University of Isfahan New Marketing Research Journal 2228-7744 16 2 2026 06 22 Opportunities, Challenges, and Requirements of Marketing in the Metaverse World Opportunities, Challenges, and Requirements of Marketing in the Metaverse World 1 30 30367 10.22108/nmrj.2026.145893.3215 FA Zahra Solgi Assistant professor, Department of Management, Faculty of Management and Economics, University of Lorestan, Khorramabad, Iran Amin Hasanvand Assistant professor, Department of Law, Faculty of Literature and Humanities, University of Lorestan, Khorramabad, Iran Journal Article 2025 07 09 This study aimed to identify the opportunities, challenges, and requirements of marketing in the metaverse. In terms of purpose, it was applied research and it was categorized as a descriptive–survey study by using data collection method. Given the nature of the data and the analytical approach, a mixed-methods design was adopted. In the qualitative phase, data were gathered through a literature review and analyzed using the meta-synthesis method with Atlas.ti software, yielding 86 open codes, 16 subcategories, and 3 main categories. In the quantitative phase, the fuzzy Delphi method was employed to screen and validate the indicators and categories derived from the qualitative stage. A panel of 18 experts was selected through purposive non-probability sampling based on their expertise in digital marketing, experience with the metaverse, and relevant professional background. Validity and reliability were confirmed using content validity and the test–retest method. The main contribution of this study was the development of a comprehensive framework for marketing in the metaverse, which classified key opportunities, challenges, and requirements. The findings indicated that immersive user experiences, personalization, community building through value creation, extensive interaction, and alignment with market trends represented major opportunities. In contrast, brand reputation risks, regulatory and legal complexities, intellectual property issues, and data security concerns were identified as key challenges. Furthermore, organizational transformation, context-appropriate marketing policies, market intelligence capabilities, and responsible platform governance were recognized as essential requirements for effective marketing in metaverse environments. Introduction The rapid and continuous evolution of digital technologies over the past two decades has fundamentally transformed the marketing landscape. Businesses have moved beyond static websites and social media platforms toward immersive, interactive, and intelligent environments. Within this trajectory, the metaverse—anchored in key innovations, such as Virtual Reality (VR), Augmented Reality (AR), and other immersive technologies—has garnered considerable attention from both researchers and marketing practitioners. By integrating physical and virtual spaces, the metaverse blurs the traditional boundaries of time, place, and physical presence, offering a platform where users can engage in immersive, collaborative, and interactive experiences with brands and other users through personalized avatars. A review of the literature reveals that most existing studies have focused on introducing the metaverse, describing its applications in gaming and entertainment or conducting preliminary examinations of consumer behavior in virtual environments. In contrast, systematic and comprehensive research addressing the opportunities, challenges, and requirements of marketing in the metaverse remains limited. This theoretical and practical gap poses a significant obstacle, particularly for businesses seeking to redesign their marketing strategies for this new space. Therefore, a comprehensive framework is urgently needed to delineate and categorize the key dimensions of metaverse marketing. Accordingly, the present study aimed to identify and explicate the opportunities, challenges, and requirements for marketing in the metaverse. The central research question was as follows: What opportunities and challenges does marketing in the metaverse entail and what organizational, technological, and governance requirements must be considered for effective engagement in this environment? Materials & Methods This study was applied in purpose and descriptive–survey in nature, employing a mixed-methods design conducted in two sequential phases: qualitative and quantitative. In the qualitative phase, a meta-synthesis approach was adopted to provide a comprehensive overview of the existing literature on the metaverse and marketing and to extract key concepts and indicators related to the opportunities, challenges, and requirements of marketing in this domain. A systematic search of reputable academic databases was conducted to identify relevant studies on marketing, branding, customer experience, and technological transformations in the metaverse. Following primary and secondary screening, eligible articles were selected and analyzed. Qualitative data analysis was performed using Atlas.ti software through a 3-stage coding process. First, open coding was applied to extract initial concepts and key statements from the articles, yielding 86 open codes. Second, these codes were grouped into 16 subcategories based on conceptual similarity. Finally, through axial and selective coding, the subcategories were organized into 3 main categories: "Marketing Opportunities in the Metaverse", "Marketing Challenges in the Metaverse", and "Requirements for Success in the Metaverse". This conceptual structure formed the basis of the initial research model. In the quantitative phase, the fuzzy Delphi method was employed to validate and prioritize the identified indicators and categories. The statistical population consisted of experts in digital marketing and the metaverse selected through purposive non‑probability sampling. Inclusion criteria comprised a relevant academic background (at least a master's degree), research or practical experience in digital marketing, familiarity with metaverse concepts and immersive technologies, and managerial or consulting experience in digital businesses. The expert panel ultimately included 18 university faculty members, managers, and marketing specialists. The data collection tool was a structured questionnaire developed based on the qualitative findings, in which experts rated the importance of each indicator using linguistic fuzzy scales. Data were analyzed following standard fuzzy Delphi procedures—including fuzzification, aggregation of expert opinions, and defuzzification—to retain indicators approaching consensus and eliminate those deemed less significant or ambiguous. This process enabled the refinement of the initial model into the final framework. Research Findings The qualitative findings indicated that the metaverse as an emerging marketing environment created a diverse range of opportunities for businesses. Identified opportunities included immersive and holistic customer experiences, deep personalization of interactions, community building and brand-centric groups, extensive bidirectional customer engagement, and market adaptation and innovation. These opportunities enabled brands to design creative campaigns, virtual stores, and events, implement experiential and participatory strategies, leverage NFTs as symbolic assets, and collect rich behavioral data. Conversely, the results showed that marketing in the metaverse was accompanied by significant challenges. Key challenges included reputational risk in the open and public metaverse environment, conceptual and technological ambiguity, legal complexities and regulatory issues, intellectual property and concerns of content rights, and threats to data security and privacy. If not managed properly, these challenges could have serious negative consequences for brand image, customer trust, and business sustainability. The third main category—requirements for success—was articulated through subcategories, such as organizational software and hardware changes, designing and ongoing updating of marketing policies tailored to the metaverse environment, development of market intelligence and analytical capabilities, empowerment of human resources in digital and creative skills, and responsible and responsive platform governance. In other words, for effective entry into the metaverse, organizations had to invest in technological infrastructure, foster an innovative organizational culture, redesign structures and processes, and pay serious attention to ethical and regulatory standards. The fuzzy Delphi results indicated that experts generally agreed with the 3‑dimensional model structure (opportunities, challenges, and requirements) and its main subcategories although some indicators were revised in terms of their relative importance. Indicators related to immersive customer experience, community building, brand–customer interaction, data security, and platform governance received the highest weights and importance. Conversely, less significant indicators—those with weaker theoretical or practical support—were removed or merged. The outcome was a refined framework based on expert consensus. Discussion of Results & Conclusion The findings suggested that the metaverse should not be viewed merely as an additional advertising channel alongside existing platforms, such as social media or websites. Rather, it represents a novel, interactive marketing environment that challenges traditional marketing logic. The immersive and participatory nature of the metaverse enables brands to create deep, personalized, and multisensory experiences for customers, thereby enhancing engagement, loyalty, and perceived value. Simultaneously, by providing rich, real-time behavioral data, this environment opens new horizons for analyzing consumer behavior and designing data-driven strategies. However, the results emphasized that successfully leveraging these opportunities is impossible without seriously addressing the identified challenges. Reputation risks—particularly in an environment where content spreads rapidly and extensively—can cause irreversible damage to a brand image in a short period if left unmanaged. Furthermore, the conceptual ambiguity and technological instability of the metaverse require organizations to adopt an experimental, flexible, and learning-oriented approach, preparing themselves for rapid change. Legal, intellectual property, and privacy issues also necessitate the formulation and implementation of transparent and up-to-date frameworks at both organizational and transnational levels. Regarding the requirements for success, this study underscored that the synergy among technology, marketing strategy, and platform governance is a necessary condition for thriving in the metaverse. Organizations must provide the requisite technological infrastructure and develop technical capabilities related to the design and management of metaverse experiences. At the same time, they must align their marketing strategies with the interactive, participatory, and experience-driven logic of this space. Finally, by defining principles of responsible governance, organizations can prevent opportunistic behavior, data misuse, and erosion of user trust. This study aimed to identify the opportunities, challenges, and requirements of marketing in the metaverse. In terms of purpose, it was applied research and it was categorized as a descriptive–survey study by using data collection method. Given the nature of the data and the analytical approach, a mixed-methods design was adopted. In the qualitative phase, data were gathered through a literature review and analyzed using the meta-synthesis method with Atlas.ti software, yielding 86 open codes, 16 subcategories, and 3 main categories. In the quantitative phase, the fuzzy Delphi method was employed to screen and validate the indicators and categories derived from the qualitative stage. A panel of 18 experts was selected through purposive non-probability sampling based on their expertise in digital marketing, experience with the metaverse, and relevant professional background. Validity and reliability were confirmed using content validity and the test–retest method. The main contribution of this study was the development of a comprehensive framework for marketing in the metaverse, which classified key opportunities, challenges, and requirements. The findings indicated that immersive user experiences, personalization, community building through value creation, extensive interaction, and alignment with market trends represented major opportunities. In contrast, brand reputation risks, regulatory and legal complexities, intellectual property issues, and data security concerns were identified as key challenges. Furthermore, organizational transformation, context-appropriate marketing policies, market intelligence capabilities, and responsible platform governance were recognized as essential requirements for effective marketing in metaverse environments. Introduction The rapid and continuous evolution of digital technologies over the past two decades has fundamentally transformed the marketing landscape. Businesses have moved beyond static websites and social media platforms toward immersive, interactive, and intelligent environments. Within this trajectory, the metaverse—anchored in key innovations, such as Virtual Reality (VR), Augmented Reality (AR), and other immersive technologies—has garnered considerable attention from both researchers and marketing practitioners. By integrating physical and virtual spaces, the metaverse blurs the traditional boundaries of time, place, and physical presence, offering a platform where users can engage in immersive, collaborative, and interactive experiences with brands and other users through personalized avatars. A review of the literature reveals that most existing studies have focused on introducing the metaverse, describing its applications in gaming and entertainment or conducting preliminary examinations of consumer behavior in virtual environments. In contrast, systematic and comprehensive research addressing the opportunities, challenges, and requirements of marketing in the metaverse remains limited. This theoretical and practical gap poses a significant obstacle, particularly for businesses seeking to redesign their marketing strategies for this new space. Therefore, a comprehensive framework is urgently needed to delineate and categorize the key dimensions of metaverse marketing. Accordingly, the present study aimed to identify and explicate the opportunities, challenges, and requirements for marketing in the metaverse. The central research question was