Users beam-time at modern FEL sources is an extremely valuable commodity. Moreover, maximization of FEL up-time must always be performed accounting for stringent requirements on the photon pulse characteristics. These may vary widely depending on the users requests, which poses issues to parallel operation of high-repetition rate facilities like the European XFEL. Therefore, both model-free or model-dependent optimization schemes, where the model might be given, or provided by machine-learning approaches, are of high importance for the overall efficiency of FEL facilities. In this contribution, we review our previous activities and we report on current efforts and progress in FEL optimization schemes at the European XFEL. Finally, we provide an outlook on future developments.