An increasing variety of products contributes to the challenge of efficient manufacturing on production lines, e.g. in the Fast Moving Consumer Goods (FMCG) sector. Due to the complexity and multitude of adjustment levers, the identification of economic actions for improvement is challenging. Reinforcement learning offers a way to deal with such complex problems with little problem-specific adaptation. This paper presents a method for decision support for economic productivity improvement of production lines. A combination of discrete event simulation and reinforcement learning is used to identify efficient, sequential trajectories of improvements. The approach is validated with a fill-and-pack line of the FMCG industry.