as follows: What opportunities and challenges does marketing in the metaverse entail and what organizational, technological, and governance requirements must be considered for effective engagement in this environment? Materials & Methods This study was applied in purpose and descriptive–survey in nature, employing a mixed-methods design conducted in two sequential phases: qualitative and quantitative. In the qualitative phase, a meta-synthesis approach was adopted to provide a comprehensive overview of the existing literature on the metaverse and marketing and to extract key concepts and indicators related to the opportunities, challenges, and requirements of marketing in this domain. A systematic search of reputable academic databases was conducted to identify relevant studies on marketing, branding, customer experience, and technological transformations in the metaverse. Following primary and secondary screening, eligible articles were selected and analyzed. Qualitative data analysis was performed using Atlas.ti software through a 3-stage coding process. First, open coding was applied to extract initial concepts and key statements from the articles, yielding 86 open codes. Second, these codes were grouped into 16 subcategories based on conceptual similarity. Finally, through axial and selective coding, the subcategories were organized into 3 main categories: "Marketing Opportunities in the Metaverse", "Marketing Challenges in the Metaverse", and "Requirements for Success in the Metaverse". This conceptual structure formed the basis of the initial research model. In the quantitative phase, the fuzzy Delphi method was employed to validate and prioritize the identified indicators and categories. The statistical population consisted of experts in digital marketing and the metaverse selected through purposive non‑probability sampling. Inclusion criteria comprised a relevant academic background (at least a master's degree), research or practical experience in digital marketing, familiarity with metaverse concepts and immersive technologies, and managerial or consulting experience in digital businesses. The expert panel ultimately included 18 university faculty members, managers, and marketing specialists. The data collection tool was a structured questionnaire developed based on the qualitative findings, in which experts rated the importance of each indicator using linguistic fuzzy scales. Data were analyzed following standard fuzzy Delphi procedures—including fuzzification, aggregation of expert opinions, and defuzzification—to retain indicators approaching consensus and eliminate those deemed less significant or ambiguous. This process enabled the refinement of the initial model into the final framework. Research Findings The qualitative findings indicated that the metaverse as an emerging marketing environment created a diverse range of opportunities for businesses. Identified opportunities included immersive and holistic customer experiences, deep personalization of interactions, community building and brand-centric groups, extensive bidirectional customer engagement, and market adaptation and innovation. These opportunities enabled brands to design creative campaigns, virtual stores, and events, implement experiential and participatory strategies, leverage NFTs as symbolic assets, and collect rich behavioral data. Conversely, the results showed that marketing in the metaverse was accompanied by significant challenges. Key challenges included reputational risk in the open and public metaverse environment, conceptual and technological ambiguity, legal complexities and regulatory issues, intellectual property and concerns of content rights, and threats to data security and privacy. If not managed properly, these challenges could have serious negative consequences for brand image, customer trust, and business sustainability. The third main category—requirements for success—was articulated through subcategories, such as organizational software and hardware changes, designing and ongoing updating of marketing policies tailored to the metaverse environment, development of market intelligence and analytical capabilities, empowerment of human resources in digital and creative skills, and responsible and responsive platform governance. In other words, for effective entry into the metaverse, organizations had to invest in technological infrastructure, foster an innovative organizational culture, redesign structures and processes, and pay serious attention to ethical and regulatory standards. The fuzzy Delphi results indicated that experts generally agreed with the 3‑dimensional model structure (opportunities, challenges, and requirements) and its main subcategories although some indicators were revised in terms of their relative importance. Indicators related to immersive customer experience, community building, brand–customer interaction, data security, and platform governance received the highest weights and importance. Conversely, less significant indicators—those with weaker theoretical or practical support—were removed or merged. The outcome was a refined framework based on expert consensus. Discussion of Results & Conclusion The findings suggested that the metaverse should not be viewed merely as an additional advertising channel alongside existing platforms, such as social media or websites. Rather, it represents a novel, interactive marketing environment that challenges traditional marketing logic. The immersive and participatory nature of the metaverse enables brands to create deep, personalized, and multisensory experiences for customers, thereby enhancing engagement, loyalty, and perceived value. Simultaneously, by providing rich, real-time behavioral data, this environment opens new horizons for analyzing consumer behavior and designing data-driven strategies. However, the results emphasized that successfully leveraging these opportunities is impossible without seriously addressing the identified challenges. Reputation risks—particularly in an environment where content spreads rapidly and extensively—can cause irreversible damage to a brand image in a short period if left unmanaged. Furthermore, the conceptual ambiguity and technological instability of the metaverse require organizations to adopt an experimental, flexible, and learning-oriented approach, preparing themselves for rapid change. Legal, intellectual property, and privacy issues also necessitate the formulation and implementation of transparent and up-to-date frameworks at both organizational and transnational levels. Regarding the requirements for success, this study underscored that the synergy among technology, marketing strategy, and platform governance is a necessary condition for thriving in the metaverse. Organizations must provide the requisite technological infrastructure and develop technical capabilities related to the design and management of metaverse experiences. At the same time, they must align their marketing strategies with the interactive, participatory, and experience-driven logic of this space. Finally, by defining principles of responsible governance, organizations can prevent opportunistic behavior, data misuse, and erosion of user trust.

https://nmrj.ui.ac.ir/article_30367_641ca3c730402223975425fd9c885e1e.pdf

University of Isfahan New Marketing Research Journal 2228-7744 16 2 2026 06 22 Designing a Competency Model of Fourth-Generation Marketing Managers (Case Study: Tourism Industry) Designing a Competency Model of Fourth-Generation Marketing Managers (Case Study: Tourism Industry) 31 66 30274 10.22108/nmrj.2026.147125.3254 FA Fatemeh Ghazanfari Fard Ph.D. student, Department of Business Administration, Faculty of Economic and Administrative Sciences, University of Mazandaran, Babolsar, Iran Morteza Movaghar Associate professor, Department of Business Administration, Faculty of Economic and Administrative Sciences, University of Mazandaran, Babolsar, Iran Abolhassan Hosseini Associate professor, Department of Business Administration, Faculty of Economic and Administrative Sciences, University of Mazandaran, Babolsar, Iran Mahmood Yahyazadehfar Professor, Department of Business Administration, Faculty of Economic and Administrative Sciences, University of Mazandaran, Babolsar, Iran Journal Article 2025 10 25 Objective: The emergence of fourth-generation technologies has profoundly transformed the tourism industry. As such, equipping marketing managers with the competencies essential for this new era is critical. This study aimed to develop a competency model for fourth-generation marketing managers within the tourism sector. Methodology: This research was applied in nature and qualitative in approach, employing grounded theory methodology. The study involved 15 experts in digital marketing specifically within the tourism industry. The sampling process utilized a purposeful and snowball approach, continuing until theoretical saturation was reached. Data were collected through semi-structured interviews, with the validity and reliability of the tool being confirmed. Data analysis was conducted using a 3-stage coding process—open, axial, and selective coding—facilitated by Maxqda software. Findings: The analysis revealed a total of 154 open codes, 34 subcategories, and 14 main categories derived from the interviews. These findings were organized into a paradigm model consisting of 6 axes: 1) Causal conditions (technological and organizational factors), 2) Contextual conditions (infrastructure and market environment), 3) Central category (ethical competence, tourism marketing, leadership, and digital skills), 4) Intervening conditions (macro-institutional and macro-environmental factors), 5) Strategies (intra-organizational and networking), and 6) Consequences (market impacts and performance outcomes). Conclusion: The insights gleaned from this study offer valuable recommendations for both managers and researchers focused on enhancing the competencies of marketing managers in the tourism industry, particularly in the context of fourth-generation marketing. Introduction Recent advancements in digital transformation, connectivity, and information accessibility have fundamentally reshaped the marketing landscape. In response, scholars have introduced Marketing 4.0 as a new paradigm that shifts marketing strategies from traditional tools to digitally driven and interactive approaches focused on building long-term customer relationships (Dash et al., 2021). Marketing 4.0 seamlessly integrates online and offline interactions within the digital economy, requiring organizations to remain flexible and adaptive to rapid technological changes (Kotler et al., 2017). These transformations have significantly impacted the tourism industry, leading to structural changes across its ecosystem. The rise of online travel agencies, review platforms, and digital accommodation services has altered competitive dynamics, shifting power toward consumers and digital intermediaries (Li et al., 2025). Digitalization has lowered entry barriers, facilitated price comparisons, revolutionized distribution channels, optimized costs, and enhanced productivity, thereby exposing tourism businesses to a more dynamic and competitive environment (Carlisle et al., 2023). Consequently, tools associated with Marketing 4.0 are essential for attracting customers, strengthening relationships, and promoting tourism products and services (Fernández Cueria et al., 2022). Despite these opportunities, the tourism industry faces significant challenges concerning managerial readiness and digital competencies. Tourism managers play a crucial role in strategic decision-making and operational performance, with digital skills increasingly recognized as vital for organizational competitiveness (Zuñiga-Collazos et al., 2025). Prior research indicates substantial gaps between current capabilities and future digital skill requirements in tourism, particularly in developing economies (Minor et al., 2025). In Iran, despite its considerable tourism potential, the industry remains underdeveloped, partly due to managerial deficiencies and limited adoption of Marketing 4.0 practices. Existing studies have primarily concentrated on digital marketing strategies rather than the specific competencies required of tourism managers. Therefore, this study aimed to develop a comprehensive model of Marketing 4.0 managerial competencies tailored to the tourism sector with a particular focus on platform-based tourism businesses, thereby addressing a critical gap in the literature and providing practical insights for emerging tourism markets. Materials & Methods This study utilized an applied purpose and adopted a qualitative approach through grounded theory methodology. The systematic framework proposed by Strauss and Corbin (1997) was employed, incorporating the paradigmatic model and the three stages of open, axial, and selective coding to develop theory from the collected data. Participants in the qualitative phase included academic experts, practitioners, and hybrid academics in marketing and tourism marketing, as well as tourism entrepreneurs and owners of small to medium-sized tourism businesses. A non-probability sampling strategy was implemented, relying on snowball sampling guided by predefined inclusion criteria specified in the interview protocol. These criteria included professional experience in tourism and tourism marketing, research experience with publications in reputable national or international journals, and authorship of academic books in related fields. Theoretical sampling was applied to determine sample size with data collection and analysis occurring concurrently until theoretical saturation was reached. After the 15th interview, the emergence of new codes ceased, confirming data saturation. Data were gathered through semi-structured interviews, which were developed based on a comprehensive literature review and consultations with experts. Each interview lasted approximately 70 minutes and was conducted over a 4-month period from February to May 2025. Data analysis was carried out by using MAXQDA software following the 3-stage coding process. To ensure the rigor and trustworthiness of the findings, credibility was established through member checking, peer debriefing, and external auditing. Inter-coder reliability was evaluated by using Cohen’s kappa coefficient, resulting in a value of 0.81, which indicated a high level of coding reliability. Minor discrepancies between coders were resolved through discussion and consensus. Finally, the validated codes and categories were synthesized to develop the comprehensive grounded theory model of Marketing 