The study includes an examination of technology adoption models such as the Technology Acceptance Model (TAM) which emphasizes the significance of perceived ease of use and perceived usefulness in determining an individual’s attitude and intention to use a particular technology, the Theory of Reasoned Action (TRA) by (Ajzen & Fishbein, 1980) focuses on the initial acceptance of new products, considering the role of individual attitudes and subjective norms in predicting behavioral intentions, The Technology Organization Environment Model (TOE) by (Tornatzky & Fleischer, 1990) underscores the influence of technological, organizational, and environmental factors on technology adoption within organizations. The Innovation Diffusion Theory (IDT) is also highlighted, emphasizing the spread of new ideas, products, or technologies within a social system and identifying different adopter categories and factors influencing the rate of adoption by (Rogers, 1985) . By considering Tanzanian context of the level of technology adopted and other researchers such as AI within HRM (Kaur et al., 2021) adaptation of BA (Horani et al., 2023) , adoption of technology (Kumar et al., 2022) , adoption of mobile banking in SMEs (Mujahed et al., 2021) , big data adoption. (Agarwal, 2022) described TOE as a “generic” theory applicable to analyzing elements impacting AI adoption in HRM, emphasizing its relevance in assessing how technology take place in fulfilling HR responsibilities.

This study proposes a framework rooted in the TOE model to explore determinants to AI adoption in HRM within MEs since it’s more based on dimensions within Organization, Technology and Environmental factors. The findings of this research study provide useful perspectives into the determinants of AI adoption and its impact on HRM effectiveness in medium-sized enterprises in Tanzania. The supported hypotheses highlight the benefits of relative advantage, compatibility, competitive pressure in the adoption of AI and it shows the valuable impact of adopting and implementing AI in stimulating HRM effectiveness (as shown in Figure 1 ).

Relative Advantage and Artificial Intelligence.

It refers towards the acknowledged superiority of a technology in comparison to other current company innovations, as well as the expected advantages, such as operation (Kurup & Gupta, 2022) . Adopting new technologies might fail due to perceived complexity and difficulty (Almaiah et al., 2022) . (Boonsiritomachai et al., 2016) indicates that technological adoption begins with relative advantage and companies that see technologies as superior to their current procedures and processes are more likely to implement it effectively. Hence,

H1. Organization factor of Relative advantage influence the adoption of AI.

Compatibility and Artificial Intelligence.

It is a degree of consistency with the firm’s present state which may be measured based on the firm’s innovation adoption. Compatibility is a key factor in driving technology adoption and influencing AI decision-making and organizations may improve their procedures and policies to promote favorable compatibility and AI adoption (Chong & Lim, 2022) . Compatibility increases the option AI advancement and adoption (Neumann et al., 2022) . Factors related to compatibility include technical compatibility and commitment (Jöhnk et al., 2021) . Aligning AI technologies with technological systems can help organizations leverage new technology and reduce uncertainty in the HR function regarding its adoption (Gangwar, 2018) . Hence,

H2. Organization factor of Compatibility influence the adoption of AI.

Complexity and Artificial Intelligence.

Complexity discusses the degree of difficulties, challenges and risks that HRM face in adopting and implementing AI technologies (Jöhnk et al., 2021) . Successful AI adoption requires coordinated activities and simplicity of technological functions across the organization towards AI implementation (Xu et al., 2023) . The flexibility of utilizing a system is connected to its applicability or accessibility, as a result, the ease in which technology can be implemented and used will encourage the company to adopt it. Complexity in technology adoption may be measured by job completion time, process integration, data processing efficiency, system functionality, and interface design. Factors such as the use of technology and automation in HRM might impact the outcome (Almaiah et al., 2022) . Hence,

H3. Organization factor of Complexity influence the adoption of AI.

Security and Privacy and Artificial Intelligence.

The perceived risk of using technology for work and data transmission (Jöhnk et al., 2021) . Technology adoption relies heavily on security and privacy considerations. Yabanci (2019) indicates that adopting technology raises concerns regarding security and privacy, particularly with regard to the potential risks to confidentiality, trustworthiness, and accessibility in human-computer interaction. These reflect the level in where AI is considered to be not safe for conducting work and exchange (Kaur et al., 2021) . Hence,

H4. Organization factor of Security/Privacy directly influence the adoption of AI.

