Explore barriers and risks that hinder AI integration into global talent management: A Systematic Review

Problem statement

AI has the potential to improve talent management techniques through the implementation of advanced automated workforce management solutions (Faqihi & Miah, 2023). AI has transformed modern workplaces in an unprecedented manner, leading to digital workstyles in HRM (Singh & Pandey, 2024). While the integration of AI in businesses has potential, it also presents problems that must be addressed to achieve successful deployment (Scheffknecht, 2025). AI has demonstrated its promise in supporting human decision-making, but there are challenges to using AI for this purpose (Booyse & Scheepers, 2024). The integration of AI in corporate contexts is rapidly increasing; however, substantial limitations prevent its successful application and adoption (Alawamleh, Shammas, Alawamleh, & Ismail, 2024). Organisations often face difficulties when integrating AI into HR processes, primarily due to limited resources such as skilled IT professionals and adequate technological infrastructure (Madanchian & Taherdoost, 2025). A shortage of experienced AI specialists affects implementation efforts, underscoring the importance of developing skills and competence (Zavodna, Überwimmer, & Frankus, 2024). Issues such as algorithmic bias, the dehumanisation of work relationships, the opacity of automated decision-making systems, and job displacement have sparked heated debate, underscoring the importance of a responsible and thoughtful approach (Rosário, 2025; Madanchian & Taherdoost, 2025). Mwita and Kitole conducted their investigation in 2025 and identified substantial obstacles, including biases towards a lack of transparency, privacy concerns, and the inability to address the qualitative components of HR processes. Although numerous studies have explored talent management and AI, a notable gap remains in scholarly research examining the barriers and risks that hinder AI adoption in GTM. Therefore, this study seeks to close this gap.

Research gap and purpose

This systematic review aimed to identify the primary barriers and risks associated with implementing AI in GTM.

Concept of AI

AI is considered a key element of the Fourth Industrial Revolution, which the organisations believe will transform the way businesses function. With the rapid advancement of AI technologies, their incorporation into various work domains has become unavoidable (Tiwari, Babu, Marda, Mishra, Bhattar, & Ahluwalia, 2024). AI is rapidly transforming the global workforce and corporate environment through task automation, enhanced decision-making, and the facilitation of innovative operational methods (Farouk, 2025). It encompasses a collection of technologies that perform computational functions often designated for humans and is crucial to contemporary global discussions regarding social and technical transformations (Benhamou, 2020). What has become a standard practice in organisations across the globe is the integration of AI into business operations to perform routine low level tasks that were performed by humans (Poisat, Cullen & Calitz, 2024). It is defined as computer systems capable of learning and executing activities generally associated with human intelligence, with the potential to ultimately surpass human capabilities (Diyin, Bhaumik, & Wang, 2024). AI employs algorithms that combine high-quality data and fast computational services, resulting in Core AI, which provides stability and precision for everyday activities (Bhardwaj, Singh, and Kumar, 2020). As a result, there is little doubt that incorporating AI into organisational work practices brings considerable advantages, leading to meaningful enhancements in various areas of business operations (Murire, 2024). AI is designed to streamline, automate, and support decision-making within organisations, because it complements the employee’s tasks (Tiwari, Babu, Marda, Mishra, Bhattar, & Ahluwalia, 2024). As noted by Rožman, Oreški, and Tominc (2022), AI enhances operational efficiency and creates opportunities for the development of new products. AI has the potential to enhance employee engagement by customising opportunities for professional growth and increasing job satisfaction (Diyin, Bhaumik, & Wang, 2024).