4.0 managerial competencies in the tourism industry. Research Findings The findings were organized within a paradigmatic model that encompassed causal conditions, central phenomenon, intervening conditions, strategies, contextual conditions, and consequences. Causal Conditions: These included both technological and organizational factors. Technological factors referred to the use of advanced digital technologies and adoption of digital marketing tools, which were essential prerequisites for developing Marketing 4.0 competencies. Organizational factors involved strategic management, organizational culture, and availability of resources—both organizational and financial—that enabled adaptability and facilitated effective digital marketing implementation. Central Phenomenon: This comprised 4 core competency dimensions. Ethical competencies emphasized a commitment to tourist privacy and sustainable marketing practices. Tourism marketing competencies encompassed specialized marketing knowledge, as well as an understanding of environmental and cultural contexts. Digital competencies involved digital literacy, data analysis skills, and media literacy, which supported data-driven and interactive marketing efforts. Leadership competencies included decision-making abilities, analytical skills, and the capacity for team leadership. Intervening Conditions: These reflected macro-institutional and macro-environmental factors, such as economic and political conditions, legal frameworks, technological advancements, and cultural-social dynamics. Strategies: Strategies were categorized into intra-organizational approaches—such as educational, innovative, and technological strategies—and networking strategies that were participatory and communicative in nature. Contextual Conditions: These involved the availability of technological and educational infrastructures, as well as a dynamic and competitive market environment. Consequences: Finally, the outcomes included market consequences, such as enhanced customer orientation and improved competitiveness, as well as performance outcomes like operational enhancement, financial growth, and improved service quality. Discussion of Results & Conclusion The findings revealed that Marketing 4.0 managerial competencies in tourism extended beyond technical digital skills, necessitating a holistic integration of ethical, marketing, digital, and leadership capabilities. Unlike prior studies that predominantly focused on digital tools or isolated skill sets, this research conceptualized managerial competency as a dynamic system influenced by organizational, technological, and environmental factors. The results emphasized the critical role of digital transformation and strategic leadership in enabling tourism organizations to effectively navigate uncertainty, intense competition, and evolving consumer behaviors. The proposed model enriches the literature by contextualizing Marketing 4.0 competencies within the unique characteristics of the tourism industry in developing economies. It illustrates that without adequate infrastructure, institutional support, and a commitment to continuous learning, individual managerial competencies alone are insufficient for achieving sustainable performance. Furthermore, the integration of ethical and sustainability-oriented competencies addresses growing concerns related to data privacy, environmental pressures, and responsible tourism practices. From a practical standpoint, the model offers a structured framework for tourism organizations and policymakers to design targeted training programs, invest in digital infrastructure, and foster collaborative networks. By enhancing managerial competencies, tourism businesses can improve service quality, elevate customer experiences, and strengthen their competitive positioning at both regional and international levels. Despite its contributions, this study was limited by its qualitative nature and specific industry focus. Future research is encouraged to quantitatively validate the proposed model by using statistical methods and explore its applicability across various tourism sub-sectors and other industries. Objective: The emergence of fourth-generation technologies has profoundly transformed the tourism industry. As such, equipping marketing managers with the competencies essential for this new era is critical. This study aimed to develop a competency model for fourth-generation marketing managers within the tourism sector. Methodology: This research was applied in nature and qualitative in approach, employing grounded theory methodology. The study involved 15 experts in digital marketing specifically within the tourism industry. The sampling process utilized a purposeful and snowball approach, continuing until theoretical saturation was reached. Data were collected through semi-structured interviews, with the validity and reliability of the tool being confirmed. Data analysis was conducted using a 3-stage coding process—open, axial, and selective coding—facilitated by Maxqda software. Findings: The analysis revealed a total of 154 open codes, 34 subcategories, and 14 main categories derived from the interviews. These findings were organized into a paradigm model consisting of 6 axes: 1) Causal conditions (technological and organizational factors), 2) Contextual conditions (infrastructure and market environment), 3) Central category (ethical competence, tourism marketing, leadership, and digital skills), 4) Intervening conditions (macro-institutional and macro-environmental factors), 5) Strategies (intra-organizational and networking), and 6) Consequences (market impacts and performance outcomes). Conclusion: The insights gleaned from this study offer valuable recommendations for both managers and researchers focused on enhancing the competencies of marketing managers in the tourism industry, particularly in the context of fourth-generation marketing. Introduction Recent advancements in digital transformation, connectivity, and information accessibility have fundamentally reshaped the marketing landscape. In response, scholars have introduced Marketing 4.0 as a new paradigm that shifts marketing strategies from traditional tools to digitally driven and interactive approaches focused on building long-term customer relationships (Dash et al., 2021). Marketing 4.0 seamlessly integrates online and offline interactions within the digital economy, requiring organizations to remain flexible and adaptive to rapid technological changes (Kotler et al., 2017). These transformations have significantly impacted the tourism industry, leading to structural changes across its ecosystem. The rise of online travel agencies, review platforms, and digital accommodation services has altered competitive dynamics, shifting power toward consumers and digital intermediaries (Li et al., 2025). Digitalization has lowered entry barriers, facilitated price comparisons, revolutionized distribution channels, optimized costs, and enhanced productivity, thereby exposing tourism businesses to a more dynamic and competitive environment (Carlisle et al., 2023). Consequently, tools associated with Marketing 4.0 are essential for attracting customers, strengthening relationships, and promoting tourism products and services (Fernández Cueria et al., 2022). Despite these opportunities, the tourism industry faces significant challenges concerning managerial readiness and digital competencies. Tourism managers play a crucial role in strategic decision-making and operational performance, with digital skills increasingly recognized as vital for organizational competitiveness (Zuñiga-Collazos et al., 2025). Prior research indicates substantial gaps between current capabilities and future digital skill requirements in tourism, particularly in developing economies (Minor et al., 2025). In Iran, despite its considerable tourism potential, the industry remains underdeveloped, partly due to managerial deficiencies and limited adoption of Marketing 4.0 practices. Existing studies have primarily concentrated on digital marketing strategies rather than the specific competencies required of tourism managers. Therefore, this study aimed to develop a comprehensive model of Marketing 4.0 managerial competencies tailored to the tourism sector with a particular focus on platform-based tourism businesses, thereby addressing a critical gap in the literature and providing practical insights for emerging tourism markets. Materials & Methods This study utilized an applied purpose and adopted a qualitative approach through grounded theory methodology. The systematic framework proposed by Strauss and Corbin (1997) was employed, incorporating the paradigmatic model and the three stages of open, axial, and selective coding to develop theory from the collected data. Participants in the qualitative phase included academic experts, practitioners, and hybrid academics in marketing and tourism marketing, as well as tourism entrepreneurs and owners of small to medium-sized tourism businesses. A non-probability sampling strategy was implemented, relying on snowball sampling guided by predefined inclusion criteria specified in the interview protocol. These criteria included professional experience in tourism and tourism marketing, research experience with publications in reputable national or international journals, and authorship of academic books in related fields. Theoretical sampling was applied to determine sample size with data collection and analysis occurring concurrently until theoretical saturation was reached. After the 15th interview, the emergence of new codes ceased, confirming data saturation. Data were gathered through semi-structured interviews, which were developed based on a comprehensive literature review and consultations with experts. Each interview lasted approximately 70 minutes and was conducted over a 4-month period from February to May 2025. Data analysis was carried out by using MAXQDA software following the 3-stage coding process. To ensure the rigor and trustworthiness of the findings, credibility was established through member checking, peer debriefing, and external auditing. Inter-coder reliability was evaluated by using Cohen’s kappa coefficient, resulting in a value of 0.81, which indicated a high level of coding reliability. Minor discrepancies between coders were resolved through discussion and consensus. Finally, the validated codes and categories were synthesized to develop the comprehensive grounded theory model of Marketing 4.0 managerial competencies in the tourism industry. Research Findings The findings were organized within a paradigmatic model that encompassed causal conditions, central phenomenon, intervening conditions, strategies, contextual conditions, and consequences. Causal Conditions: These included both technological and organizational factors. Technological factors referred to the use of advanced digital technologies and adoption of digital marketing tools, which were essential prerequisites for developing Marketing 4.0 competencies. Organizational factors involved strategic management, organizational culture, and availability of resources—both organizational and financial—that enabled adaptability and facilitated effective digital marketing implementation. Central Phenomenon: This comprised 4 core competency dimensions. Ethical competencies emphasized a commitment to tourist privacy and sustainable marketing practices. Tourism marketing competencies encompassed specialized marketing knowledge, as well as an understanding of environmental and cultural contexts. Digital competencies involved digital literacy, data analysis skills, and media literacy, which supported data-driven and interactive marketing efforts. Leadership competencies included decision-making abilities, analytical skills, and the capacity for team leadership. Intervening Conditions: These reflected macro-institutional and macro-environmental factors, such as economic and political conditions, legal frameworks, technological advancements, and cultural-social dynamics. Strategies: Strategies were categorized into intra-organizational approaches—such as educational, innovative, and technological strategies—and networking strategies that were participatory and communicative in nature. Contextual Conditions: These involved the availability of technological and educational infrastructures, as well as a dynamic and competitive market environment. Consequences: Finally, the outcomes included market consequences, such as enhanced customer orientation and improved competitiveness, as well as performance outcomes like operational enhancement, financial growth, and improved service quality. Discussion of Results & Conclusion The findings revealed that Marketing 4.0 managerial competencies in tourism extended beyond technical digital skills, necessitating a holistic integration of ethical, marketing, digital, and leadership capabilities. Unlike prior studies that predominantly focused on digital tools or isolated skill sets, this research conceptualized managerial competency as a dynamic system influenced by organizational, technological, and environmental factors. The results emphasized the critical role of digital transformation and strategic leadership in enabling tourism organizations to effectively navigate uncertainty, intense competition, and evolving consumer behaviors. The proposed model enriches the literature by contextualizing Marketing 4.0 competencies within the unique characteristics of the tourism industry in developing economies. It illustrates that without adequate infrastructure, institutional support, and a commitment to continuous learning, individual managerial competencies alone are insufficient for achieving sustainable performance. Furthermore, the integration of ethical and sustainability-oriented competencies addresses growing concerns related to data privacy, environmental pressures, and responsible tourism practices. From a practical standpoint, the model offers a structured framework for tourism organizations and policymakers to design targeted training programs, invest in digital infrastructure, and foster collaborative networks. By enhancing managerial competencies, tourism businesses can improve service quality, elevate customer experiences, and strengthen their competitive positioning at both regional and international levels. Despite its contributions, this study was limited by its qualitative nature and specific industry focus. Future research is encouraged to quantitatively validate the proposed model by using statistical methods and explore its applicability across various tourism sub-sectors and other industries.

