Artificial Intelligence In Human Resource Recruitment, Utilization, and Management: an Overview and Responsible Implementation Framework for Vietnamese Businesses

1. Introduction

The development of computer learning, natural language processing, and AI generation is changing how businesses conduct HR activities. In recruitment, AI can assist in drafting and distributing job postings, searching for candidates, screening resumes, scheduling appointments, interacting via chatbots, and analyzing evaluation results. In other activities, AI is used to support HR policy formulation, work scheduling, skill needs identification, training content recommendations, performance analysis, and providing information for human resource planning (Tambe et al., 2019; Berg & Johnston, 2025; Tusquellas et al., 2024).

Human resource management (HRM) doesn't just focus on a high-potential group of employees. This field encompasses policies and processes applied to all employees throughout their employment, from recruitment, job placement, training, evaluation, and compensation to maintaining labor relations. Therefore, research on AI needs to fully consider all HR functions and decisions that directly impact employees.

The technical capabilities of AI do not always produce the expected managerial effectiveness. Human resource issues are influenced by many factors, such as job characteristics, organizational relationships, individual motivations, and internal regulations. Meanwhile, enterprise data may be scarce, inconsistent, or not fully reflect all employee groups. Therefore, a good predictive model based on outdated data may still produce unfavorable results for some groups or become outdated as jobs and workforces change (Tambe et al., 2019; Köchling & Wehner, 2020; Köchling et al., 2021).

In Vietnam, the National Strategy on AI Research, Development and Application until 2030 provides direction for promoting the application of the technology (Prime Minister, 2021). The Law on Personal Data Protection and its implementing guidelines, effective from January 1, 2026, set requirements for the collection, use, sharing, and storage of data of candidates and employees (National Assembly, 2025; Government, 2025). However, empirical research in Vietnam is still limited. The study by Tri Minh Cao and Loc Thi Vy Nguyen (2025) focuses on the decision to apply AI in recruitment at medium-sized enterprises, while Vietnamese-language articles mainly present a general overview of applications and propose implementation directions (Khong Van Hai, 2025; Pham Hong Long et al., 2024).

The article has three goals: (1) systematize the main application groups of AI in recruitment and human resource management; (2) analyze benefits, risks and factors affecting implementation effectiveness; and (3) propose a control framework suitable for Vietnamese businesses. The article connects research on assessment quality, algorithmic fairness, candidate and employee reactions, data governance, and organizational capacity to build a responsibility management process throughout the lifecycle of an AI system.

2. Research Methodology

This article uses a structured literature review method to select, analyze, and synthesize relevant studies. This method is used to clarify key issues and build a framework for responsible AI deployment. The study does not perform a pooled analysis and does not claim to have covered all published works on this topic.

The review process consists of three steps. First, documents are searched using keyword groups related to artificial intelligence, human resource management, recruitment and selection, performance management, employee development, candidate response, algorithmic fairness, and AI governance. Second, information about the author, year of publication, journal or institution name, DOI, or official publication address is checked on the publisher's or issuing authority's website. Third, only documents whose content directly supports the claims in the article are selected.

The review scope includes documents published from 2019 to July 2026. After considering relevance and verifying publication information, 19 documents were selected for analysis, including 12 international scientific articles, two articles on the Vietnamese context, a research report from the International Labour Organization (ILO), a governance framework from the US National Institute of Standards and Technology (NIST), and three official Vietnamese documents. The documents were analyzed according to four content groups: AI application areas; benefits; risks and stakeholder responses; and governance mechanisms. This classification helps to clearly distinguish research results from descriptive assessments and governance recommendations.

3. Applying Ai In Recruitment And Human Resource Management

3.1. Recruitment and selection of personnel

AI can support many steps in the recruitment process. In the candidate acquisition stage, the system can analyze job descriptions, search resumes, and distribute recruitment information. In the screening stage, natural language processing tools can match resumes to the position requirements; chatbots assist in gathering information and scheduling appointments. In the evaluation stage, AI can assist in grading tests or standardizing certain interview questions. Hunkenschroer and Luetge (2022) argue that these applications can speed up and improve the consistency of the process, but at the same time raise issues of privacy, discrimination, transparency, and accountability.

Automation does not equate to objectivity. Models may relearn unfair decisions from the past, use indirect indicators related to sensitive personal characteristics, or optimize a criterion that does not fully reflect job requirements. Köchling and Wehner (2020) point out that biased, incomplete, or inaccurate data reflecting applicant groups can perpetuate discrimination. Research on video analytics in recruitment also shows that the overall accuracy of the model is insufficient to demonstrate that the results are fair to all groups (Köchling et al., 2021).

Candidate response is a key consideration when evaluating recruitment effectiveness. Empirical studies suggest that AI-powered processes may be rated lower by candidates in terms of fairness, interactivity, and relevance compared to human-powered processes (Acikgoz et al., 2020; Mirowska & Mesnet, 2022). Clearly explaining the purpose and use of AI can improve acceptance, but effectiveness depends on the content and delivery. Providing excessive technical details does not necessarily build trust with candidates (Langer et al., 2021; Köchling & Wehner, 2023).