Integration of AI in human resources

Human Resource Management (HRM) plays a pivotal role in organisations, encompassing key functions such as recruitment, employee retention, and motivation (Ochieng, 2023). Traditional HRM approaches have faced criticism for being time-intensive and influenced by human bias, subjectivity, and personal judgment (Bujold et al., 2024). The adoption of AI in HRM has transformed organisational functions, reflecting its ability to analyse candidate data more quickly while remaining free of human prejudice or inaccuracy (Liu, Li, & Xia, 2021; Rosário, 2025; Pandey & Khaskel, 2019). AI has undeniably become a valuable asset in handling HRM functions, including managing employees, refining recruitment processes, selection, evaluating candidate-job compatibility, performance management, workforce engagement and supporting practical employee training and development initiatives (Mir, 2024; Tewari & Pant, 2020; El-Ghoul, Almassri, El-Habibi, Al-Qadi, Abou Eloun, Abu-Nasser, & Abu-Naser, 2024; Budhwar, Malik, De Silva, & Thevisuthan, 2022). This tool has a significant impact on employee engagement, streamlining repetitive administrative tasks and enabling HR professionals to concentrate on more strategic and value-adding responsibilities (Chukwuka & Dibie, 2024; Madanchian & Taherdoost, 2025).

The global rise of AI has sparked discussions about its impact on human jobs. While some are concerned about job displacement, others argue that AI can work alongside human labour to boost productivity, innovation, and efficiency (Spulbar & Mitrache, 2025). In the context of Human Resource Management (HRM), AI presents both opportunities and challenges (Pereira, Hadjielias, Christofi, & Vrontis, 2023). Recognising its advantages is key to encouraging its adoption (Madanchian & Taherdoost, 2025). Table 1 below outlines the pros and cons of AI.

AI in talent management

Hiring employees with the right skills is essential to an organisation’s success, as its overall performance is closely tied to the efforts and effectiveness of its workforce. For this reason, AI provides unique benefits by reinventing TM, with its significant potential to revolutionise how organisations attract, manage, and retain their staff (Al-Mughairi, 2026). There is evidence in the literature that the introduction of AI is altering existing TM processes, with implications that could be negative, positive, or both, and substantial potential benefits for organisational performance (Gao & Segumpan, 2024). AI integration in TM implies optimising several aspects of the employee lifecycle, from recruiting and selection to development, engagement, and performance tracking (Khan, 2024; Lourens, Vijai, Nijhawan, Garwal, Gakhreja & Saud, 2024). It also has the potential to enhance personnel management by automating repetitive tasks, reducing errors, and increasing efficiency in critical areas such as recruitment, performance evaluation, employee retention, and talent acquisition (Mir, 2024). TM is the process of placing individuals with appropriate talents and capabilities in roles that best match their strengths (Ekhsan, Parashakti, & Perkasa, 2023). The primary goal of TM is to attract, recruit, place, develop, and retain top talent, while top talent, thus ensuring a positive vision for the future of work within an organisation (Sowmya, Polisetty, & Dash, 2024; Briki & Gherrab, 2024).

Guiding theoretical frameworks

Several theoretical frameworks were evaluated for their applicability to the study. The first theory that holds promise for explaining the findings is the Technology Acceptance Model (TAM) (Venkatesh et al., 2003; Venkatesh, Thong, & Xu, 2012). TAM posits that users adopt technology more quickly when they perceive it as both helpful and easy to use. It holds promise to help explain why HR practitioners may resist using AI tools to support TM in their organisations, even though they may be aware of some of the benefits. The TAM theory was extended by the Unified Theory of Acceptance and Use of Technology (UTAUT), which added facilitating conditions, including organisational support and technical infrastructure, as well as social influence (Venkatesh, 2012). Another theory that shows promise to make sense of AI adoption in TM is the STST (Socio-Technical Systems Theory), which emphasises that technology implementation may fail due to socio-technical misalignment (Baxter & Sommerville, 2011). Technical issues, for example, may negatively affect employees’ trust in the technology and hinder its adoption. This holistic approach helps explain that without social interventions, such as appropriate training and change management, AI solutions are likely to fail.