https://nmrj.ui.ac.ir/article_30274_2611c03576c18d36489a1b239c0a2756.pdf

University of Isfahan New Marketing Research Journal 2228-7744 16 2 2026 06 22 Effects of Augmented Reality (AR) Advertising on Purchase Intention and Shared Social Experience: An Experience Economy Approach Effects of Augmented Reality (AR) Advertising on Purchase Intention and Shared Social Experience: An Experience Economy Approach 67 94 30403 10.22108/nmrj.2026.147788.3283 FA Mohsen Rajabzadeh Assistant professor, Department of Executive Management, Faculty of Economics and Administrative Sciences, University of Mazandaran, Babolsar, Iran 0000-0002-1089-455X Seyed Mehdi Khakzadian Assistant professor, Department of Industrial Management (Entrepreneurship), Faculty of Economic and Administrative Sciences, University of Mazandaran, Babolsar, Iran Fereshteh Rahchamandi Master's Degree in Information Technology Management, Faculty of Economics and Administrative Sciences, Ferdowsi University of Mashhad, Mashhad, Iran 0009-0000-5663-1379 Journal Article 2025 12 15 Nowadays, marketing has increasingly moved away from traditional approaches and become deeply integrated with emerging technologies. Among these technologies, Augmented Reality (AR) plays a pivotal role in enhancing brand–consumer interactions by creating interactive and immersive experiences, thereby improving the effectiveness of marketing activities. Drawing on experiential economy theory, this study aimed to examine the impact of AR-based advertising on shared social experience and purchase intention. This research was applied in terms of purpose and adopted a descriptive–survey methodology. The study population consisted of users who had been exposed to AR advertisements, from which a sample of 384 respondents was selected. Data were collected by using a standardized questionnaire, the validity and reliability of which were confirmed. Data analysis was conducted using SPSS and SmartPLS software. The findings indicated that the aesthetic dimension had a positive and significant effect on entertainment, education, and escapism. Furthermore, entertainment and escapism were found to have a positive and significant impact on satisfaction with AR advertising. However, escapism did not exert a significant effect on shared social experience. Finally, the results revealed that novel brand experience had a positive and significant influence on both shared social experience and purchase intention. Overall, the findings suggested that organizations could design more effective advertising campaigns by strategically employing AR advertising—not only to enhance attractiveness and entertainment, but also to deeply engage consumers' cognition. Such engagement facilitated the formation of electronic word-of-mouth and ultimately strengthened consumers' purchase intentions. Introduction Marketing management has consistently emphasized the importance of strengthening interactions between consumers and brands as active customer engagement enhances transactions, purchase behavior, and consumers' enthusiasm toward brands (Sung, 2021), while also fostering long-term relationships and sustainable competitive advantage (Kumar & Pansari, 2016; Venkatesan, 2017). In today's highly competitive markets, the emergence of modern marketing approaches has been significantly influenced by continuous technological development and innovation (Feiz et al., 2022). Consequently, organizations increasingly seek innovative technologies capable of creating richer and more immersive customer experiences. Among these technologies, Augmented Reality (AR) has emerged as one of the most transformative and disruptive innovations in contemporary marketing (Hackl & Wolfe, 2017). By integrating digital content into users' real-world environments, AR creates immersive and interactive experiences that enhance consumer engagement with brands (Flavián et al., 2019; Georgiou & Kyza, 2017; tom Dieck et al., 2018b). Moreover, rapid proliferation of smartphones and artificial intelligence technologies has further accelerated the adoption of AR in marketing and advertising contexts (Hackl & Wolfe, 2017). AR applications are increasingly utilized across various industries, including e‑commerce, advertising, education, entertainment, and retailing (Emadi Sadeghi et al., 2023). Major retailers, such as IKEA and Sephora, have adopted AR‑based applications that enable customers to virtually experience products before making a purchase. Beyond virtual try‑on features, AR can also enrich product packaging and storytelling by delivering immersive brand‑related content through interactive digital markers (Barta et al., 2025). Given the growing dominance of visual content and strong interest of Generation Z in emerging technologies, AR is expected to become deeply integrated into future marketing communications and promotional campaigns (Hilken et al., 2017; Rauschnabel, 2021; Rese et al., 2017). Despite increasing scholarly attention to AR marketing, previous studies have primarily focused on consumer motivations for using AR applications, AR quality, social communication, shopping behavior, and retail atmospherics (Hilken et al., 2017; Poncin & Mimoun, 2014; Scholz & Duffy, 2018; Yim et al., 2017). However, empirical evidence regarding customer responses to mobile AR advertising remains limited (Sung, 2021). Addressing this research gap, the present study investigated customer responses to mobile AR advertising applications through the lens of experiential economy and viral marketing approaches. Specifically, this research examined the extent to which satisfaction with AR advertising influenced consumers' purchase intention and shared social experience. The findings were expected to contribute to the literature on AR‑based advertising while providing practical insights for organizations seeking to implement AR‑driven mobile marketing campaigns and enhance competitive advantage through immersive customer experiences. Materials & Methods The present study was applied in nature and employed a descriptive-survey design with a correlational approach. In terms of data analysis, the study followed a quantitative methodology. The statistical population consisted of students at the University of Mazandaran. To enhance the accuracy and validity of the collected data, participants were first exposed to an AR advertising experience before completing the survey questionnaire. Specifically, respondents were provided with a link to a mobile AR advertisement related to sunglasses. They were then asked to customize the product by selecting their preferred face shape, frame type, lens design, and sunglasses style. Using the front camera of their smartphones, they were able to virtually visualize the selected sunglasses on their own faces through AR technology. Following this immersive advertising experience, respondents completed the research questionnaire. Given the unlimited size of the statistical population, the sample size was determined by using Morgan's sampling table for infinite populations, which yielded a sample of 384 respondents. Data were collected by using a non-probability convenience sampling method. The theoretical foundations and literature review were developed through a library research approach, while survey techniques were employed to collect participants' opinions and responses. The data collection instrument was a standardized questionnaire adopted from Sung et al. (2022). As the questionnaire had been previously validated, face validity was assessed through expert evaluation and the instrument was confirmed to possess acceptable validity. Construct validity was examined by using Confirmatory Factor Analysis (CFA) conducted with SmartPLS software. The results indicated that all factor loadings exceeded the acceptable threshold of 0.40 and all t‑values were greater than 1.96, confirming satisfactory construct validity. Convergent validity was also evaluated by using the Average Variance Extracted (AVE) criterion with all AVE values exceeding 0.40, indicating adequate convergent validity. To assess reliability, Cronbach's alpha and composite reliability coefficients were calculated using SPSS 18 based on a pilot sample of 30 respondents. The findings confirmed satisfactory reliability for all constructs. Furthermore, discriminant validity was assessed by using cross‑loadings and the Fornell–Larcker criterion, both of which supported the adequacy of discriminant validity. The Kolmogorov–Smirnov (K‑S) test results indicated that the research data were not normally distributed. Finally, confirmatory factor analysis was employed to evaluate the relationship between the questionnaire items and their corresponding constructs, while the coefficients of determination (R²) demonstrated the explanatory power of the latent variables in the proposed research model. Research Findings The descriptive findings revealed that among the 384 respondents, 128 were female and 256 were male. Regarding age distribution, the majority of respondents (283 individuals) were between 18 and 23 years old followed by 28 participants aged 24–29, 18 aged 30–35, 30 aged 36–40, 21 aged 41–50, and 4 respondents over 50 years old. In terms of educational background, 43 respondents held a high school diploma, 6 had an associate degree, 270 possessed a bachelor’s degree, and 65 held a master’s degree. After confirming the adequacy of the measurement model through confirmatory factor analysis, the structural model and the proposed hypotheses were examined by using Structural Equation Modeling (SEM). The results demonstrated that aesthetics had a significant positive effect on entertainment (β=0.609, t=16.523, p<0.001), education (β=0.439, t=10.613, p<0.001), and escapism (β=0.517, t=13.673, p<0.001). Furthermore, entertainment positively influenced customer satisfaction (β=0.214, t=5.578, p<0.001), and education also exerted a significant positive effect on satisfaction (β=0.118, t=3.312, p=0.001). Among the dimensions of the experience economy, escapism demonstrated the strongest impact on satisfaction (β=0.407, t=13.518, p<0.001). The findings further indicated that escapism did not have a significant effect on shared social experience (β=-0.039, t=0.744, p=0.457), leading to the rejection of this hypothesis. However, customer satisfaction had a strong positive influence on both shared social experience (β=0.832, t=17.029, p<0.001) and purchase intention (β=0.743, t=25.775, p<0.001). Additionally, authenticity significantly affected the creation of a new brand experience (β=0.725, t=28.872, p<0.001). The mediating role of satisfaction was also confirmed. The results demonstrated that a new brand experience indirectly influenced shared social experience through satisfaction (β=0.280, t=8.519, p<0.001). Similarly, satisfaction significantly mediated the relationship between new brand experience and purchase intention (β=0.250, t=10.772, p<0.001). Overall, the findings highlighted the critical role of immersive AR advertising experiences in shaping customer satisfaction, enhancing social interaction, and strengthening consumers' purchase intentions. Discussion of Results & Conclusion Augmented Reality (AR) has emerged as an advanced and transformative technology increasingly utilized across various domains to enhance user experiences (Lee et al., 2025). Among these domains, marketing and advertising have attracted considerable attention due to the ability of AR to create immersive and interactive consumer experiences. Interactive AR advertising represents a modern marketing approach capable of increasing advertising effectiveness and stimulating positive consumer responses (Sung et al., 2022). The findings of the present study demonstrated that mobile AR advertising significantly influenced consumers' purchase intention and their willingness to share experiences through social networks. Drawing on the experience economy framework, the study revealed that experiential dimensions—namely aesthetics, entertainment, education, and escapism—along with satisfaction derived from AR advertising, positively shaped consumers' behavioral intentions. The findings confirmed that aesthetics played a crucial role in enhancing the attractiveness of AR advertising and significantly contributed to entertainment, education, and escapism experiences. Consistent with prior studies (Shukla et al., 2024; Sung, 2021; tom Dieck et al., 2018b), aesthetically appealing AR environments increased user satisfaction by creating engaging and immersive interactions. In particular, entertainment and educational features were found to positively influence customer satisfaction. AR advertising not only provided enjoyable and novel experiences, but also facilitated product-related learning and customization in an interactive manner. This educational aspect stimulated consumers' curiosity and improved their understanding of products and brands, thereby increasing satisfaction and strengthening purchase intention (Hosany & Witham, 2010; Yang et al., 2020). The results further indicated that escapism positively affected satisfaction by enabling users to immerse themselves in the AR environment and temporarily escape everyday routines. Such immersive experiences enhanced emotional and cognitive engagement, intensified the sense of presence, and created enjoyable interactions with the advertised product (Chang & Suh, 2025; Tsita et al., 2023). However, contrary to the findings of Sung (2021), escapism did not significantly influence shared social experience. This inconsistency might be attributed to cultural and contextual factors, particularly within the Iranian social media environment, where social platforms were often presentation‑oriented and internet restrictions might reduce users' motivation to share immersive experiences online. Consequently, escapism might function more as an individual emotional experience rather than a socially interactive one (Soraci et al., 2025). Another important finding was the positive effect of authenticity on creating a new brand experience. Because AR advertising integrated digital elements into real-world settings, consumers tended to perceive AR content as more authentic, credible, and immersive (Sung, 2021). Furthermore, the study confirmed the mediating role of satisfaction in the relationship between new brand experience, shared social experience, and purchase intention. Immersive AR experiences enhanced perceived value, strengthened brand attitudes, reduced perceived purchase risk, and encouraged consumers to actively engage with and share brand-related experiences (Yang & Lin, 2024). From a managerial perspective, the findings suggested that businesses should design AR advertising campaigns that emphasize interactivity, personalization, aesthetics, entertainment, and educational value. Features like customizable product options, gamification elements, and simple interactive scenarios can significantly improve user satisfaction, purchase intention, and social sharing behavior (Huang & Liao, 2015; Hollebeek & Macky, 2019). Managers are also encouraged to facilitate easy sharing mechanisms through social media platforms to maximize the viral potential of AR campaigns. However, given the non-significant relationship between escapism and shared social experience, marketers should avoid overemphasizing escapist dimensions and instead focus on practical engagement and meaningful social interaction. Despite its contributions, this study had several limitations. The measurement of escapism might not fully capture the complexity of the construct and the use of a relatively homogeneous sample