Crowdsourcing is a paradigm which enables and harnesses the power and wisdom of a usually large, diverse crowd in innovation, problem solving and knowledge acquisition. The scale, benefits and application areas of this traditional model are amplified by the advances of information and communication technology such as the advent of Web 2.0, which, at the same time, has increased the complexity and, hence, the need for systematic development approaches. While crowdsourcing has been successfully applied in several projects and de-facto platforms already exist for them, the research on engineering principles, methods and tools for developing a crowdsourcing project is still in the early stages. In this paper, we study the adaptation of crowdsourcing settings to fit the nature of the problem being crowdsourced. As a method, we review the literature and complement that with an online expert survey involving practitioners and researchers active in the field of crowdsourcing. We then interpret the obtained results and identify a set of recommendations on how to set up crowdsourcing to fit each of the five common categories of problems. Our results inform future crowdsourcing developers with best practice experiences on planning and configuring their projects.