Protocol and reporting standards

A systematic literature review and a qualitative research approach were used as the research methods. A systematic review is a comprehensive literature review and analysis of results on a specific topic. Furthermore, it helps interpret findings or results. It begins with the conceptualisation of clear questions for critique by framing these questions, after which data are collected, categorised, evaluated, summarised, proofread, and discussed (Kang & Ahn, 2018). As suggested by Molefe, Sehularoa and Koen (2024:55), five steps were followed to conduct a systematic review: 1) Conceptualisation of clear questions for critique; 2) Collection and categorisation of data; 3) Evaluation; 4) Summary and proofreading; and 5) Discussion. Searches were done between 22 April and 10 June 2025. The PRISMA 2020 rules were employed to search eight databases: Google Scholar, Scopus, Web of Science Core Collection, ScienceDirect, ProQuest, Emerald Insight, Sabinet, and El and Elsevier. Searches were conducted between 2020 and 2025.

Quality appraisal

The researchers conducted a critical appraisal to evaluate the quality of the literature considered for eligibility. The methodological transparency, relevance, and peer review status of each publication were evaluated. Studies were weighted in the narrative synthesis rather than being disqualified on the basis of quality. Important metrics were monitored, such as regional applicability, AI domain specificity, and empirical design.

Information source and search strategy

A comprehensive literature search was conducted across Google Scholar, Scopus, Web of Science, ScienceDirect, ProQuest, Emerald Insight, Sabinet, and Elsevier. Articles published between 2020 and 2025 in peer-reviewed journals were considered. Two independent reviewers conducted screening and data extraction. A narrative synthesis and thematic analysis were applied between April and May 2025. The search string used was: (“artificial intelligence” OR AI OR “machine learning” OR “deep learning” OR automation OR “predictive analytics”) AND (“global talent management” OR “international HR” OR “global human resource management”) AND (integration OR adoption OR implementation). The initial search yielded 1,099 results across databases, as detailed in the systematic review tracker. Duplicates were removed using Zotero’s duplicate detection tool (1087 articles remained), and articles were then exported for screening in Rayyan.

Inclusion and Exclusion criteria

Studies were evaluated for eligibility and excluded if they did not align with the research objectives. Table 2 below shows the inclusion criteria used to choose publications for secondary data analysis.

Screening and selection process

Stage 1: Title and Abstract Screening

A PRISMA appraisal checklist and diagram were utilised to confirm that the original guidelines were followed. Two independent reviewers completed the initial screening in Rayyan to confirm that the results were relevant to the research questions. Conflicts were resolved by discussion and reassessment of the inclusion criteria. It was unnecessary to involve a third reviewer to address additional disagreements. The checklist links the PRISMA 2020 items to the pages and supplemental files in the systematic study “Integrating AI in GTM”. In this study, 1,099 articles were identified, of which 1,087 were subsequently eliminated prior to the commencement of the screening and selection process.

Stage 2: Full-Text Screening

A total of 272 publications were chosen for comprehensive screening, assessed according to the specified inclusion criteria:

Of 272 articles, 23 met all requirements and were selected for a comprehensive thematic synthesis. This screening process resulted in the exclusion of 249 articles. The reasons for elimination were a lack of focus on talent management, insufficient methodological transparency, and the absence of reference to AI implementation. Of these, 23 articles passed all criteria and were chosen for a comprehensive thematic synthesis. This source provides recommendations for the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA diagram) used in this review. The publication selection process diagram is presented in Figure 1 below.

Figure 1. Publication selection process diagram.

Source: Researchers’ own compilation.

Data extraction and thematic synthesis

The main reviewer reviewed all 23 chosen articles in full, while a secondary coder checked the theme assignments for accuracy. The articles were personally annotated before being posted to Zotero and Rayyan. The following procedures were used to identify themes using thematic synthesis:

Relevant text about AI use, its facilitators and obstacles that is open-coded.

Code grouping into broad themes and sub-topics.

Linking each theme to the contributions of articles.

Thematic pattern frequency analysis.

A visual representation of the connections between themes and sub-themes.

To ensure there was no overlap and to assess conceptual clarity, the final themes were examined. The robustness of this method was increased by a comprehensive codebook, frequency table, synthesis matrix, and PRISMA-compliant visual aids.

