Technological factors and AI adoption

The Technology-Organization-Environment (TOE) framework was instrumental in analyzing technological determinants, including relative advantage, complexity, and cost. Relative advantage (H1b: β = 0.18, p = 0.01) emerged as a significant factor, aligning with previous studies^219,220. SMEs in Pakistan perceive AI as superior to traditional technologies, enhancing their competitive positioning, efficiency, and organizational reputation. This aligns with²²¹, where perceived technological benefits drive technology adoption. In contrast, implementation cost (H1a: β = 0.03, p = 0.68) did not significantly affect AI adoption. This unexpected result may indicate that financial considerations are overshadowed by the strategic importance of AI adoption^73,222. SMEs may prioritize AI’s long-term value creation over immediate financial costs, particularly in competitive markets. However, it is also plausible that respondents perceive financial barriers as secondary to technological capability or external pressures, as supported by¹⁰². Complexity (H1c: β = 0.12, p = 0.02) had a positive impact, contrary to studies suggesting complexity hinders adoption. This result highlights a contextual difference, where SMEs in Pakistan, particularly in manufacturing and retail sectors, do not find AI overly complicated. The TOE framework assumes that when technologies are perceived as manageable, adoption becomes more likely.

Organizational factors and AI adoption

The impact of organizational factors on AI is investigated in the second part of the TOE framework. The organizational factors that impact the adoption of new technologies by SMEs have been the subject of numerous research³⁹. Top management support (H2a: β = 0.27, p = 0.01) was a critical organizational factor, corroborating findings^39,138,223. Effective leadership is essential for resource allocation, risk mitigation, and fostering an innovation-driven culture. In alignment with the TOE framework, strong leadership enhances organizational readiness and facilitates AI adoption. Additionally, the availability of financial resources (H2b: β=-0.02, p = 0.68) did not significantly influence AI adoption, contradicting studies by¹⁰². This could be due to the limited financial capacities of SMEs in Pakistan, where resource constraints are persistent. Moreover, SMEs may seek alternative funding mechanisms, such as vendor support, or rely on incremental technology adoption to bypass cost barriers. Employee capability (H2c: β = 0.25, p < 0.001) significantly influenced AI adoption, supporting findings from Cruz-Jesus et al.³⁹. Skilled employees, particularly in manufacturing, agriculture, and retail sectors, play a critical role in technology integration. Variations in employee competence across industries highlight the need for specific training programs to enhance AI proficiency.

Environmental factors and AI adoption

Environmental factors, including customer pressure and vendor support, positively influenced AI adoption. Customer pressure (β = 0.22, p < 0.001) highlights the role of external stakeholders in driving technological innovation, consistent with ²²⁴ and ²²⁵. SMEs operating in competitive markets are compelled to adopt AI to meet evolving customer demands and enhance service delivery. Vendor support (β = 0.29, p < 0.001) also emerged as a significant driver, aligning with studies by^102,226. Vendors provide critical technical and operational assistance, reducing implementation complexities and fostering confidence in AI adoption. Interestingly, competitor pressure (H3a: β=-0.05, p = 0.86) was insignificant, diverging from prior research²²⁴. This may reflect SMEs’ financial constraints and limited exposure to competitive technological advancements. Unlike large enterprises, SMEs may not perceive competitor-driven AI adoption as an immediate necessity.

AI and sustainable performance

The study confirmed the substantial influence of AI on SMEs’ economic, social, and environmental performance (H4a-c). Consistent with^79,227,228. AI adoption enhances resource efficiency, reduces operational costs, and improves economic outcomes. For SMEs in Pakistan, technological integration optimizes processes, driving profitability and growth. From a social perspective, AI fosters workplace safety, predictive maintenance, and personalized customer experiences, aligning with²²⁹. These advancements contribute to employee satisfaction and improved customer relationships, reinforcing AI’s social value for SMEs. Environmental performance also benefitted significantly, as AI facilitates waste reduction, resource optimization, and sustainable practices²³⁰. These findings highlight the role of AI as a catalyst for SMEs’ transition toward environmentally responsible operations.

