This study proposes a Reinforcement Learning-based replenishment strategy aimed at improving the responsiveness of a real-world case study from the semiconductor industry, operating in a complex environment characterized by perishable products, stochastic lead time, and minimum order quantity. The agent explicitly determines the optimal order quantity while managing product perishability. To validate the effectiveness of the proposed replenishment policy, it has been compared with the Smoothing Order-Up-To (SOUT) policy. Such comparison analysis is based on two performance measures, such as the fill rate and the amount of expired products. For the sake of fairness, the SOUT replenishment parameters were calibrated by means of a Response Surface Methodology analysis. An extensive experimental campaign was carried out and numerical results show the outperformance of the Reinforcement Learning over the SOUT by mitigating the effect of variability in the lead time, which in turn implies inventory cost reduction and fill-rate increase.