consisting solely of university students limited generalizability of the findings. Additionally, the self-reported nature of the data might introduce response bias. Future studies are encouraged to investigate moderating variables, such as personality traits, social anxiety, and motivations, for AR usage (McLean & Wilson, 2019). Moreover, qualitative approaches and broader demographic samples can provide deeper insights into users' perceptions of AR advertising and its influence on social interaction and purchasing behavior. Nowadays, marketing has increasingly moved away from traditional approaches and become deeply integrated with emerging technologies. Among these technologies, Augmented Reality (AR) plays a pivotal role in enhancing brand–consumer interactions by creating interactive and immersive experiences, thereby improving the effectiveness of marketing activities. Drawing on experiential economy theory, this study aimed to examine the impact of AR-based advertising on shared social experience and purchase intention. This research was applied in terms of purpose and adopted a descriptive–survey methodology. The study population consisted of users who had been exposed to AR advertisements, from which a sample of 384 respondents was selected. Data were collected by using a standardized questionnaire, the validity and reliability of which were confirmed. Data analysis was conducted using SPSS and SmartPLS software. The findings indicated that the aesthetic dimension had a positive and significant effect on entertainment, education, and escapism. Furthermore, entertainment and escapism were found to have a positive and significant impact on satisfaction with AR advertising. However, escapism did not exert a significant effect on shared social experience. Finally, the results revealed that novel brand experience had a positive and significant influence on both shared social experience and purchase intention. Overall, the findings suggested that organizations could design more effective advertising campaigns by strategically employing AR advertising—not only to enhance attractiveness and entertainment, but also to deeply engage consumers' cognition. Such engagement facilitated the formation of electronic word-of-mouth and ultimately strengthened consumers' purchase intentions. Introduction Marketing management has consistently emphasized the importance of strengthening interactions between consumers and brands as active customer engagement enhances transactions, purchase behavior, and consumers' enthusiasm toward brands (Sung, 2021), while also fostering long-term relationships and sustainable competitive advantage (Kumar & Pansari, 2016; Venkatesan, 2017). In today's highly competitive markets, the emergence of modern marketing approaches has been significantly influenced by continuous technological development and innovation (Feiz et al., 2022). Consequently, organizations increasingly seek innovative technologies capable of creating richer and more immersive customer experiences. Among these technologies, Augmented Reality (AR) has emerged as one of the most transformative and disruptive innovations in contemporary marketing (Hackl & Wolfe, 2017). By integrating digital content into users' real-world environments, AR creates immersive and interactive experiences that enhance consumer engagement with brands (Flavián et al., 2019; Georgiou & Kyza, 2017; tom Dieck et al., 2018b). Moreover, rapid proliferation of smartphones and artificial intelligence technologies has further accelerated the adoption of AR in marketing and advertising contexts (Hackl & Wolfe, 2017). AR applications are increasingly utilized across various industries, including e‑commerce, advertising, education, entertainment, and retailing (Emadi Sadeghi et al., 2023). Major retailers, such as IKEA and Sephora, have adopted AR‑based applications that enable customers to virtually experience products before making a purchase. Beyond virtual try‑on features, AR can also enrich product packaging and storytelling by delivering immersive brand‑related content through interactive digital markers (Barta et al., 2025). Given the growing dominance of visual content and strong interest of Generation Z in emerging technologies, AR is expected to become deeply integrated into future marketing communications and promotional campaigns (Hilken et al., 2017; Rauschnabel, 2021; Rese et al., 2017). Despite increasing scholarly attention to AR marketing, previous studies have primarily focused on consumer motivations for using AR applications, AR quality, social communication, shopping behavior, and retail atmospherics (Hilken et al., 2017; Poncin & Mimoun, 2014; Scholz & Duffy, 2018; Yim et al., 2017). However, empirical evidence regarding customer responses to mobile AR advertising remains limited (Sung, 2021). Addressing this research gap, the present study investigated customer responses to mobile AR advertising applications through the lens of experiential economy and viral marketing approaches. Specifically, this research examined the extent to which satisfaction with AR advertising influenced consumers' purchase intention and shared social experience. The findings were expected to contribute to the literature on AR‑based advertising while providing practical insights for organizations seeking to implement AR‑driven mobile marketing campaigns and enhance competitive advantage through immersive customer experiences. Materials & Methods The present study was applied in nature and employed a descriptive-survey design with a correlational approach. In terms of data analysis, the study followed a quantitative methodology. The statistical population consisted of students at the University of Mazandaran. To enhance the accuracy and validity of the collected data, participants were first exposed to an AR advertising experience before completing the survey questionnaire. Specifically, respondents were provided with a link to a mobile AR advertisement related to sunglasses. They were then asked to customize the product by selecting their preferred face shape, frame type, lens design, and sunglasses style. Using the front camera of their smartphones, they were able to virtually visualize the selected sunglasses on their own faces through AR technology. Following this immersive advertising experience, respondents completed the research questionnaire. Given the unlimited size of the statistical population, the sample size was determined by using Morgan's sampling table for infinite populations, which yielded a sample of 384 respondents. Data were collected by using a non-probability convenience sampling method. The theoretical foundations and literature review were developed through a library research approach, while survey techniques were employed to collect participants' opinions and responses. The data collection instrument was a standardized questionnaire adopted from Sung et al. (2022). As the questionnaire had been previously validated, face validity was assessed through expert evaluation and the instrument was confirmed to possess acceptable validity. Construct validity was examined by using Confirmatory Factor Analysis (CFA) conducted with SmartPLS software. The results indicated that all factor loadings exceeded the acceptable threshold of 0.40 and all t‑values were greater than 1.96, confirming satisfactory construct validity. Convergent validity was also evaluated by using the Average Variance Extracted (AVE) criterion with all AVE values exceeding 0.40, indicating adequate convergent validity. To assess reliability, Cronbach's alpha and composite reliability coefficients were calculated using SPSS 18 based on a pilot sample of 30 respondents. The findings confirmed satisfactory reliability for all constructs. Furthermore, discriminant validity was assessed by using cross‑loadings and the Fornell–Larcker criterion, both of which supported the adequacy of discriminant validity. The Kolmogorov–Smirnov (K‑S) test results indicated that the research data were not normally distributed. Finally, confirmatory factor analysis was employed to evaluate the relationship between the questionnaire items and their corresponding constructs, while the coefficients of determination (R²) demonstrated the explanatory power of the latent variables in the proposed research model. Research Findings The descriptive findings revealed that among the 384 respondents, 128 were female and 256 were male. Regarding age distribution, the majority of respondents (283 individuals) were between 18 and 23 years old followed by 28 participants aged 24–29, 18 aged 30–35, 30 aged 36–40, 21 aged 41–50, and 4 respondents over 50 years old. In terms of educational background, 43 respondents held a high school diploma, 6 had an associate degree, 270 possessed a bachelor’s degree, and 65 held a master’s degree. After confirming the adequacy of the measurement model through confirmatory factor analysis, the structural model and the proposed hypotheses were examined by using Structural Equation Modeling (SEM). The results demonstrated that aesthetics had a significant positive effect on entertainment (β=0.609, t=16.523, p<0.001), education (β=0.439, t=10.613, p<0.001), and escapism (β=0.517, t=13.673, p<0.001). Furthermore, entertainment positively influenced customer satisfaction (β=0.214, t=5.578, p<0.001), and education also exerted a significant positive effect on satisfaction (β=0.118, t=3.312, p=0.001). Among the dimensions of the experience economy, escapism demonstrated the strongest impact on satisfaction (β=0.407, t=13.518, p<0.001). The findings further indicated that escapism did not have a significant effect on shared social experience (β=-0.039, t=0.744, p=0.457), leading to the rejection of this hypothesis. However, customer satisfaction had a strong positive influence on both shared social experience (β=0.832, t=17.029, p<0.001) and purchase intention (β=0.743, t=25.775, p<0.001). Additionally, authenticity significantly affected the creation of a new brand experience (β=0.725, t=28.872, p<0.001). The mediating role of satisfaction was also confirmed. The results demonstrated that a new brand experience indirectly influenced shared social experience through satisfaction (β=0.280, t=8.519, p<0.001). Similarly, satisfaction significantly mediated the relationship between new brand experience and purchase intention (β=0.250, t=10.772, p<0.001). Overall, the findings highlighted the critical role of immersive AR advertising experiences in shaping customer satisfaction, enhancing social interaction, and strengthening consumers' purchase intentions. Discussion of Results & Conclusion Augmented Reality (AR) has emerged as an advanced and transformative technology increasingly utilized across various domains to enhance user experiences (Lee et al., 2025). Among these domains, marketing and advertising have attracted considerable attention due to the ability of AR to create immersive and interactive consumer experiences. Interactive AR advertising represents a modern marketing approach capable of increasing advertising effectiveness and stimulating positive consumer responses (Sung et al., 2022). The findings of the present study demonstrated that mobile AR advertising significantly influenced consumers' purchase intention and their willingness to share experiences through social networks. Drawing on the experience economy framework, the study revealed that experiential dimensions—namely aesthetics, entertainment, education, and escapism—along with satisfaction derived from AR advertising, positively shaped consumers' behavioral intentions. The findings confirmed that aesthetics played a crucial role in enhancing the attractiveness of AR advertising and significantly contributed to entertainment, education, and escapism experiences. Consistent with prior studies (Shukla et al., 2024; Sung, 2021; tom Dieck et al., 2018b), aesthetically appealing AR environments increased user satisfaction by creating engaging and immersive interactions. In particular, entertainment and educational features were found to positively influence customer satisfaction. AR advertising not only provided enjoyable and novel experiences, but also facilitated product-related learning and customization in an interactive manner. This educational aspect stimulated consumers' curiosity and improved their understanding of products and brands, thereby increasing satisfaction and strengthening purchase intention (Hosany & Witham, 2010; Yang et al., 2020). The results further indicated that escapism positively affected satisfaction by enabling users to immerse themselves in the AR environment and temporarily escape everyday routines. Such immersive experiences enhanced emotional and cognitive engagement, intensified the sense of presence, and created enjoyable interactions with the advertised product (Chang & Suh, 2025; Tsita et al., 2023). However, contrary to the findings of Sung (2021), escapism did not significantly influence shared social experience. This inconsistency might be attributed to cultural and contextual factors, particularly within the Iranian social media environment, where social platforms were often presentation‑oriented and internet restrictions might reduce users' motivation to share immersive experiences online. Consequently, escapism might function more as an individual emotional experience rather than a socially interactive one (Soraci et al., 2025). Another important finding was the positive effect of authenticity on creating a new brand experience. Because AR advertising integrated digital elements into real-world settings, consumers tended to perceive AR content as more authentic, credible, and immersive (Sung, 2021). Furthermore, the study confirmed the mediating role of satisfaction in the relationship between new brand experience, shared social experience, and purchase intention. Immersive AR experiences enhanced perceived value, strengthened brand attitudes, reduced perceived purchase risk, and encouraged consumers to actively engage with and share brand-related experiences (Yang & Lin, 2024). From a managerial perspective, the findings suggested that businesses should design AR advertising campaigns that emphasize interactivity, personalization, aesthetics, entertainment, and educational value. Features like customizable product options, gamification elements, and simple interactive scenarios can significantly improve user satisfaction, purchase intention, and social sharing behavior (Huang & Liao, 2015; Hollebeek & Macky, 2019). Managers are also encouraged to facilitate easy sharing mechanisms through social media platforms to maximize the viral potential of AR campaigns. However, given the non-significant relationship between escapism and shared social experience, marketers should avoid overemphasizing escapist dimensions and instead focus on practical engagement and meaningful social interaction. Despite its contributions, this study had several limitations. The measurement of escapism might not fully capture the complexity of the construct and the use of a relatively homogeneous sample consisting solely of university students limited generalizability of the findings. Additionally, the self-reported nature of the data might introduce response bias. Future studies are encouraged to investigate moderating variables, such as personality traits, social anxiety, and motivations, for AR usage (McLean & Wilson, 2019). Moreover, qualitative approaches and broader demographic samples can provide deeper insights into users' perceptions of AR advertising and its influence on social interaction and purchasing behavior.

