5.1 Thematic analysis of clustering result
Based on a comparison of three clustering algorithms, we ultimately selected hierarchical clustering for our subsequent analysis. While the section on topic modeling algorithm selection has already provided an initial explanation of the BERTopic results based on hierarchical clustering, this section delves deeper into the conceptual structure of AI embeddedness. To achieve this, we used a literature review approach to examine papers with more than 100 citations within each topic, thereby laying a solid foundation for the formation of the AI embeddedness concept. To clarify the mapping logic between the 9 topics and the three dimensions of AI embeddedness (Link, Fit, Sacrifice), we constructed a transparent coding scheme and validated its reliability, as shown in the Table 5. An independent marketing scholar who is unrelated to the research team, re-mapped the 9 topics using the above scheme, with Cohen’s kappa = 0.82 (p < 0.001), meeting the standard for substantial reliability and ensuring the mapping is auditable.
5.1.1 Class 1: AI-Integrated in financial and marketing operations
Topic 1 primarily describes the impact of deep learning algorithms on financial marketing and stock market prediction. Fischer and Krauss (2018) applied long-short term memory (LSTM) networks to a large-scale financial market prediction task, successfully demonstrating that LSTM networks can effectively extract meaningful information from noisy time series data to predict price trends. Similar studies have used neural networks and other intelligent technologies for stock market prediction, indicating a close relationship between employees, consumers, and digital technology. Consequently, artificial intelligence and similar technologies have been long-term applications in investor behavior research (Chen et al., 2003; Cao et al., 2005; Leippold et al., 2022). Additionally, by applying large-scale deep learning methods to stock market research, it is possible to more accurately predict market price changes and make timely adjustments to company marketing strategies (Tambe, 2014; Cont and Sirignano, 2019).
Topic 2 primarily describes the application of neural networks in marketing. As early as 2004, Huang et al. conducted a market comparative study on credit rating analysis using neural networks, finding that neural network algorithm models provide better interpretability for credit ratings, a conclusion also supported by Leigh et al. (2002). Subsequently, researchers built comprehensive bankruptcy databases by extracting features from textual data using neural network layers, demonstrating that deep learning models based on neural networks excel in predicting bankruptcy through textual disclosures, thereby offering more intelligent models for investors, regulators, and researchers (Mai et al., 2019). Moreover, neural network models have been widely applied in direct marketing. For example, Cui et al. (2006) used Bayesian networks to model consumer responses, aiding in management decision-making, and He et al. (2021) employed neural network algorithms to delve into consumer neuroscience. These studies collectively highlight that neural networks play a crucial role in direct marketing by enhancing the accuracy of customer targeting, predicting consumer behavior, and optimizing marketing strategies through advanced data analysis and pattern recognition.
Following a comprehensive literature review of the topics encompassed in class 1, we have designated this class as "AI Integration in Financial and Marketing Operations." This category captures the application of deep learning and neural networks within the financial and marketing sectors, highlighting the integration of artificial intelligence in these domains and the pivotal role of employees in leveraging these technologies.
5.1.2 Class 2: AI-Enhanced marketing tactics and innovations
Topic 3 primarily explores the application of data mining in marketing contexts such as product recommendation and sales forecasting. Cortez et al. (2009) introduced a data mining approach to predict human wine taste preferences using various algorithms, including support vector machines, neural networks, and multiple regression. This method aids targeted marketing by simulating consumer tastes in niche markets. Researchers have employed data mining techniques to extract customer knowledge for product development across different market segments, thereby aligning customer needs with product development processes to reduce trial-and-error costs and enhance sales performance (Agard and Kusiak, 2004; Su et al., 2006; Yu et al., 2020). By leveraging data mining, companies can better understand and predict online consumer product demands. Previous studies have utilized online promotional marketing and online reviews as predictive factors, employing deep learning algorithms such as neural networks to analyze online data comprehensively. This approach yields valuable insights to inform corporate marketing strategies (Chong et al., 2017; Loureiro et al., 2018). The extensive application of data mining in marketing has significantly strengthened the integration of companies, consumers, and their stakeholders with digital technologies, including artificial intelligence, in the digital age.
Topic 4 primarily delves into the application of digital technologies in social media marketing. Liu and Shih (2005) pioneered a marketing model that integrates hierarchical analysis and data mining techniques to deliver precise product recommendations tailored to each customer segment. This seminal study has exerted considerable influence in this domain. Building on their foundation, subsequent researchers have leveraged sentiment analysis, big data architectures, and advanced deep learning algorithms to investigate how online reviews and online promotional marketing strategies can predict sales performance. They discovered that sentiment significantly interacts with both the volume and valence of online reviews, thereby substantially influencing and forecasting product sales (Chong et al., 2016; Kauffmann et al., 2020). Furthermore, contemporary scholars have enhanced social media marketing through the implementation of a visual data analytics framework. By employing deep learning algorithms, they have explored various contexts, including the study of CEO oral communication, to strengthen the impact of social media marketing efforts (Choudhury et al., 2019; Shin et al., 2020).