Top Management and Artificial Intelligence.

This is very vital for the effective implementation of AI in HRM. It is critical in implementing any significant change to an organization (Lutfi et al., 2022) . Support from top management is evident in the way they assess and frame the importance of technological innovation in creating value for businesses. In order to handle change-related obstacles and any opposition to the implementation of new technologies, this support guarantees the commitment of resources and fosters a constructive environment. The implementation of technologies in HRM requires the backing of upper management. Hence,

H5. Technology factor of Top management support directly influence the adoption of AI.

Organization Readiness and Artificial Intelligence.

This includes processes, structures, and resources that encourage the use of technology, including automation and AI (Alam et al., 2016) . In an organization, HR professional’s should conduct an assessment of how ready they think the organization is in terms of the level of knowledge, loyalty, resource allocation and successful implementation of adopting a technology (Chong & Lim, 2022) . To be efficiently deployed in enterprises, AI in HRM requires effective infrastructure, the necessary HRM knowledge, and financial allocation of resources (Pillai & Sivathanu, 2020) . Organizational technological preparation is thought to influence technology implementation. In order for businesses to use AI to enhance their HRM process, they must retrain and restructure their staff. Hence,

H6. Technology factor of Organization Readiness directly influence the adoption of AI.

Competitive Pressure and Artificial Intelligence.

It discusses the perceived pressure a company feels from its competitors (Almaiah et al., 2022) . Businesses must assess, derive lessons from past experiences, and come up with creative solutions to challenging issues in the global economy (Shet et al., 2021) . More efficient and economical use of resources can lead to better human resource management (Bhatiasevi & Naglis, 2018) . Market competition pushes firms to implement human resource technologies (Alam et al., 2016) . When several firms adopt different AI systems in their HRM activities, they gain a strategic advantage over their competitors. This creates pressure and compels the corporation to adopt the technical upgrade (Nguyen et al., 2022) . Hence,

H7. Environmental factors of Competitive pressure directly influence the adoption of AI.

External Support and Artificial Intelligence.

Third parties provide guidance for business enterprises in adopting technology solutions including consulting, training and technical assistance in overcoming problems related to the adoption of new technology (Horani et al., 2023) . Since AI is termed as a relatively new product, the positive involvement of AI suppliers throughout the introduction process is an important element affecting its acceptance (Malik et al., 2021) . For soft adoption, the supplier is expected to offer everything needed throughout every stage, including training and post-implementation assistance (Singh & Pandey, 2024) . Depending on the organization’s unique demands, the HRM function may need a supplier to create and implement customized AI solutions. Hence,

H8. Environmental factors of external support directly influence the adoption of AI.

Government Support and Artificial Intelligence.

Government rules and regulations, such as motivations, technical requirements, and legislation, may alternatively promote or restrict technology adoption (Horani et al., 2023) . Government restrictions may encourage or prevent firms from implementing technology breakthroughs (Dincbas et al., 2021) . Financial assistance, necessary tools, and tax incentives can all help to encourage adoption (Chong & Olese, 2017) . Hence,

H9. Environmental factors of Government support directly influence the adoption of AI.

Artificial Intelligence and HRM effectiveness.

The implementation of AI presents the field of HRM with a number of advantages and prospects (Agarwal, 2022) . AI can increase satisfaction with work and retention rates, quicken hiring, enhance onboarding for staff, and offer individualized opportunities for professional growth (Meister, 2023) . Through the analysis of several variables including job happiness, work-life balance, and career advancement, AI can also assist HR managers in creating strategies that will effectively draw in and keep top applicants (Visier, 2013) . AI can give HR directors insight into the knowledge and abilities that will be in high demand, enabling them to plan ahead for their hiring needs (Visier, 2013) . However, HR professionals must be conscious about some of the weaknesses of using AI in HR, such as AI bias during the hiring process, which can lead to major consequences (Heinze, 2023) .

H10. The Artificial intelligence adoption has a significant impact on effective HRM in the medium enterprises in Tanzania.

To create a questionnaire aligned with hypotheses, the study referred to relevant questionnaire from different literatures, which is presented in Table 1 including name of the variables, item of measurement from questionnaire, and corresponding references.

