Data Envelopment Analysis (DEA) as a data-oriented benchmarking tool is considered a powerful and promising instrument for performance evaluation in various application areas. In DEA, the set of all decision-making units (DMUs) is divided into efficient and inefficient subsets. Inefficient DMUs are improved by reducing the input and/or increasing the output, and as far as we know, efficient DMUs are abandoned with the conclusion that they are all technically and relatively efficient, and no further analysis has been suggested in the literature. In this article, we first show that there is a gap between the actual efficiency and the efficiency estimated using benchmarking tools such as DEA. This means that there is no guarantee that the efficient DMUs characterized by DEA are really efficient. Thus, there is a gap in improving the technical efficiency of efficient DMUs. In this paper, we attempt to close this gap by introducing a method to improve efficient DMUs. First, we introduce a random variable as a corrector of efficiency evaluation, and then an inverse DEA model (IDEA) is proposed to improve efficient DMUs. To demonstrate the actual applicability of the proposed approach, we present an illustrative empirical application using 106 Chinese bank data from 2021.