The endorsement of governmental bodies and other organizations is important for the effective implementation of AI in SMEs. The government can help small businesses a lot by making the laws and rules better, letting more people know about the digital shift, giving them money and technical help, and making the infrastructure for data exchange stronger^231,232. Prior research has identified government policies, such as offering financial incentives, scientific resources, pilot projects, and training programs, as influential elements in motivating small and medium-sized enterprises (SMEs) to embrace new technology and environmentally friendly practices^139,233,234. Furthermore, the ANN findings and rankings of the important factors are not substantially distinct from the SEM results when it comes to drawing conclusions. This is because ANN can analyse both linear and non-linear correlations, whereas SEM is limited to detecting just linear relationships. Nevertheless, NNA offers additional validation of the SEM results.

Theoretical implications

This research fills a research gap in the literature regarding the use of AI in small businesses. Most existing research focuses on large organizations and developing countries. By examining SMEs in Pakistan, a developing nation, this study provides valuable empirical evidence on how AI technologies are adopted in different economic and cultural contexts. Previous studies have highlighted that the adoption of innovative technologies is influenced by various factors such as the legal environment, technological infrastructure, economy, and culture. This study extends this knowledge by exploring these factors in the context of Pakistani SMEs.

Furthermore, this research makes a methodological contribution by combining Partial Least Squares Structural Equation Modelling (PLS-SEM) with Artificial Neural Networks (ANN). While previous studies have primarily used SEM to identify linear, our approach provides understanding by netting both linear and non-linear relationships among the variables. The integration of ANN with PLS-SEM yields more robust results and demonstrates the high precision of ANN, contributing to the methodological advancement in the field.

Practical implications

This study underscores the need for strong top management support and skilled employees to facilitate AI adoption among SMEs. Prioritizing training programs, upskilling initiatives, and collaboration with technology vendors, research institutes, and AI specialists can help SMEs overcome capability and resource gaps. Top management plays a critical role by providing leadership, committing resources, fostering an innovative culture, and ensuring ethical AI practices. Financial barriers remain a significant challenge, particularly in emerging economies, where high implementation costs, rising energy prices, and limited funding restrict innovation. Policymakers must address these issues by offering interest-free loans, grants, subsidies, tax incentives, and infrastructure support to reduce financial burdens and enable SMEs to adopt AI technologies. SMEs should also align AI initiatives with sustainability goals to promote green practices, monitor environmental impact, and comply with regulations, particularly in regions affected by climate change. AI adoption can enhance SMEs’ competitiveness by improving operational efficiency, market image, and eco-conscious consumer engagement while contributing to societal and environmental sustainability. By strategically addressing these challenges and opportunities, SMEs can achieve long-term growth, resilience, and positive societal impact.

Conclusion

This study finds the impact of AI adoption on the sustainable performance of small and medium-sized enterprises (SMEs) in Pakistan, employing a combined Partial Least Squares (PLS) and Artificial Neural Network (ANN) approach. Data from 305 SME owners and managers reveal that top management support, employee capability, customer pressure, complexity, vendor support, and relative advantage significantly influence AI adoption. PLS-SEM results show that relative advantage (β = 0.18), top management support (β = 0.27), employee capability (β = 0.25), customer pressure (β = 0.16), and vendor support (β = 0.08) positively affect AI adoption, while complexity has a negative effect (β = 0.12). AI adoption positively impacts economic (β = 0.64), social (β = 0.57), and environmental (β = 0.56) performance. ANN analysis confirms these findings, ranking employee capability, relative advantage, and top management support as the most influential factors.

The ANN analysis further validated these findings by ranking the importance of different predictors. Employee capability (ECP) emerged as the most influential predictor (100%), followed by relative advantage (78%), top management support (57%), vendor support (40%), customer pressure (39%), and competitor pressure (18%).

The quantitative methodology applied in this research, particularly the SEM-ANN model, allowed for the validation of our hypotheses and the identification of both linear and non-linear relationships among the variables. Key findings reveal that all the investigated factors significantly and positively influence AI adoption. Moreover, the adoption of AI technologies substantially improves the economic, social, and environmental performance of SMEs, highlighting its broad spectrum of benefits.

