A two-phase multiobjective optimization for benchmarking and evaluating service quality in banks

Service quality is an evaluation of how well a delivered service meets customers’ expectations. The purpose of this paper is to provide a reliable scale of measurement for service quality in banks.

The SERVQUAL model was adopted based on a Banking Service Quality (BSQ) model and a two-phase multiobjective optimization model was designed. A structured questionnaire with five-point Likert scale was administered with a 93 percent response rate of 270 sample size. A total of 22 variables were considered based on the BSQ model and the significance of these variables to customers’ satisfaction were investigated. Factor analysis was used to extract the most influential factors on the measure of service quality and four factors were selected namely: they deliver when promised, precision on account statements, queues that move rapidly and sufficient number of ATMs per branch. In order to determine the reliability of the multiple Likert questions in the survey, Cronbach’s α was used indicating a scale reliability of 0.743. Moreover, multiple regression analysis was carried out on the selected factors to design an objective function for the design and evaluation of service quality model. The model design used for benchmarking was done using multiobjective genetic algorithm in MATLAB. Similarly, the model evaluation was done in a java interface using multiobjective particle swamp optimization.

The evaluation results validated the designed model and showed that the factors they deliver when promised and queues that move rapidly are a more reliable scale of measurement for customer’s satisfaction than the factors precision on account statements and sufficient number of ATMs per branch.

The implication of the results is that effectiveness and assurance combined with access is a more significant factor for measuring customers’ satisfaction than tangibles based on the BSQ model.

The introduction of a two-phase optimization model for model benchmarking and evaluation as compared to ordinary factor analysis of the dimension constructs.