Geographic distribution of included studies

The systematic review’s final 23 articles cover a wide variety of geographical, sectoral, and methodological viewpoints. The findings of this study were derived from five regions. With eight publications from Asia (China, Singapore, the United Arab Emirates, and India), five from Europe (the United Kingdom, Germany, and the Netherlands), three from Africa (South Africa and Egypt), four from North America (the United States and Canada), and three from global/cross-regional studies. Asia was where most of the investigations started, then Europe.

Methodological distribution of studies

Three methodological techniques were used in the study: conceptual, empirical, and qualitative. With eleven submissions, empirical research was the largest group, while conceptual and qualitative studies were equally represented with six each.

Sectoral distribution of studies

The sectoral context of the collected papers was also investigated in this study. Thirteen of the 23 studies that were reviewed were conducted in the private sector, four of which looked into competitive or cross-sectoral settings, and four more of which looked into public or government organisations. With significant representation from both developed and developing economies, these distributions demonstrate a global orientation that promotes transferability and comparative insights.

Synthesising the findings

The number of main themes and sub-themes identified in peer-reviewed papers supporting the incorporation of AI into GTM is summarised in Table 3. The PRISMA diagram above shows that themes and sub-themes were developed from 23 articles. While AI offers considerable promise, its implementation in global talent GTM is shaped by governance and accountability, implementation barriers, and ethical and technical risks. The primary themes and subthemes are summarised in Table 4, along with definitions and examples of each theme and subtheme. Additionally, Figure 2 illustrates why AI integration stalls in GTM.

Figure 2. Barrier Breakdown: Why AI integration stalls in GTM.

Source: Researchers’ own compilation.

Theme 1: Governance and Accountability

This theme encompasses the ethical, legal, and procedural frameworks needed to ensure AI use in HR is fair, transparent, and accountable.

The risk is that “black box” algorithms lack transparency and may replicate historical inequities in hiring and promotion (Varma et al., 2024; Roppelt et al., 2025). Organisations must audit algorithms and introduce fairness controls, or risk eroding gains in inclusion (Sulaiman et al., 2025). Organisations must audit algorithms and introduce fairness controls or risk eroding equity and inclusion gains in GTM.

The ethical concern is that over-reliance on algorithms treats humans as objects to be acted upon (“patiency”) rather than autonomous agents (Varma et al., 2024). There is also a risk of dehumanisation in processes requiring high empathy, such as candidate rejection (Varma et al., 2024). Without explainable outputs, trust declines, especially in rejection decisions or promotion denials. Transparent communication about how AI influences HR decisions must be embedded in practice. Ethical lapses or a lack of transparency can erode trust and legitimacy. Governance structures must evolve in tandem with AI innovation to ensure fairness and regulatory compliance.

Theme 2: Implementation Barriers

This theme captures the organisational, cultural, and environmental obstacles, such as resistance to change and infrastructure deficits, that hinder the smooth adoption and integration of AI systems across global and local contexts.

Regional factors, including infrastructure limitations, language processing gaps, and traditional hiring preferences, hinder the implementation of standard AI models. In South Asian corporate sectors, particularly in Pakistan, implementation is significantly hindered by a cultural preference for personal connections in hiring, which 76.6% of respondents identified as a significant factor (Sulaiman et al., 2025). This relationship-oriented business culture, where referrals and personal networks traditionally dominate, creates tension with standardised, algorithm-based selection tools (Sulaiman et al., 2025; Hussain et al., 2022). Furthermore, hierarchical decision-making structures emerge as a major cultural barrier, affecting how AI recommendations are integrated into final selection processes (Sulaiman et al., 2025). Technical infrastructure limitations, such as internet connectivity issues and frequent power outages, represent practical barriers for 76% of organisations in these regions (Sulaiman et al., 2025). Language processing gaps also pose a critical technical barrier, as 60% of senior managers report that current AI systems struggle to accurately interpret bilingual resumes that mix English with regional languages such as Urdu or culturally specific terminology (Sulaiman et al., 2025). Finally, 84% of respondents cite data quality issues arising from inconsistent standards in regional education and non-standardised employment verification records, which complicates algorithm training and validation (Sulaiman et al., 2025).