https://nmrj.ui.ac.ir/article_30403_2c1d4eb775fb40ad57f7ea51ec702244.pdf

University of Isfahan New Marketing Research Journal 2228-7744 16 2 2026 06 22 Predicting Customer Satisfaction in Online Business-to-Business (B2B) Wholesaler: An Ensemble with Genetic Algorithm (GA) Tuning and SHAP Explainability Predicting Customer Satisfaction in Online Business-to-Business (B2B) Wholesaler: An Ensemble with Genetic Algorithm (GA) Tuning and SHAP Explainability 95 124 30500 10.22108/nmrj.2026.144327.3159 FA Amir Hosein Esmaielpour Ph.D. student, Department of Industrial Engineering, Faculty of Engineering, Kharazmi University, Tehran, Iran Mariam Ameli Assistant professor, Department of Industrial Engineering, Faculty of Engineering, Kharazmi University, Tehran, Iran Ehsan Golchin M.Sc. student, Department of Business Management, Faculty of Management, Kharazmi University, Tehran, Iran Journal Article 2025 02 11 The rapid growth of Business-to-Business (B2B) e-commerce platforms necessitates robust, data-driven frameworks for anticipating and managing customer experiences. This study addressed the challenge of predicting customer satisfaction within an online industrial B2B wholesaler in Iran by proposing an optimized machine learning ensemble approach. Using an operational dataset comprising 349 distinct orders, we formulated the prediction task as a multi-class ordinal classification problem with 3 categories: Satisfied, Neutral, and Dissatisfied. A majority-voting ensemble architecture was developed, integrating 3 diverse base learners—CatBoost, Random Forest (RF), and a Multi-Layer Perceptron (MLP) neural network. To maximize predictive accuracy, the hyperparameters of each base algorithm were independently tuned by using a Genetic Algorithm (GA). Empirical evaluation on a held-out validation set showed that the proposed GA-tuned majority-voting ensemble achieved outstanding performance with an Accuracy of 0.9575, Recall of 0.9476, F1-score of 0.9379, and Matthews Correlation Coefficient (MCC) of 0.8966. Moreover, an equal-weight TOPSIS evaluation across multiple metrics ranked the ensemble model first among all alternative configurations. To address the "black-box" nature of the ensemble, a SHAP explainability analysis was conducted, revealing that operational parameters—such as Days Since Last Purchase and indicators of basket volume and weight—served as the primary drivers of satisfaction, while demographic attributes and purely financial variables exhibited negligible predictive power. This scalable pipeline provides actionable strategic insights for wholesale managers to implement proactive customer retention interventions. Introduction In the contemporary digital commerce paradigm, understanding and anticipating customer satisfaction is widely recognized as a cornerstone of corporate longevity, market sustainability, and effective customer relationship management. Within the Business-to-Business (B2B) e-commerce sector, proliferation of digital storefronts and wholesale platforms has dramatically intensified market competition. Unlike Business-to-Consumer (B2C) dynamics, B2B e-commerce involves highly complex operational pipelines characterized by large order volumes, high financial stakes, specialized transaction conditions, and extended procurement cycles. Consequently, the quality of the customer experience delivered by a wholesaler extends far beyond commodity pricing, encompassing multifaceted operational dimensions, such as delivery timeliness, packaging durability, shipping accuracy, logistical transparency, and after-sales service responsiveness. Despite the critical importance of these factors, traditional B2B operations often rely on retrospective feedback mechanisms—such as periodic phone surveys or post-delivery online ratings—to gauge client sentiment. These reactive approaches inherently limit the ability of an organization to implement timely, mitigating interventions before client dissatisfaction escalates into churn. To address this limitation, contemporary academic literature has shifted from theoretical, post-hoc explanations of customer loyalty toward proactive, data-driven predictive modeling enabled by machine learning and deep learning algorithms. By leveraging real-time transactional, behavioral, and demographic features, predictive pipelines can identify vulnerable client accounts before formal grievances are recorded. However, deploying predictive models within emerging markets, such as Iran, introduces distinct macro-environmental complexities. Wholesale enterprises operating in Iran contend with macroeconomic volatility, supply chain disruptions, fluctuating regulatory frameworks, and unique localized purchasing behaviors. In this context, predicting customer sentiment requires robust analytical models capable of extracting subtle behavioral patterns from relatively small, heterogeneous datasets without succumbing to overfitting. Although sophisticated deep learning architectures typically demand large datasets for effective generalization, machine learning ensembles and intelligent optimization algorithms—such as Genetic Algorithms (GAs)—offer a mathematically rigorous and computationally efficient alternative for modeling localized operational realities. This research directly addressed a notable gap in the existing literature: the absence of tailored, explainable machine learning frameworks for predicting multi-class customer satisfaction in small-sample B2B online wholesale environments operating under economic constraints. The primary objective of this study was to construct, optimize, and evaluate a comprehensive machine learning architecture capable of predicting customer satisfaction prior to the receipt of formal feedback. By validating this model on operational data from an industrial B2B wholesaler in Iran, this study established a portable framework that balanced high predictive accuracy with clinical interpretability. In doing so, it would equip corporate decision-makers with the analytical infrastructure needed to optimize resource allocation, enhance logistical workflows, and systematically foster long-term B2B client loyalty. Materials & Methods The methodological workflow of this study followed a highly structured and scientifically rigorous pipeline comprising data acquisition, preprocessing, architectural design of the ensemble model, evolutionary hyperparameter optimization, and multi-criteria evaluation. The empirical foundation of this research was built upon an operational dataset of 349 unique customer orders extracted from the Enterprise Resource Planning (ERP) and Customer Relationship Management (CRM) databases of Arman Roshan Fartak, a prominent industrial wholesaler specializing in industrial adhesives in Iran. The target variable representing customer satisfaction was meticulously collected through a synchronized dual-channel feedback mechanism that integrated systematic online post-purchase questionnaires with follow-up telephonic surveys conducted by specialized account managers. To align with the internal evaluation metrics of organizations, the target variable was encoded as a multi-class ordinal variable with 3 distinct levels: Class 0 ("Satisfied" clients), Class 1 ("Neutral" clients), and Class 2 ("Dissatisfied" clients). The predictive feature space encompassed 3 primary dimensions: procurement demographics, purchase volume and value indicators, and relational interaction signals. Specifically, the features included: the gender of the corporate purchasing officer, the age of the purchasing officer, the geographical location (categorically binned into metropolitan hubs versus regional municipalities), the cumulative financial expenditure incurred by the client up to the point of the current transaction ("Total Spend"), the aggregate physical weight of all products purchased ("KG"), the commission structure applied to the sales agent ("Commission"), the exact number of days elapsed since the client's immediately preceding transaction ("Days Since Last Purchase"), and the corporate legal status of the client enterprise (private sector versus public or governmental entity). The data preprocessing phase was executed with utmost care to ensure data integrity and prevent any form of data leakage. Of the initial 349 records, a small subset of 3 records (approximately 0.86%) exhibiting systematic and unrecoverable omissions in critical operational fields was entirely removed from the sample, yielding a final analytical dataset of 346 records. Categorical variables were transformed by using standard label encoding and one-hot encoding frameworks. To accommodate the specific structural requirements of the neural network component, continuous numerical variables were standardized by using a StandardScaler protocol, mapping the data to a mean of zero and a standard deviation of 1, thereby eliminating scaling biases. To establish a rigorous defense against overfitting and ensure robust generalization given the limited sample size, all subsequent training and optimization steps were embedded within a stratified 5-fold cross-validation framework, maintaining exact class proportions across all validation folds. The core predictive engine consisted of a heterogeneous majority-voting ensemble architecture designed to leverage the complementary mathematical strengths of 3 distinct base learners: CatBoost, Random Forest (RF), and a Multi-Layer Perceptron (MLP). CatBoost, a symmetric gradient-boosting tree framework, was selected for its inherent mathematical superiority in processing heterogeneous tabular data and its algorithmic defenses against target leakage. Random Forest was incorporated as a robust, non-linear bagging baseline known for its low variance and resilience to localized noise. The MLP neural network was integrated to capture high-dimensional, non-linear interactions within the feature space. Rather than relying on computationally blind hyperparameter selection methods, such as Grid Search or Random Search, this study deployed an evolutionary GA to independently discover the optimal hyperparameter configurations for each base learner. The GA optimization employed a population size of 20 chromosomes over a maximum of 40 generations with a tournament selection size of 3, a crossover probability of 0.85, a mutation probability of 0.10, and an early stopping criterion set to 10 consecutive generations without fitness improvement. The fitness function guiding the evolutionary search was defined as the stratified cross-validated weighted F1-score. To aggregate the individual base model outputs into a single final prediction, a hard voting (majority voting) protocol was chosen over soft voting following a rigorous preliminary sign test, which confirmed that hard voting offered superior mathematical stability and lower susceptibility to uncalibrated probability estimates given the sample size of this study. Research Findings The empirical results derived from the cross-validated training and testing phases demonstrated the substantial performance advantages achieved by integrating evolutionary optimization with an ensemble architecture. During the baseline evaluation of individual uncombined learners, CatBoost emerged as the most powerful standalone algorithm, exhibiting superior performance across nearly all primary evaluation metrics compared to RF and the MLP network. However, following the implementation of the hard-voting aggregation protocol, the finalized majority-voting ensemble delivered the most balanced, robust, and highly accurate performance, establishing its statistical dominance. On the held-out validation datasets, the proposed ensemble achieved an overall accuracy of 0.9575, a precision of 0.9396, a recall of 0.9476, and a weighted F1-score of 0.9379. To validate classification performance beyond simple accuracy, we computed advanced metrics, yielding a Matthews Correlation Coefficient (MCC) of 0.8966 and a Cohen's Kappa (CK) coefficient of 0.8957. These values confirmed that the model's predictive power was highly stable and entirely independent of any underlying class imbalances within the operational dataset. To mathematically validate the superiority of the ensemble approach without relying on subjective, single-metric interpretations, a Multi-Criteria Decision-Making (MCDM) evaluation was performed using the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS). The evaluation matrix integrated 6 core performance dimensions—Accuracy, Precision, Recall, F1-score, MCC, and CK—calculated across all models. To ensure completely unbiased results, an equal-weighting scheme was applied across all criteria. The calculated TOPSIS closeness coefficients (CiCi), which represented each model's Euclidean proximity to the mathematically ideal solution and distance from the anti-ideal solution, explicitly yielded the following ranking: Ensemble (C=0.9695C=0.9695, Rank 1) > CatBoost (C=0.9187C=0.9187, Rank 2) > Random Forest (C=0.3437C=0.3437, Rank 3) > Multi-Layer Perceptron (C=0.1037C=0.1037, Rank 4). This rigorous multi-criteria analysis confirmed that the ensemble architecture systematically minimized classification errors when evaluated across all operational performance indicators simultaneously. An in-depth analysis of the classification mechanics was conducted by inspecting the cross-validated confusion matrices for all four model variants. The ensemble architecture demonstrated an extraordinary capacity to identify vulnerable client segments, achieving a true positive count of 25 for Class 2 ("Dissatisfied")—meaning it attained