The operations efficiency, service quality and resources productivity, are the core aspects of the call centres competitive advantage in massive market competition. Thus, subjective evaluation is the leniency, perception and bias in performance evaluation which impact the efficiency of the operations and leads to frustrated customers. The study aims to determine the subjective performance evaluation in call centres to get a more objective measurement. It can be achieved by identifying factors affecting resources performance evaluation through the development of a conceptual model to reduce or eliminate the effect of subjective factors contained in the performance evaluation. The research approach is based on quantitative methodology through cross-sectional self-reports for 224 participants’ work in eight outsource call centres located in Egypt. The research aims to determine the subjective evaluation factors biases the true performance. It is followed by a machine learning practical application using neural networks for auto-detection the subjective context in the recorded calls to be considered through the evaluation process. The key findings of the study are nine subjective factors out of fifteen that have a direct influence on subjective performance evaluation. The actual performance is the performance evaluation after eliminating the subjective performance. Two different methods have concluded the actual performance. The first method excludes the subjective factors from the resulting evaluation to determine the actual performance. The second method is a prediction model defining subjectivity percent as a call centre baseline for future performance evaluation. Furthermore, the study highlights the potential subjective variables and the degree of influence for each variable. The theoretical contribution is determining the subjective factor and proposing the model to measure and predict the subjectivity in the call centre. The study recommended a restatement for the resource-based theory considering the subjective evaluation effect on performance evaluation. The practical application contribution is based on automating the detection and prediction of subjectivity using a machine learning approach through cascaded Convolutional Neural Networks, which achieved 75% accuracy in classifying the subjectivity for two study constructs: agents and customer behaviour.