The fear of job loss, loss of autonomy, and workplace surveillance among employees leads to reduced engagement or deliberate tool avoidance. A primary driver of resistance is “automation anxiety,” particularly among “white-collar” workers who fear that AI will eventually replace them for cost-efficiency reasons (Smith et al., 2025; Roppelt et al., 2025). Public sector organisations, such as the UK Civil Service, are often perceived as risk-averse and slow to respond to change, which further inhibits the adoption of transformative technologies despite staff’s personal readiness (Smith et al., 2025). Within the management layer, a generational gap in trust exists where 68% of senior leaders raised in traditional business environments remain sceptical of algorithmic decisions, preferring to rely on their own “gut feeling” and experience (Sulaiman et al., 2025).

Additionally, employees may perceive AI as intrusive, which can diminish trust and reduce engagement if transparency regarding the system’s purpose is lacking (Smith et al., 2025; Varma et al., 2024). This resistance is often compounded by an attitudinal “deskilling” in which workers feel treated as objects to be acted upon (“patiency”) rather than as autonomous agents, potentially leading to the atrophy of professional skills (Varma et al., 2024). Successful implementation, therefore, requires addressing organisational change management and targeted upskilling rather than just technical deployment (Smith et al., 2025; Roppelt et al., 2025).

Theme 3: Ethical and Technical Risks

This theme captures the inherent risks and moral dilemmas associated with applying machine learning and algorithmic logic to sensitive human domains, encompassing both technical failures in systems and the erosion of ethical standards in people management.

The “garbage in, garbage out” (GIGO) syndrome, where inconsistent data standards, missing fields, or poor-quality datasets erode the reliability, validity, and accuracy of AI-driven predictive models. A critical technical barrier to effective AI integration is the reliance on raw data that may be fragmented or inaccurate. The phenomenon of “garbage in, garbage out” remains a pervasive concern; if automated processes utilise erroneous inputs, they will efficiently reach entirely incorrect conclusions regarding candidate suitability or workforce forecasts (Kingsley, 2025; Smith et al., 2025). Organisations often face challenges with inconsistent labelling and siloed systems that hinder model accuracy (Kingsley, 2025; Roppelt et al., 2025). In emerging markets like Pakistan, this is exacerbated by a lack of standardised documentation, such as non-standardised experience letters or bilingual resumes that current systems struggle to interpret accurately (Sulaiman et al., 2025). Furthermore, when predictive algorithms are trained on small or non-representative datasets, they risk producing results that are both inaccurate and inherently biased (Roppelt et al., 2025).

The emergence of negative consequences and systemic vulnerabilities following adoption, including cybersecurity threats, privacy violations, and the potential for technological tools to dehumanise human resource processes. Beyond initial implementation, organisations must navigate “Harmful Forms of Practice” (HFP) that can manifest as data privacy breaches, information leaks to third parties, and system downtimes (Roppelt et al., 2025). Fears regarding unauthorised access to sensitive performance data and the potential for “surveillance-like” analytics often erode stakeholder trust and adversely affect adoption intentions (Kingsley, 2025; Roppelt et al., 2025). Beyond technical security, there is a profound risk of dehumanisation within the HR function. Over-reliance on predictive algorithms can subordinate worker dignity, treating humans as objects to be acted upon (“patiency”) rather than autonomous agents (“agency”) (Varma et al., 2024). This is particularly critical in processes requiring high empathy, such as candidate rejection or professional development coaching; replacing human discretion with “black box” algorithms risks alienating users and undermining the core human-centric values of talent management (Varma et al., 2024; Aguinis et al., 2024). Therefore, human-in-the-loop systems remain essential to mitigate the risk of treating human capital as a mere commodity (Varma et al., 2024; Roppelt et al., 2025).