a perfect 100% detection rate for dissatisfied accounts with zero false negatives in this critically important operational category. For Class 1 ("Neutral"), the ensemble maintained high precision, misclassifying only 2 out of 36 instances as "Satisfied". For Class 0 ("Satisfied"), the model exhibited a minimal margin of error with only a single instance incorrectly flagged as "Neutral". In contrast, the standalone MLP showed severe confusion, frequently failing to differentiate the subtle boundaries separating Class 1 and Class 2 instances. To unlock the "black-box" architecture of the optimized ensemble and extract actionable strategic insights, we applied SHAP (SHapley Additive exPlanations) values, which were grounded in cooperative game theory. The global feature importance hierarchy was established by calculating the mean absolute SHAP values averaged comprehensively across all three classification classes. The resulting SHAP summary plot revealed an extremely pronounced skew in predictive influence. The behavioral variable of "Days Since Last Purchase" emerged as the dominant driver of customer satisfaction predictions, achieving a mean ∣SHAP∣∣SHAP∣ value exceeding 1.6—a magnitude that mathematically quadrupled the influence of any other feature in the space. The volume and value metrics, namely "Total Spend" and the total physical weight of purchased goods ("KG"), exhibited moderate predictive influence acting as secondary drivers. Crucially, purely financial variables—such as the agent's "Commission"—along with all consumer demographic attributes—including the purchasing officer's age, gender, geographic city classification, and organizational company type—displayed negligible SHAP values, exerting almost no discernible impact on the model's ultimate classification decisions. Discussion of Results & Conclusion The empirical insights generated by this research carry both profound theoretical value for the domain of digital B2B marketing and immediate, actionable utility for operational management within wholesale enterprises. By demonstrating that a heterogeneous ensemble engine, optimized via evolutionary GAs, achieved an Accuracy of 95.75% and an MCC of 0.8966 on a compact operational dataset, this study successfully refuted the conventional assumption that sophisticated machine learning modeling is reserved exclusively for enterprises possessing massive, multi-million-record big data repositories. The perfect detection rate achieved by the ensemble for dissatisfied accounts (Class 2) held profound managerial significance: it proved that operational data natively generated during routine wholesale transactions contained highly structured, predictive behavioral signals that could be systematically exploited to deploy preemptive customer recovery strategies before relationship dissolution occurred. When contextualized within contemporary academic literature, the findings of this study provided both crucial validations and distinct points of divergence. In contrast to the prominent research conducted by Zaghloul et al. (2024), which utilized an immense dataset exceeding 100,000 e-commerce orders to predict customer satisfaction via standalone RF and Support Vector Machine (SVM) models—achieving an accuracy of approximately 92%—our framework demonstrated that an evolutionarily tuned ensemble architecture could extract significantly higher predictive accuracy (95.75%) and a vastly superior multi-criteria balance (validated via TOPSIS) from a small data footprint (n=346n=346). Furthermore, our SHAP-based feature importance hierarchy strongly reinforced the empirical conclusions of Tang (2025) and Hui et al. (2025), who argued that within B2B industrial environments, operational logistics signals, transaction recency, and purchase volume metrics carry exponentially greater predictive weight regarding customer sentiment than traditional demographic segmentation or basic pricing mechanisms. The extraordinary dominance of the feature of "Days Since Last Purchase" within our SHAP explainability pipeline underscored a vital behavioral reality in the Iranian industrial wholesale sector. In this context, an increasing temporal gap between orders did not merely signify a passive pause in purchasing; rather, owing to intense competition and macro-environmental supply disruptions, it served as an active, early-warning indicator of client alienation, low engagement, or silent switching to alternative suppliers. Conversely, the moderate significance of "KG" (physical weight) and "Total Spend" confirmed that clients who consolidated large, heavy industrial orders established deeper, more integrated operational ties with the wholesaler, rendering them highly sensitive to logistics execution quality, packaging reliability, and delivery schedules. The finding that demographics and commission structures exerted almost zero predictive influence sent a strong signal to wholesale executives: broad financial discounting policies or rigid demographic marketing campaigns were fundamentally ineffective levers for securing long-term loyalty. Instead, corporate interventions had to be aggressively redirected toward recency-based customer activation systems and the systematic optimization of fulfillment workflows. In conclusion, this study successfully established an optimized, highly accurate, and mathematically explainable machine learning pipeline for predicting customer satisfaction within the online B2B industrial wholesale sector. By fusing CatBoost, RF, and MLP models through a hard-voting ensemble, and optimizing their internal structures via a G A, the framework provided a highly reliable predictive tool that remained fully transparent through SHAP value interpretation. From a managerial perspective, the model offered an analytical blueprint for shifting corporate strategies from reactive grievance resolution to proactive client retention, prioritizing interaction recency and fulfillment reliability over superficial financial adjustments. However, certain structural limitations must be acknowledged. This study relied on a compact, single-organization dataset from an industrial setting in Iran, which might limit the direct generalizability of the exact parameter values to substantially different economic sectors. Additionally, the classification pipeline employed a standard multi-class formulation without deploying specialized ordinal classification algorithms and the feature space lacked micro-level logistical timestamps and textual client reviews. To address these constraints, future research trajectories should focus on: (1) expanding data collection to cross-organizational, multi-sector wholesale databases to validate model generalizability; (2) implementing dedicated ordinal machine learning models to capture the intrinsic mathematical progression of satisfaction levels; (3) enriching the feature space by integrating automated Natural Language Processing (NLP) text-mining models applied to raw telephonic and digital customer review transcripts; and (4) executing randomized, controlled field pilots to empirically measure the financial and operational Return On Investment (ROI) of SHAP-guided, proactive retention interventions. The rapid growth of Business-to-Business (B2B) e-commerce platforms necessitates robust, data-driven frameworks for anticipating and managing customer experiences. This study addressed the challenge of predicting customer satisfaction within an online industrial B2B wholesaler in Iran by proposing an optimized machine learning ensemble approach. Using an operational dataset comprising 349 distinct orders, we formulated the prediction task as a multi-class ordinal classification problem with 3 categories: Satisfied, Neutral, and Dissatisfied. A majority-voting ensemble architecture was developed, integrating 3 diverse base learners—CatBoost, Random Forest (RF), and a Multi-Layer Perceptron (MLP) neural network. To maximize predictive accuracy, the hyperparameters of each base algorithm were independently tuned by using a Genetic Algorithm (GA). Empirical evaluation on a held-out validation set showed that the proposed GA-tuned majority-voting ensemble achieved outstanding performance with an Accuracy of 0.9575, Recall of 0.9476, F1-score of 0.9379, and Matthews Correlation Coefficient (MCC) of 0.8966. Moreover, an equal-weight TOPSIS evaluation across multiple metrics ranked the ensemble model first among all alternative configurations. To address the "black-box" nature of the ensemble, a SHAP explainability analysis was conducted, revealing that operational parameters—such as Days Since Last Purchase and indicators of basket volume and weight—served as the primary drivers of satisfaction, while demographic attributes and purely financial variables exhibited negligible predictive power. This scalable pipeline provides actionable strategic insights for wholesale managers to implement proactive customer retention interventions. Introduction In the contemporary digital commerce paradigm, understanding and anticipating customer satisfaction is widely recognized as a cornerstone of corporate longevity, market sustainability, and effective customer relationship management. Within the Business-to-Business (B2B) e-commerce sector, proliferation of digital storefronts and wholesale platforms has dramatically intensified market competition. Unlike Business-to-Consumer (B2C) dynamics, B2B e-commerce involves highly complex operational pipelines characterized by large order volumes, high financial stakes, specialized transaction conditions, and extended procurement cycles. Consequently, the quality of the customer experience delivered by a wholesaler extends far beyond commodity pricing, encompassing multifaceted operational dimensions, such as delivery timeliness, packaging durability, shipping accuracy, logistical transparency, and after-sales service responsiveness. Despite the critical importance of these factors, traditional B2B operations often rely on retrospective feedback mechanisms—such as periodic phone surveys or post-delivery online ratings—to gauge client sentiment. These reactive approaches inherently limit the ability of an organization to implement timely, mitigating interventions before client dissatisfaction escalates into churn. To address this limitation, contemporary academic literature has shifted from theoretical, post-hoc explanations of customer loyalty toward proactive, data-driven predictive modeling enabled by machine learning and deep learning algorithms. By leveraging real-time transactional, behavioral, and demographic features, predictive pipelines can identify vulnerable client accounts before formal grievances are recorded. However, deploying predictive models within emerging markets, such as Iran, introduces distinct macro-environmental complexities. Wholesale enterprises operating in Iran contend with macroeconomic volatility, supply chain disruptions, fluctuating regulatory frameworks, and unique localized purchasing behaviors. In this context, predicting customer sentiment requires robust analytical models capable of extracting subtle behavioral patterns from relatively small, heterogeneous datasets without succumbing to overfitting. Although sophisticated deep learning architectures typically demand large datasets for effective generalization, machine learning ensembles and intelligent optimization algorithms—such as Genetic Algorithms (GAs)—offer a mathematically rigorous and computationally efficient alternative for modeling localized operational realities. This research directly addressed a notable gap in the existing literature: the absence of tailored, explainable machine learning frameworks for predicting multi-class customer satisfaction in small-sample B2B online wholesale environments operating under economic constraints. The primary objective of this study was to construct, optimize, and evaluate a comprehensive machine learning architecture capable of predicting customer satisfaction prior to the receipt of formal feedback. By validating this model on operational data from an industrial B2B wholesaler in Iran, this study established a portable framework that balanced high predictive accuracy with clinical interpretability. In doing so, it would equip corporate decision-makers with the analytical infrastructure needed to optimize resource allocation, enhance logistical workflows, and systematically foster long-term B2B client loyalty. Materials & Methods The methodological workflow of this study followed a highly structured and scientifically rigorous pipeline comprising data acquisition, preprocessing, architectural design of the ensemble model, evolutionary hyperparameter optimization, and multi-criteria evaluation. The empirical foundation of this research was built upon an operational dataset of 349 unique customer orders extracted from the Enterprise Resource Planning (ERP) and Customer Relationship Management (CRM) databases of Arman Roshan Fartak, a prominent industrial wholesaler specializing in industrial adhesives in Iran. The target variable representing customer satisfaction was meticulously collected through a synchronized dual-channel feedback mechanism that integrated systematic online post-purchase questionnaires with follow-up telephonic surveys conducted by specialized account managers. To align with the internal evaluation metrics of organizations, the target variable was encoded as a multi-class ordinal variable with 3 distinct levels: Class 0 ("Satisfied" clients), Class 1 ("Neutral" clients), and Class 2 ("Dissatisfied" clients). The