Topic 6 focuses on data-driven marketing models utilizing deep learning algorithms, such as neural networks, for predictive analysis. Researchers have employed enhanced hybrid ensemble machine learning approaches and text-based deep learning methods to forecast crude oil prices and assess the credit risk of SMEs in supply chain finance, thereby offering scientific guidance for developing corporate marketing strategies (Li et al., 2019; Zhu et al., 2019). Currently, marketing models based on deep learning algorithms are extensively applied in studies on customer demand analysis, product innovation, and green technology innovation. These applications highlight the effective integration of human resources and artificial intelligence in marketing, providing a robust theoretical foundation for business performance improvement (Li et al., 2018; Zhang et al., 2018; Yaoteng & Xin, 2022). Moreover, recent research has explored financial performance and marketing innovation through deep learning algorithms in various contexts, such as supply chain strategy and intercity connections, contributing to a deeper understanding of these domains (Luo et al., 2023; Wang et al., 2023a, b).
Topic 8 primarily explores how platform companies develop service marketing strategies through product and technological innovation. Current research on platform companies focuses on the impact of two major digital technologies, the Internet of Things (IoT) and blockchain, on the platform ecosystem (Liu et al., 2021). Basaure et al. (2020) used agent-based modeling to analyze the effects of consumer switching costs and provider multi-homing on market structure and competition. Simulation results indicate that when switching costs decrease due to factors such as consumer data portability, provider multi-homing plays a critical role in enhancing market competition. Additionally, cloud computing has also played a significant role in the formulation of service marketing strategies for platform companies. Studies have shown that cloud computing can effectively help companies track consumer adoption behavior in strategic management, make timely adjustments in service failure scenarios to retain customers, and improve user satisfaction and trust (Shin et al., 2014; Huang et al., 2015; Chen et al., 2019).
Following an extensive literature review of the four topics within class 2, we have named this category "AI-Enhanced Marketing Tactics and Innovations". This designation underscores the application of artificial intelligence in data mining, social media marketing, predictive analysis, and the innovation of service marketing strategies, specifically illustrating AI in enhancing marketing strategies and driving innovation.
5.1.3 Class 3: AI-Driven data analytics and VR innovations
Topic 0 primarily describes the widespread application of big data in marketing, encompassing areas such as operations management, e-commerce, and strategic management. Grover et al. (2018), based on the resource-based view, proposed a big data value creation framework, arguing that the better the alignment between employees and consumers with big data analytics, the more significant the improvements in company performance and consumer well-being. Existing research indicates that big data can provide unique insights into market trends, customer purchasing patterns, and maintenance cycles, as well as methods to reduce costs and make more targeted business decisions. Therefore, employees or consumers with big data analytical capabilities can overcome computational and data challenges, further enhancing job performance and consumer well-being (Wang et al., 2016; Choi et al., 2018). The operational performance improvements brought about by big data have been further confirmed in studies targeting manufacturing organizations and e-commerce scenarios (Akter & Wamba, 2016; Dubey et al., 2020).
Topic 5 primarily examines the impact of data analytics capabilities on corporate performance and market innovation. Research indicates that big data analytics capabilities enable firms to generate insights that enhance their dynamic capabilities, thereby positively influencing both marketing and technological competencies (Mikalef et al., 2020). With advances in artificial intelligence and the growing availability of data, data network effects manifest in the positive direct relationship between a platform's AI capabilities and users' perceived value of the platform. When data network effects are activated, customers perceive the value brought by AI, further strengthening the company's competitive advantage (Gregory et al., 2021). Existing studies, based on the knowledge-based view theory, demonstrate that the business value realized through investments in big data analytics can provide financial performance advantages to companies. AI technologies enabled by big data enhance firm performance through knowledge creation processes involving customers, users, and the external market (Raguseo & Vitari, 2018; Bag et al., 2021).
Topic 7 mainly discusses the application of virtual reality technology in tourism services and its impact on consumer experience and purchase intentions. The rapid development of virtual reality technology offers opportunities for widespread consumption of tourism content. It also presents challenges in understanding the effectiveness of virtual experiences in fostering more favorable attitudes towards tourist destinations and shaping visitation intentions (Tussyadiah et al., 2018). Bogicevic et al. (2019) found that virtual reality previews can provide more detailed explanations of experiential mental imagery and generate a stronger sense of presence, thereby enhancing brand experience. Through digital technologies like virtual reality, consumers resonate with tourism brands, which benefits companies in adjusting marketing strategies and choosing strategic directions. Additionally, applying augmented reality technology in scenarios such as museums and cultural heritage sites enables consumers to have immersive experiences, thereby influencing their purchase motivations, stimulating purchase interest, and driving consumption behavior (Javornik, 2016; He et al., 2018; Jung et al., 2018).