3.2. Human resources and operations support

Beyond recruitment, AI is also used to answer repetitive employee questions, guide internal processes, assist with shift scheduling, and inform decisions about compensation or performance evaluation. Berg and Johnston (2025) analyzed the use of AI in four activities: recruitment, compensation, work scheduling, and performance management. This study showed that risks stem not only from algorithms but also from how businesses define goals, select data, and translate management objectives into measurable indicators.

AI typically offers clearer benefits for tasks with stable processes, clear data sources, and correctable errors. Conversely, when systems are used for job allocation, compensation recommendations, or disciplinary decision support, businesses require a higher level of reliability, explainability, and human oversight. Therefore, the same level of automation should not be applied to simple administrative tasks and decisions that significantly impact employee rights.

3.3. Training, performance evaluation, and workforce retention.

In training and development, AI can support the building of skill profiles, identify competency gaps, suggest learning content, and personalize development paths. Tusquellas et al.'s (2024) overview notes applications in training, skill demand forecasting, and identifying turnover patterns. Pereira et al. (2023) showed that the impact of AI on workplace outcomes depends on organizational processes, the level of analysis, and how employees interact with the technology.

Models that predict performance, engagement, or turnover only show correlations within existing data; they do not self-identify causes and do not replace consideration of specific circumstances. Labeling an employee as “high risk” can negatively impact their training or promotion opportunities. Therefore, model results should only be considered as information for further investigation, not as definitive conclusions about an individual.

The competence of HR professionals is a crucial condition when implementing AI. Deepa et al. (2024) emphasize that HR managers need to combine technical knowledge with managerial understanding and interpersonal skills. Businesses will find it difficult to create sustainable value if they only invest in software without developing the ability to identify problems, interpret data, assess errors, coordinate with technology and legal departments, and explain decisions to affected parties.

(Source: Compiled by the author from the documents listed in the table)

4. Benefits, Risks, And Factors Affecting Implementation

4.1. Benefits and conditions for creating value

First, AI can help HR departments handle high-volume tasks with relatively clear criteria, such as request classification, answering repetitive questions, assisting with resume screening, or suggesting learning content. The benefits lie not only in reducing manual work but also in the ability to implement consistent processes and track progress. To achieve this, the system needs to record input data, model versions, and instances of user adjustments to the results.

Secondly, AI can assist in analyzing the skills and needs of individuals, thereby helping businesses design training programs and allocate resources more appropriately (Tusquellas et al., 2024). However, personalization is only meaningful when the data accurately reflects job requirements and employees can check, correct, or add information about themselves.

Third, properly designed and tested tools can reduce some of the discrepancies caused by inconsistent human evaluation. However, AI does not automatically eliminate bias. Biases can still arise from the system's objectives, data labeling, input information, and rules for generating results.

4.2. Main risks

Relevance and fairness. Risks can stem from how desired outcomes are defined, historical data, labeling methods, indirect indicators, the representativeness of each group, or job variability. Businesses need to examine the relevance of outcomes to job requirements, overall error, and disparities between groups. A single accurate indicator is insufficient to assess the quality of the system.

The issue of interpretability and vendor dependence. When using a ready-made solution, businesses may not fully understand the training data or how the model operates. Transparency doesn't necessarily mean publicly releasing the source code, but businesses need to clearly understand the intended use, input data, system limitations, evaluation criteria, fair testing methods, update procedures, and responsibilities for handling errors.

Experience and acceptance of the decision. Candidates and employees may react negatively if they are unaware of where AI is used, lack the opportunity to communicate with the person in charge, or cannot request a review. This can impact their intention to reapply, their trust in the decision, and their acceptance of it (Acikgoz et al., 2020; Köchling & Wehner, 2023).

Personal data and the risk of over-monitoring. Candidate profiles, work history, evaluation results, communication content, behavioral data, and model-generated judgments can affect privacy. In Vietnam, processing procedures need to comply with the Law on Personal Data Protection and Decree No. 356/2025/ND-CP. Businesses need to clearly define the purpose of use, who has access rights, the retention period, the responsibilities of the data processor, and how data subjects exercise their rights (National Assembly, 2025; Government, 2025).

There is a tendency to over-rely on system results. AI-generated scores or rankings can lead users to believe the results are more certain than they actually are. Human oversight is only meaningful when decision-makers have sufficient information, time, capacity, and authority to question, modify, or reject the system's recommendations.

4.3. Factors affecting implementation effectiveness

AI is better suited to tasks with clear objectives, reliable data, verifiable evaluation criteria, and rectifiable errors. Risks increase when the system attempts to infer subtle traits, reuses data for other purposes, or automatically makes decisions that significantly impact job opportunities and benefits. Therefore, chatbots that answer internal questions and automated candidate screening models require varying levels of control.

