Employee intentions steer AI adoption in Chinese apparel factories

Artificial intelligence is reshaping manufacturing across China, yet its arrival in the labour-intensive apparel sector has been slowed by limited digitalisation and relatively low technological readiness. A new study of 200 small and medium-sized Chinese apparel manufacturers now shows that the path into AI runs through the workforce itself: employees’ stated willingness to use AI tools acts as a bridge between individual perceptions and the firm-level decision to adopt the technology. The research, published in Fashion and Textiles, combined two established frameworks—the unified theory of acceptance and use of technology (UTAUT) and the technological-organisational-environmental (TOE) framework—to model how personal and organisational forces jointly shape adoption decisions.

The study, led by Chen Qu and Eunyoung Kim, collected paired survey responses from 200 managers and 200 technicians representing 200 Chinese apparel manufacturing firms. Each manager answered questions about their organisation’s adoption decision, while a technician from the same firm reported on their own intentions to use AI tools. This paired design allowed the researchers to examine employee-level responses alongside firm-level organisational and environmental conditions within the same companies. The resulting model was evaluated using partial least squares structural equation modelling, a statistical technique well suited to testing networks of hypothesised relationships among latent constructs.

The headline finding concerns mediation. When the researchers traced the pathways through employees’ behavioural intentions, three indirect effects emerged as statistically significant. Performance expectancy—the degree to which a user expects a technology to improve job performance—showed a significant negative indirect effect on organisational AI adoption decisions (β = −0.081, p = .037). Peer influence, by contrast, exerted a significant positive indirect effect (β = 0.232, p = .002), and regional context produced the strongest indirect effect of all (β = 0.501, p < .001). For the organisational adoption decision itself, the model accounted for 71.4% of the observed variance, a substantial share for a field setting.

Several direct paths defied expectations. Performance expectancy had a significant but negative effect on employees’ behavioural intentions (β = −0.166, p < .05), the opposite of the positive direction hypothesised under UTAUT. Effort expectancy and facilitating conditions—employees’ perceptions of available training, technical support, and resources—showed no significant relationship with behavioural intention. The direct path from individual characteristics to behavioural intention was also not significant (β = −0.041, p = .683). Yet individual characteristics were strongly and positively associated with performance expectancy (β = 0.682, p < .001), effort expectancy (β = 0.793, p < .001), and facilitating conditions (β = 0.686, p < .001), suggesting that personal capability beliefs feed into downstream expectations even when they do not directly drive intention.

Peer influence stood out as the clearest positive force on the employee side. Encouragement and support from colleagues and supervisors showed a significant positive effect on technicians’ behavioural intentions (β = 0.480, p < .001). The authors interpret this as evidence that social dynamics within the workplace shape whether technical staff warm to AI tools, consistent with a broader research agenda on social and individual factors in employee AI adoption. On the organisational side, competitive pressure had a significant positive effect on adoption decisions (β = 0.315, p < .001), and dynamic capability—the firm’s ability to integrate, develop, and reconfigure resources in changing environments—also mattered (β = 0.193, p = .002). Behavioural competency, however, showed a negative but statistically insignificant relationship with adoption decisions (β = −0.064, p = .291), so that hypothesis was rejected.

Geography proved to be a powerful variable. With North China coded 0 and South China coded 1, region showed significant positive effects on both employees’ behavioural intentions (β = 1.036, p < .001) and organisations’ AI adoption decisions (β = 1.127, p < .001). The positive coefficients indicate higher model-predicted outcomes for the South China group after adjustment for the other predictors. The authors note that standardised coefficients for binary predictors can occasionally exceed 1.0 because of the standardisation procedure and the variance characteristics of dummy-coded variables, and they report that inner variance inflation factor values (1.707 and 1.457) did not indicate problematic multicollinearity. Regional context also operated through both direct and indirect paths in the same direction, a pattern the researchers describe as complementary mediation.

The sample itself carries important caveats. Most participating firms were located in South China—185 of the 200 firms, or 92.5%—with only 15 firms from North China, reflecting geographic concentration in areas such as the Yangtze River Delta and Pearl River Delta. The authors acknowledge that this regional imbalance is substantial and that a more balanced sample would help determine whether the regional associations persist across a broader range of locations. Firms were recruited purposively through the Alibaba.com manufacturer database and quota sampling was applied to reflect the database’s distribution of firm sizes and manufacturing models, with original design manufacturers accounting for 87% of the sample and medium-sized firms 62%.

The survey ran from April to May 2024. After two follow-ups, 232 manager responses and 223 technician responses arrived; after removing missing, suspicious, and inconsistent entries, the final dataset comprised 200 paired responses. Employee respondents were technicians with substantial experience—68% reported over 15 years of work experience, and all had at least five years. The study received institutional review board approval and collected no personally identifying information. Common method bias was examined using a marker-variable approach with general work tenure, and nonresponse bias was checked by comparing the first 100 responses with the last 100 using chi-square and t-tests, which revealed no significant differences.

Measurement quality was assessed with standard psychometric criteria. Cronbach’s alpha and composite reliability supported internal consistency, average variance extracted supported convergent validity, and both Fornell–Larcker comparisons and heterotrait–monotrait ratios below 0.85 supported discriminant validity. Structural diagnostics showed variance inflation factor values below 5.0 and a standardised root mean square residual of 0.070, below the 0.080 reference value. Hypothesis tests used 5,000 bootstrap resamples, and the model explained between 42.7% and 71.4% of the observed variance across its endogenous constructs, including 46.5% for performance expectancy, 62.9% for effort expectancy, 47.0% for facilitating conditions, and 42.7% for employees’ behavioural intentions.

The authors frame their contribution as a shift from purely organisational analyses toward an individual-level perspective in China’s apparel manufacturing industry. They caution that the cross-sectional design records intentions and adoption decisions at a single point in time and cannot establish temporal ordering or evolution, that the sample of 200 firms is modest relative to the full population of Chinese apparel manufacturers, and that the findings concern Chinese apparel manufacturing SMEs and should not be assumed to describe other sectors or national settings. The study also did not explicitly consider government policy and governance, employees’ education level, suppliers’ cooperation, or AI-related ethical concerns. Within those limits, the results position employees’ behavioural intentions as a measurable mediating mechanism—one that managers and policymakers can target through training, peer support, and region-specific strategies as AI spreads through apparel production.

Subject of Research: How employees' behavioural intentions mediate organisational AI adoption decisions in Chinese small and medium-sized apparel manufacturing enterprises

Article Title: Employees’ behavioural intentions as a mediator of artificial intelligence adoption: a UTAUT-TOE study of Chinese small and medium-sized apparel manufacturing enterprises

Article References: Qu, C., & Kim, E. (2026). Employees’ behavioural intentions as a mediator of artificial intelligence adoption: a UTAUT-TOE study of Chinese small and medium-sized apparel manufacturing enterprises. Fashion and Textiles, 13(1), Article 47. https://doi.org/10.1186/s40691-026-00497-4

Image Credits: AI Generated

DOI: 10.1186/s40691-026-00497-4

Keywords: artificial intelligence, apparel manufacturing, UTAUT, TOE framework, behavioural intentions, small and medium-sized enterprises, China, PLS-SEM, peer influence, regional context, technology adoption, Employees