Computing a lower bound for the solution of a Robotic Process Automation (RPA) problem using network flows

Robotic process automation (RPA) helps companies reduce the time required to process tasks by using software or robots to mimic human actions on graphic interfaces. In this paper, the RPA problem is solved for a financial institution. A set of different types of financial transactions are to be processed with different processing times, volumes, market hours and clearance delays. In a previous work, a two-phase linear integer model was used to solve the problem on small instances. In this study, a network flow algorithm is used to compute a lower bound for the problem, thus reducing the computational time required to obtain a solution. The method is tested on a real case provided by a bank in North America and on synthetic test cases containing a greater number of transaction types. Results show that combining the computation of the lower bound with a linear integer model is faster and more practical.