predictive feature space encompassed 3 primary dimensions: procurement demographics, purchase volume and value indicators, and relational interaction signals. Specifically, the features included: the gender of the corporate purchasing officer, the age of the purchasing officer, the geographical location (categorically binned into metropolitan hubs versus regional municipalities), the cumulative financial expenditure incurred by the client up to the point of the current transaction ("Total Spend"), the aggregate physical weight of all products purchased ("KG"), the commission structure applied to the sales agent ("Commission"), the exact number of days elapsed since the client's immediately preceding transaction ("Days Since Last Purchase"), and the corporate legal status of the client enterprise (private sector versus public or governmental entity). The data preprocessing phase was executed with utmost care to ensure data integrity and prevent any form of data leakage. Of the initial 349 records, a small subset of 3 records (approximately 0.86%) exhibiting systematic and unrecoverable omissions in critical operational fields was entirely removed from the sample, yielding a final analytical dataset of 346 records. Categorical variables were transformed by using standard label encoding and one-hot encoding frameworks. To accommodate the specific structural requirements of the neural network component, continuous numerical variables were standardized by using a StandardScaler protocol, mapping the data to a mean of zero and a standard deviation of 1, thereby eliminating scaling biases. To establish a rigorous defense against overfitting and ensure robust generalization given the limited sample size, all subsequent training and optimization steps were embedded within a stratified 5-fold cross-validation framework, maintaining exact class proportions across all validation folds. The core predictive engine consisted of a heterogeneous majority-voting ensemble architecture designed to leverage the complementary mathematical strengths of 3 distinct base learners: CatBoost, Random Forest (RF), and a Multi-Layer Perceptron (MLP). CatBoost, a symmetric gradient-boosting tree framework, was selected for its inherent mathematical superiority in processing heterogeneous tabular data and its algorithmic defenses against target leakage. Random Forest was incorporated as a robust, non-linear bagging baseline known for its low variance and resilience to localized noise. The MLP neural network was integrated to capture high-dimensional, non-linear interactions within the feature space. Rather than relying on computationally blind hyperparameter selection methods, such as Grid Search or Random Search, this study deployed an evolutionary GA to independently discover the optimal hyperparameter configurations for each base learner. The GA optimization employed a population size of 20 chromosomes over a maximum of 40 generations with a tournament selection size of 3, a crossover probability of 0.85, a mutation probability of 0.10, and an early stopping criterion set to 10 consecutive generations without fitness improvement. The fitness function guiding the evolutionary search was defined as the stratified cross-validated weighted F1-score. To aggregate the individual base model outputs into a single final prediction, a hard voting (majority voting) protocol was chosen over soft voting following a rigorous preliminary sign test, which confirmed that hard voting offered superior mathematical stability and lower susceptibility to uncalibrated probability estimates given the sample size of this study. Research Findings The empirical results derived from the cross-validated training and testing phases demonstrated the substantial performance advantages achieved by integrating evolutionary optimization with an ensemble architecture. During the baseline evaluation of individual uncombined learners, CatBoost emerged as the most powerful standalone algorithm, exhibiting superior performance across nearly all primary evaluation metrics compared to RF and the MLP network. However, following the implementation of the hard-voting aggregation protocol, the finalized majority-voting ensemble delivered the most balanced, robust, and highly accurate performance, establishing its statistical dominance. On the held-out validation datasets, the proposed ensemble achieved an overall accuracy of 0.9575, a precision of 0.9396, a recall of 0.9476, and a weighted F1-score of 0.9379. To validate classification performance beyond simple accuracy, we computed advanced metrics, yielding a Matthews Correlation Coefficient (MCC) of 0.8966 and a Cohen's Kappa (CK) coefficient of 0.8957. These values confirmed that the model's predictive power was highly stable and entirely independent of any underlying class imbalances within the operational dataset. To mathematically validate the superiority of the ensemble approach without relying on subjective, single-metric interpretations, a Multi-Criteria Decision-Making (MCDM) evaluation was performed using the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS). The evaluation matrix integrated 6 core performance dimensions—Accuracy, Precision, Recall, F1-score, MCC, and CK—calculated across all models. To ensure completely unbiased results, an equal-weighting scheme was applied across all criteria. The calculated TOPSIS closeness coefficients (CiCi), which represented each model's Euclidean proximity to the mathematically ideal solution and distance from the anti-ideal solution, explicitly yielded the following ranking: Ensemble (C=0.9695C=0.9695, Rank 1) > CatBoost (C=0.9187C=0.9187, Rank 2) > Random Forest (C=0.3437C=0.3437, Rank 3) > Multi-Layer Perceptron (C=0.1037C=0.1037, Rank 4). This rigorous multi-criteria analysis confirmed that the ensemble architecture systematically minimized classification errors when evaluated across all operational performance indicators simultaneously. An in-depth analysis of the classification mechanics was conducted by inspecting the cross-validated confusion matrices for all four model variants. The ensemble architecture demonstrated an extraordinary capacity to identify vulnerable client segments, achieving a true positive count of 25 for Class 2 ("Dissatisfied")—meaning it attained a perfect 100% detection rate for dissatisfied accounts with zero false negatives in this critically important operational category. For Class 1 ("Neutral"), the ensemble maintained high precision, misclassifying only 2 out of 36 instances as "Satisfied". For Class 0 ("Satisfied"), the model exhibited a minimal margin of error with only a single instance incorrectly flagged as "Neutral". In contrast, the standalone MLP showed severe confusion, frequently failing to differentiate the subtle boundaries separating Class 1 and Class 2 instances. To unlock the "black-box" architecture of the optimized ensemble and extract actionable strategic insights, we applied SHAP (SHapley Additive exPlanations) values, which were grounded in cooperative game theory. The global feature importance hierarchy was established by calculating the mean absolute SHAP values averaged comprehensively across all three classification classes. The resulting SHAP summary plot revealed an extremely pronounced skew in predictive influence. The behavioral variable of "Days Since Last Purchase" emerged as the dominant driver of customer satisfaction predictions, achieving a mean ∣SHAP∣∣SHAP∣ value exceeding 1.6—a magnitude that mathematically quadrupled the influence of any other feature in the space. The volume and value metrics, namely "Total Spend" and the total physical weight of purchased goods ("KG"), exhibited moderate predictive influence acting as secondary drivers. Crucially, purely financial variables—such as the agent's "Commission"—along with all consumer demographic attributes—including the purchasing officer's age, gender, geographic city classification, and organizational company type—displayed negligible SHAP values, exerting almost no discernible impact on the model's ultimate classification decisions. Discussion of Results & Conclusion The empirical insights generated by this research carry both profound theoretical value for the domain of digital B2B marketing and immediate, actionable utility for operational management within wholesale enterprises. By demonstrating that a heterogeneous ensemble engine, optimized via evolutionary GAs, achieved an Accuracy of 95.75% and an MCC of 0.8966 on a compact operational dataset, this study successfully refuted the conventional assumption that sophisticated machine learning modeling is reserved exclusively for enterprises possessing massive, multi-million-record big data repositories. The perfect detection rate achieved by the ensemble for dissatisfied accounts (Class 2) held profound managerial significance: it proved that operational data natively generated during routine wholesale transactions contained highly structured, predictive behavioral signals that could be systematically exploited to deploy preemptive customer recovery strategies before relationship dissolution occurred. When contextualized within contemporary academic literature, the findings of this study provided both crucial validations and distinct points of divergence. In contrast to the prominent research conducted by Zaghloul et al. (2024), which utilized an immense dataset exceeding 100,000 e-commerce orders to predict customer satisfaction via standalone RF and Support Vector Machine (SVM) models—achieving an accuracy of approximately 92%—our framework demonstrated that an evolutionarily tuned ensemble architecture could extract significantly higher predictive accuracy (95.75%) and a vastly superior multi-criteria balance (validated via TOPSIS) from a small data footprint (n=346n=346). Furthermore, our SHAP-based feature importance hierarchy strongly reinforced the empirical conclusions of Tang (2025) and Hui et al. (2025), who argued that within B2B industrial environments, operational logistics signals, transaction recency, and purchase volume metrics carry exponentially greater predictive weight regarding customer sentiment than traditional demographic segmentation or basic pricing mechanisms. The extraordinary dominance of the feature of "Days Since Last Purchase" within our SHAP explainability pipeline underscored a vital behavioral reality in the Iranian industrial wholesale sector. In this context, an increasing temporal gap between orders did not merely signify a passive pause in purchasing; rather, owing to intense competition and macro-environmental supply disruptions, it served as an active, early-warning indicator of client alienation, low engagement, or silent switching to alternative suppliers. Conversely, the moderate significance of "KG" (physical weight) and "Total Spend" confirmed that clients who consolidated large, heavy industrial orders established deeper, more integrated operational ties with the wholesaler, rendering them highly sensitive to logistics execution quality, packaging reliability, and delivery schedules. The finding that demographics and commission structures exerted almost zero predictive influence sent a strong signal to wholesale executives: broad financial discounting policies or rigid demographic marketing campaigns were fundamentally ineffective levers for securing long-term loyalty. Instead, corporate interventions had to be aggressively redirected toward recency-based customer activation systems and the systematic optimization of fulfillment workflows. In conclusion, this study successfully established an optimized, highly accurate, and mathematically explainable machine learning pipeline for predicting customer satisfaction within the online B2B industrial wholesale sector. By fusing CatBoost, RF, and MLP models through a hard-voting ensemble, and optimizing their internal structures via a G A, the framework provided a highly reliable predictive tool that remained fully transparent through SHAP value interpretation. From a managerial perspective, the model offered an analytical blueprint for shifting corporate strategies from reactive grievance resolution to proactive client retention, prioritizing interaction recency and fulfillment reliability over superficial financial adjustments. However, certain structural limitations must be acknowledged. This study relied on a compact, single-organization dataset from an industrial setting in Iran, which might limit the direct generalizability of the exact parameter values to substantially different economic sectors. Additionally, the classification pipeline employed a standard multi-class formulation without deploying specialized ordinal classification algorithms and the feature space lacked micro-level logistical timestamps and textual client reviews. To address these constraints, future research trajectories should focus on: (1) expanding data collection to cross-organizational, multi-sector wholesale databases to validate model generalizability; (2) implementing dedicated ordinal machine learning models to capture the intrinsic mathematical progression of satisfaction levels; (3) enriching the feature space by integrating automated Natural Language Processing (NLP) text-mining models applied to raw telephonic and digital customer review transcripts; and (4) executing randomized, controlled field pilots to empirically measure the financial and operational Return On Investment (ROI) of SHAP-guided, proactive retention interventions.

https://nmrj.ui.ac.ir/article_30500_393c514db9da07c583f0cf4178751a11.pdf