Continued

Source: Author’s analysis.

A structured questionnaire survey in English was utilized to examine the hypothesized connection indicated in the model as Figure 1 indicates. It intends to evaluate the determinants of AI adoption in HRM within Tanzanian MEs and its impacts on effective HRM. The total number of respondents of this study is based on (Hair et al., 2013) calculation, which is 5 - 10 times the number of questionnaire items. The number of items included in this research questionnaire was 37 items, this study chose the multiplication of 5, so the required population is 37 items × 5 = 185 respondents includes both HR Profesionals IT profesionals and CEO from different sectors. The questionnaire incorporated a Likert scale, specifically utilizing a 5 point scale (1 “strongly disagree” to 5 “strongly agree”) (Goswami et al., 2023) . For the accuracy of data collection, the study collects 200 total number of respondents, in where after the deletion of incomplete questionnaires only remain with 185 respondents. To assess the response rate, the study, calculated the percentage of people who completed and returned the questionnaire out of the total number of people contacted which is the number of completed questionnaires (185)/Number of contacted people (200) × 100 where the total response rate was 92.5% which is efficient according to (Fowler Jr., 2009) .

This study is based on the individual who belongs to a group specifically, human resource professionals, IT professionals and CEOs from medium enterprises. According to (Sekaran & Bougie, 2016) , the unit of analysis can be at the individual, dyad, and group levels. The present study comprises a diverse population of the HR, IT managers and CEOs. The sample in this study was taken from different sectors in the category of medium enterprises which are categorized by considering the number of employees starting from 50 - 99 (Ministry of Industry and Trade, 2003) . The sample is part of the total population selected for research, and the researcher can draw/generalize conclusions about the entire population, so the sample must represent the population (Tabachnick et al., 2013) . It must be considered that there are two aspects determined in the sample: the number of samples (sampling size) and the sampling technique. The number of samples in this study is based on (Hair et al., 2013) calculation, which is 5 - 10 times the number of questionnaire items. Number of items in this study questionnaire was 37 items. The technique of sampling for the present study applied purposive and convenience sampling rather than probability sampling (Cooper & Schindler, 2014) , Purposive sampling aims to discover information-rich scenarios that can give the most useful information for achieving the study’s goals which divides the population into strata depending on important factors such as experience, (Campbell et al., 2020) .

In the hypothesis testing phase of research, inferential statistical analysis plays a crucial role (Sekaran & Roger, 2016) . In the context of this study, inferential analysis is conducted using Structural Equation Modeling (SEM)-PLS with the assistance of SmartPLS 4 software. Evaluating PLS-SEM involves assessing the outer model, inner model, and hypothesis testing, as outlined by (Maroufkhani et al., 2022) .

For the outer model, reflective model evaluation includes criteria such as indicator reliability, discriminant validity, and internal consistency. Indicator reliability is assessed with outer loadings ideally between 0.4 and 0.7. Indicators within this range may be considered for removal if their deletion improves composite reliability or average variance extracted (AVE) above the threshold, though their contribution to content validity must also be considered (Hair et al., 2014) . Discriminant validity ensures that indicators of a construct are not highly correlated with those of other constructs. This is evaluated through cross-loading tests and the Fornell-Larcker criterion, where indicators should load higher on their respective latent variable than on others, and each construct’s AVE should ideally exceed its correlations with other latent variables (Agarwal, 2022) . Internal consistency is tested using composite reliability and Cronbach’s alpha, with a preferred composite reliability limit of 0.6 (Hair et al., 2019) . Composite reliability does not assume equal weighting for each indicator, whereas Cronbach’s alpha does (Hair et al., 2014; Hair et al., 2011) . These criteria ensure the robustness and validity of the PLS-SEM analysis in hypothesis testing within HRM research contexts.

In the inner model analysis, the focus is on the accuracy of the structural model. Multicollinearity is assessed using the Variance Inflation Factor (VIF) with a low Tolerance value (below 0.10) or a high VIF value (above 10) indicating problematic multicollinearity. The coefficient of determination (R²) measures the extent to which an exogenous construct explains an endogenous construct, with higher R² values indicating a stronger model. Path coefficient values range from −1 to +1, indicating the strength and direction of relationships between constructs, with values closer to +1 denoting strong positive relationships and values closer to −1 indicating strong negative relationships (Maroufkhani et al., 2022) . These metrics, along with additional criteria, are essential for evaluating the overall structural integrity and explanatory capacity of the model, ensuring its reliability and validity in HRM research.