The findings emphasize significant policy and managerial contributions. Policymakers should offer financial support, such as grants and low-interest loans, and invest in training programs to address skill gaps in SMEs. Managers should focus on enhancing employee capabilities, securing top management support, and forming strategic partnerships with AI experts to strengthen technological competencies. These efforts will enable SMEs to harness AI’s potential, improving their economic, social, and environmental performance, and enhancing their competitiveness and sustainability in an AI-driven market.

Limitations and future research avenues

This study’s results and implications are significant, several limitations need to be acknowledged. One primary limitation is the sample size, which could be expanded to include respondents from different geographical areas within Sindh province and across Pakistan, thus enhancing the generalizability of the findings. Future research could replicate this study in other emerging countries for comparative analysis. Another limitation is the exclusive use of quantitative research techniques. Future studies should incorporate qualitative methods to gain a deeper understanding of the factors influencing AI adoption. Additionally, the study utilized cross-sectional data, which provides a snapshot at one point in time. Longitudinal studies would be valuable to measure the long-term impact of AI adoption on the sustainable performance of SMEs.

The use of snowball sampling, a non-probability sampling technique, is another limitation. Snowball sampling relies on existing participants to recruit future participants from among their acquaintances, which may introduce bias and limit the representativeness of the sample. Future research should employ more robust sampling methods, such as combining snowball sampling with random or stratified sampling to ensure a more diverse and representative sample.

In terms of methodology, this research innovatively combines Structural Equation Modeling (SEM) and Artificial Neural Networks (ANN) to examine AI adoption in SMEs. While this combined SEM-ANN approach offers significant advantages, including the ability to capture non-linear relationships and manage multiple latent variables, it also has limitations. Specifically, the ANN model used in this study may face challenges in generalizing beyond the sample studied due to its reliance on the specific data used for training the model. ANN models are highly sensitive to the training data, and as a result, their performance may degrade when applied to different or unseen datasets. Future studies should consider validating the ANN model in different contexts or with a larger, more diverse sample to assess its generalizability. Lastly, the combined SEM-ANN methodology allows for a broader understanding of AI adoption in SMEs by highlighting key factors that influence adoption, providing valuable insights for practitioners. Future research could further explore this hybrid approach to AI adoption, refining the model and its applicability across various industries and regions.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

R. S, wrote the main manuscript, Methodology, Investigation, Data curation, Writing—review & editingW M.R, wrote the main manuscript, Methodology, Investigation, Writing—review & editingN.A.D, Formal analysis, Investigation, Writing—review & editingL.A, Methodology, Writing—original draft, Writing—review & editing, Funding acquisitionA.S.A, Software, Resources, Writing—original draft, Writing—review & editing, Funding acquisitionA. A, Formal analysis, Data curation, Writing—review & editingAll authors have read and agreed to the published version of the manuscript.

Funding

The authors extend their appreciation to the Princess Nourah bint Abdulrahman University Researchers Supporting Project number (PNURSP2025R349), Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia. In addition, this work was supported by the King Saud University, Riyadh, Saudi Arabia, through Re-searchers Supporting Project no. RSP-2025/R417.

Data availability

The author confirms that all data generated or analyzed during this study are included in this published article. Furthermore, all analyzed data supporting this study’s findings were publicly available at the time of submission.

Declarations

Competing interests

The authors declare no competing interests.

Ethical approval

In accordance with ethical standards, I hereby confirm that the research study mentioned above involved the collection of data from Sindh Pakistan, and prior ethical approval was duly obtained, under Reference No. RMC/Q.J130000.21A2.07E10. Additionally, formal permissions were granted by the University of Shah abdul Latif University via Letter No. IBA/SALU/120/Dated: 28-2-2024 and the University Teknologi Malaysia (UTM) through Letter Reference No. UTM.J.13.01/13.14/1/88 Jld.23(75)/Dated: 1-06-2023 and under RMC research project No. Q.J130000.21A2.07E10. Copies of these approval letters are attached herewith for reference and verification, confirming that all necessary ethical and regulatory requirements have been met throughout the course of this research project.

Informed consent

Informed consent was obtained in written form from all participants in the research. Participants were informed about the use of the data (e.g., scientific publication) and their right to decide what happens to the (identifiable) personal data gathered.