Following a thorough literature review of the topics within class 1, we have designated this category as "AI-Driven Data Analytics and VR Innovations." This title reflects the integration of AI-driven data analytics and virtual reality technology in marketing, highlighting the significant impact these innovations have on corporate performance and consumer experience.
5.2 Definition of AI embeddedness
Through thematic clustering analysis of 640 cleaned documents and systematic extraction from 52 highly cited papers (≥ 100 citations; see Sect. 3.3 for details), it is evident that with the rapid advancement of digital technology, the integration between employees and artificial intelligence in the marketing sector has become increasingly seamless. Management practices such as Generated-AI, advanced chatbots, and customer customization have given rise to the concept of AI embeddedness. Drawing on Zhang et al. (2012)'s review of job embeddedness, this section uses the three dimensions of job embeddedness theory — link, fit, and sacrifice — combined with clustering results and findings from the analysis of highly cited papers, to comprehensively explain the concept of AI embeddedness.
5.2.1 Link dimension of AI embeddedness
The link dimension in job embeddedness refers to the formal and informal relationships between employees and the organization and its members. For AI embeddedness, this dimension is derived from the analysis of 52 highly cited papers: As data becomes a crucial production factor, AI has deeply penetrated domains such as consumer behavior, customer relationship management, service marketing, and brand management through extensive data training sets (76% of the 21 highly cited papers in Cluster 2 mention such penetration). These penetrations result in two types of links: formal relationships where enterprises utilize AI for marketing purposes, and informal relationships where employees use AI to enhance their job performance. Therefore, the link dimension of AI embeddedness refers to the degree of connection between AI systems and various organizational processes, employees, and external stakeholders. This includes how AI technology is integrated into workflows to enhance collaboration and communication within the organization, as well as the depth of connection between AI tools and other systems, such as Customer Relationship Management platforms, marketing automation tools and employee communication channels.
5.2.2 Fit dimension of AI embeddedness
The fit dimension in job embeddedness refers to the alignment between employees' growth paths and organizational development plans, as well as the congruence between personal values and organizational beliefs. For AI embeddedness, core findings of this dimension come from systematic extraction from highly cited papers: Amazon employs AI algorithms to analyze users' browsing and purchase histories, offering personalized product recommendations; Salesforce leverages AI to analyze customer data, providing insights into customer behavior and sales forecasts to help marketing teams develop more effective customer engagement strategies (Acharya et al., 2018; Chen & Guo, 2022), these cases are all from the top 10 most cited papers in Clusters 1 and 2. This alignment between employees and AI in marketing not only enhances user satisfaction and purchase rates but also reflects corporate values and promotes brand culture. Therefore, in the concept of AI embeddedness, the fit dimension refers to the degree to which AI technology aligns with business needs, employee work styles, and organizational culture in the marketing field. By improving the efficiency and effectiveness of personalized recommendations, dynamic ad generation, customer relationship management, and content creation, AI applications ensure that they meet and support the company's strategic goals and operational processes. This conclusion is consistent with 83% of findings in the 52 highly cited papers.
5.2.3 Sacrifice dimension of AI embeddedness
The sacrifice dimension in job embeddedness refers to the tangible or intangible losses employees might face upon leaving, such as loss of salary, career advancement opportunities, and personal relationships. For AI embeddedness, the connotation of this dimension is based on the focused discussion of risks in highly cited papers. Algorithmic black boxes (Basukie et al., 2020), AI-related errors (Falk, 2021), and big data price discrimination (Valletti & Wu, 2020) have emerged as critical issues, making it crucial to address the welfare dilemmas posed by AI embeddedness. These studies all come from the 13 highly cited papers in Cluster 3, with 92% emphasizing the inevitability of such risks. Beyond technical and economic risks, there are also socio-technical costs that have been under-addressed in prior discussions.
One key socio-technical cost is deskilling among marketing employees. As AI automates repetitive yet skill-dependent marketing tasks — like manual customer segmentation and heuristic market analysis. employees may gradually lose proficiency in these core competencies. This weakens their career resilience and increases organizational dependence on AI systems, creating hidden operational risks if AI fails (Puntoni et al., 2021). Another is algorithmic bias in AI marketing systems. AI recommendation or pricing systems often inherit biases from historical training data; for example, overrepresenting high-income consumer preferences may lead to discriminatory recommendations or worsened big data price discrimination (Valletti & Wu, 2020). Such bias violates marketing fairness, triggers regulatory scrutiny, and erodes consumer trust, resulting in intangible losses like brand image damage that are harder to mitigate than direct financial costs.
Therefore, in the concept of AI embeddedness, the sacrifice dimension involves potential losses or costs from adopting and implementing AI technology. This includes financial investments, time for training and integration, risk of certain job roles being displaced, data privacy and security issues, employee skill degradation, and reputational or regulatory costs from algorithmic bias. Understanding these sacrifices helps organizations balance benefits and costs. For instance, by designing AI-human collaboration training programs to mitigate deskilling or conducting regular algorithmic audits to reduce bias, leading to informed decisions about AI integration.
