Minimizing the number of robots required for a Robotic Process Automation (RPA) problem

Abstract

Robotic Process Automation (RPA) is used in various fields of human activity in order to implement faster and more secure processes through a reduction in the risks or errors but also an increase in the productivity rates. The increase of its use and importance calls for evermore efficient solution methods for this problem. In this paper, the RPA is addressed in the context of a financial institution. The problem consists in assigning transactions to software robots, where each transaction type has a different clearance date and a different processing time. First, four heuristics are used to compute an upper bound on the number of required software robots. Then, this bound is given as a parameter to an integer linear program, which is used to assign the transactions to the different robots. The quality of the solutions is assessed by an extensive experimental study on a set of 39,000 instances. The results show that two heuristics outperform the others and allow for a faster resolution by the integer linear program which in turn finds the optimal solution for most of the instances within a timeout of 60 seconds.