The 200 responses were received from different sectors including Manufacturing, food processing, Mining, Agriculture, Education, Financial, Entertainment, Hospitality, Health, Law aid, ICT and Turism sector. After the elimination of incomplete and invalid questionnaires, only 185 responses were finalized for data analysis. Table 2 shows the demographic results.

Source: Author’s analysis.

In the evaluation of the results derived from the PLS SEM, the initial phase includes investigating of the measurement model as suggested by (Hair et al., 2014) . It involves a thorough examination of the indicator reliability (Hair et al., 2013) . The reliability of an item is confirmed by factor loadings that exceed 0.5, which also implies that the construct explains 50% of the variance (Hair et al., 2014) . In the context of this study, all item loadings are above the threshold, thereby, confirm item reliability ( Table 3 ).

Source: Author’s analysis.

Positioning The discriminant validity assessment aims to determine whether the indicators of a construct exhibit low correlation with indicators from other constructs (Hair et al., 2019) . This evaluation was conducted using two methods namely Heterotrait–Monotrait (HTMT) criterion and Fornell-Larcker criterion. HTMT is a comparison of the mean of the heterotrait-heteromethod correlations to the geometric mean of the monotrait-heteromethod and monotrait-monomethod correlations, In terms of acceptance criteria, an HTMT value of less than 0.85 is generally considered to indicate adequate discriminant validity (Agarwal, 2022) as Table 4 .

Source: Author’s analysis.

The subsequent discriminant validity valuation employs the Fornell-Larcker measure, which involves comparing the Average Variance Extracted (AVE) values with the relationship between constructs. This comparison is depicted in Table 5 . According to the Fornell-Larcker criterion, each latent variable’s AVE value should exceed the correlations with other latent variables, ensuring discriminant validity.

Source: Author’s analysis.

Positioning the following step in the assessment of internal consistency reliability is the calculation of composite reliability, with the acceptance threshold set between 0.6 and 0.7. The current study exhibits Composite Reliability (CR) values that range from 0.949 to 0.872, thereby indicating a high degree of reliability as suggested by (Hair et al., 2011) . Furthermore, Cronbach’s alpha, another measure of composite reliability, is employed to measure the internal consistency of the constructs utilized. The acceptable level for Cronbach’s alpha is set at 0.7 (Hair et al., 2019) . As per Table 3 , all constructs have higher Cronbach’s alphas that exceed the 0.7 level, thereby implying that these measures possess high reliability and are suitable for measuring each of the stated constructs.

Additionally, the following step in the evaluation of the measurement model is the assessment of the convergent reliability of each construct measure (Hair et al., 2019) . The convergent validity can be assessed using the Average Variance Extracted (AVE) value, which indicates a high correlation between the items and the factor, and confirms their belonging to the construct. The minimum acceptable level for AVE is set at 0.5 (Hair et al., 2011; Agarwal, 2022) . As per Table 6 , all constructs formed in this study have high AVE values that exceed 0.5, thereby indicating that the construct elucidates at least 50% of the variance of its items.

Source: Author’s analysis.

After confirming the adequacy of the measurement model, the study evaluates the structural model which encompasses several key metrics, including the coefficient of determination (R square), and their statistical significance, as outlined by (Hair et al., 2019) . The outcomes of these assessments are detailed in Table 7 and Table 8 . The model demonstrates no collinearity, as evidenced by Variance Inflation Factor (VIF) values under 3 (Agarwal, 2022) . The model also displays a respectable predictive capability, with an R-square value nearing 0.5, which is perceived moderately high by (Hair et al., 2011) .

Source: Author’s analysis.

Source: Author’s analysis.

Table 9 shows hypothesis testing results of this study based on the t values and p values that was obtained after the calculation.

Source: Author’s analysis. Note: *p < 0.10, **p < 0.05, and ***p < 0.01 (two tailed test).