In order to be able to 'predict' the extent or likelihood of success of an intervention, it is essential to know that exposure and effect are causally related. The randomised controlled clinical trial (RCT) is commonly referred to as the golden standard to establish causality. Apart from ethical constraints or rare diseases it is rarely impossible to do an RCT. Sometimes, however, RCTs may not be feasible (at least at first glance) due to e. g. patients' or therapists' reluctance to participate, lack of funding, organisational limitations, or because technical skills to deliver a therapy are not available. A variety of design alternatives have been proposed to overcome these problems, e. g. clinician-preferred treatment, single consent design, double consent randomised design, prerandomisation, cluster randomisation, and response-adaptive designs ('play the winner'). For complex ('combined modality') therapies it seems reasonable to study the effect of the whole treatment package on the patient rather than the value of individual components. Change to open label or optional cross-over allow 'individualized treatment' as well as 'individualized outcome' within the framework of the RCT. For clinicians and researchers it seems wise to put more emphasis on the search for a suitable and appropriate de-sign for their current problem rather than to artificially tailor their study to an awkward standard design (or even abandon the idea to seek an answer to an open question): Much more alternatives exist than one would imagine.