Organizational readiness also influences outcomes. Research at medium-sized enterprises in Vietnam shows that perceived value and readiness of the human resources department are related to the decision to adopt AI (Cao & Nguyen, 2025). Therefore, before purchasing or expanding a system, businesses need to assess data, human resource capabilities, workflows, organizational culture, and accountability mechanisms.

5. Context And Impact On Vietnamese Businesses

Decision No. 127/QD-TTg establishes the development and application orientation of AI until 2030 (Prime Minister, 2021). In human resource management, this orientation creates opportunities for businesses to experiment with automation tools and data analysis. However, implementation needs to be placed within the labor management system, data protection, and internal control, rather than being considered a separate technology project.

The Law on Personal Data Protection No. 91/2025/QH15 and Decree No. 356/2025/ND-CP will take effect from January 1, 2026. Because personnel data is often linked to the identity, work history, performance evaluations, and behavior of individuals, businesses need to clearly define the roles of the parties processing the data, the purpose of use, access rights, retention periods, and how to handle requests from data subjects (National Assembly, 2025; Government, 2025).

Direct research in Vietnam on this topic is limited. Cao and Nguyen (2025) provide quantitative evidence on factors influencing the decision to adopt AI in recruitment at medium-sized enterprises. Two domestic articles present the applications, opportunities, and limitations of AI in human resource management (Khong Van Hai, 2025; Pham Hong Long et al., 2024). However, these documents are insufficient to establish causal relationships or generalize to all Vietnamese businesses.

For small and medium-sized enterprises (SMEs), a cautious approach is to start with narrow-scope use cases, clear data sources, and measurable results, such as assisting with policy lookups, requirements classification, or training content suggestions. Applications like automated candidate screening, interview scoring, turnover forecasting, compensation recommendations, or promotion decision support have a larger impact and require rigorous testing.

The contract with the vendor should clearly define the scope and purpose of data processing, storage location, inspection rights, responsibility for notifying when the model changes, quality criteria, complaint resolution and troubleshooting support, and the obligation to delete or return data upon contract termination. Purchasing an AI solution does not transfer all responsibility for personnel decisions to the vendor.

6. Framework For Deploying Responsible Personnel In Human Resource Management

The proposed framework is based on the NIST AI RMF's continuous risk management approach, comprising four main activities: governance, context identification, measurement, and risk management. The framework is adapted to suit human resource decisions (Tabassi, 2023). The six components below constitute the control process from purpose selection, deployment, operation to decommissioning of the system.

(Source: The author's proposal is based on Tabassi (2023), Hunkenschroer and Luetge (2022), Berg and Johnston (2025), and studies on candidate responses).

For high-impact decisions, businesses should adopt the principle of "AI makes suggestions, humans make decisions." This principle is only meaningful when the person in charge has sufficient information and authority to review or reject the outcome. For low-risk administrative tasks, a higher level of automation can be achieved. The system needs to be evaluated before deployment, after significant changes in data, models, or usage, and periodically as appropriate to the level of risk.

Businesses also need to consider the option of not using AI. When data is of insufficient quality, job requirements are unclear, results are inexplicable, or there are no corrective measures if errors occur, postponing deployment may be more appropriate than adopting the technology simply because it's trending.

7. Research Gap And Future Directions

First, more on-site and long-term research is needed to assess the impact of AI on recruitment quality, performance, skill development, trust, labor relations, and turnover rates. Studies based on hypothetical scenarios or perceptions cannot replace data collected after the system is implemented in practice.

Secondly, research needs to differentiate between model accuracy, job suitability, disparities in results between groups, participants' perceptions of fairness, and process legitimacy. A system might improve one criterion but degrade the quality of another. Therefore, evaluation criteria need to be linked to the impact on both the business and individuals.

Third, the influence of culture, types of work, and the legal environment needs to be examined within the Vietnamese context. It should not be assumed that research findings from developed economies can be applied indiscriminately. The level of AI acceptance may vary depending on trust in the organization, expectations regarding communication, experience using the technology, and the ability to protect employee rights.

Fourth, AI generation raises new issues regarding data sources, content accuracy, the risk of insider information leaks, and the use of unverified results to evaluate humans. Future research needs to develop governance criteria tailored to each specific use case, rather than simply applying a single set of general principles.

8. Conclusion

AI is expanding the automation and analytical capabilities in recruitment and human resource management. However, the technology only creates value when deployed with clear objectives, relevant data, carefully designed processes, and clearly defined human responsibilities. Existing research shows that AI can help increase processing capabilities, improve consistency, and support the personalization of certain activities. At the same time, the overall accuracy of the model does not guarantee fair, understandable, or acceptable results for candidates and employees.

For Vietnamese businesses, the appropriate roadmap starts with clear intended use, readiness assessment, data governance, and the application of controls commensurate with the level of impact. The six-component framework helps translate principles of responsible AI into control questions, documentation to be maintained, and responsible departments. Sustainable benefits only materialize when AI supports human judgment, rather than obscuring or replacing managerial responsibilities.