Robotic Process Automation (RPA) is known for its value in cost-saving, error-free, scalable and efficient solutions for real-world business problems. However, RPA modelling on accountants’ decisions in Enterprise Resource Planning (ERP) system is still in the infant stage to the best of our knowledge. This research proposes Smart Virtual Robot Automation (SVRA) conceptual framework with deep learning for ERP applications. The research methodology’s first step is to gather the human expert’s action as label and ERP application information as input data. Furthermore, the trained supervised deep learning model will predict human experts’ actions during simulation. For each simulation, a human expert will perform the simulated SVRA action verification process. The objective of this work is to study the proposed approach’s similarity to human expert accuracy or efficiency in action. The significance of the SVRA conceptual framework is to enable automation of the supervised deep learning approach.