Artificial intelligence (AI) has become an important aspect of recruitment with automated decision-making systems operating to screen, rank and select candidates using prediction algorithms. In this respect, Type II errors (also referred to as 'false reject errors') happen when eligible job applicants are denied an opportunity to work in an organisation. Although AI systems can be specifically developed to be efficient and accurate, little is known about the cost of such false negatives. This paper is conceptual and analytical as it uses the Statistical Decision Theory and Signal Detection Theory to investigate the cost of false-reject errors. These models bring out the fundamental trade-off between accuracy and recall in classification systems, in which high rates of accuracy would result in the high possibility that good candidates are excluded. The results show that this type of optimisation favouritism is a serious threat defrauding talent output and organisational functionality. In addition, once training data trained through algorithms causes bias, it worsens the risks of exclusion. The research concludes that to reduce the occurrence of the hidden losses, equitable hiring, balanced model calibration, and human oversight are necessary as governance mechanisms...