Forward flux sampling (FFS) schemes have become a valuable tool for studying the dynamics of complex systems; however, its applicability is highly limited by sampling efficiency, which is dictated primarily by the choice of the order parameter but also by how the phase space is partitioned and sampled. To address these shortcomings, we propose a simple adaptive optimization algorithm which optimizes the phase space staging by concentrating the sampling in the "bottlenecks" of the FFS-type simulation. The algorithm is based on an iterative procedure that systematically identifies the kinetic bottlenecks along the reaction coordinate (i.e., order parameter used to partition the phase space via interfaces) and improves the estimates of the conditional probabilities to reach subsequent interfaces (P_i) to then improve the sampling of the phase space. Hence, the approach can optimize for either the number and position of the interfaces (i.e., optimized λ phase staging), and/or the number M of fired trial runs per interface (i.e., the {M_i} set) to minimize the statistical error in the rate constant estimation per simulation period (i.e., increase the computational efficiency). The method is demonstrated for the reassembly and folding in split proteins systems that are representative of numerous protein-protein interactions encountered in many biological applications. The kinetic studies of such systems by FFS-type simulations are usually limited by the choice of a suitable reaction coordinate (i.e., a model on relevant variables) to describe the intrinsic system's dynamics. It is shown that the proposed approach leads to an optimized λ staging and/or {M_i} set which optimize the flux of partial trajectories between interfaces based on an input desirable distribution of P_i values. Indeed, the λ phase staging could be optimized first to position interfaces at the transition state and FFS's "bottleneck" regions and the {M_i} set then be optimized to improve the sampling and reduce the effort needed to obtain committor probability data for the estimation of refined reaction coordinate models. When taken together, these algorithms provide a gain in efficiency of over two orders of magnitude when compared to traditional FFS simulations.