The automation paradox: Efficiency gains vs. workforce concerns in human-AI collaboration

Human-AI collaboration (HAC) has attracted growing attention in operations and supply chain management due to its potential to enhance decision-making and organisational performance. Despite this promise, automation preferences remain uneven across organisations. This study examines preferences for full, partial, and no automation using a discrete choice experiment. Data were collected from 398 managers, generating 1592 choice sets and 4776 alternative-level observations. Alternative-specific conditional logit results show that automation preferences are primarily shaped by a trade-off between performance gains and workforce concerns. Operational efficiency and accuracy significantly increase the likelihood of selecting automation, whereas job insecurity exerts a strong negative influence. Anthropomorphism generates positive but context-dependent utility, while social interaction and investment cost are not statistically significant in the overall model. Managers prefer collaborative human-AI arrangements, choosing partial automation in 55.7% of scenarios and selecting no automation in only 3.4% of choice scenarios, indicating that the central decision concerns the degree of automation rather than its adoption. Further subgroup analyses reveal context-specific patterns across gender, sector, and organisational size, indicating subgroup-specific associations between socio-technical attributes and automation preferences. By identifying subgroup-specific patterns in the utility of AI attributes, this study extends socio-technical systems and contingency perspectives on technology adoption and highlights the need for context-sensitive automation strategies that balance efficiency gains with workforce security.