In an effort to achieve agile control in a just-in-time (JIT) environment, reactive JIT ordering systems have been proposed for multistage production systems with unstable changes in demand. In reactive JIT ordering systems, the time series data for the demand are monitored, and unstable changes in demand are detected using statistical control charts. Then, in these systems, the number of Kanbans and the buffer size at each stage are adjusted as a response to the detected unstable changes. Recently, an adaptive Kanban system based on inventory levels was proposed and applied to unstable demand conditions. However, its performance has not yet been analysed and compared with the other reactive system. Therefore, the first purpose of this paper is to analyse and compare the performance of the reactive Kanban systems, both the control chart-based and the inventory-based systems. In addition to comparing their performance, as the second purpose, a new inventory-based system is proposed. In the proposed system, the time series data for the demand is not monitored, but instead the inventory level of a product is monitored to detect unstable changes in demand, and the number of Kanbans is adjusted as a response to these detected unstable changes. The performance of the proposed system is investigated and compared with those of the previous systems by using simulation experiments that model unstable demand conditions.