Repositioning operations are fundamental on bike-sharing systems. Its optimization is necessary in order to achieve the best level of service with minimum agency cost. Literature propose routing models that preventively avoid full and empty stations according to demand forecasting with good results. However, simulations show that reactive methods could improve the performance in some scenarios, because they can adapt to unexpected demand variations quicker. This paper describes a mixed repositioning model for station-based systems that includes both the preventive routing optimization and the real-time reactive adaptability. Results obtained on a simulated case of study (Barcelona ‘Bicing’ system) are positive. Model acts as a flexible repositioning clustering method, which breaks the cluster to control user costs if demand deviates too much from expected.