The interoperability of applications depends on a successful mapping between their domain models. Nowadays, common file formats serve as a mediator between the different domain models but cause friction losses during the conversion of data. These loses could be mitigated whenever the models are already working on the same concepts but are only using different representations for them. We propose the concept of deducing classes which interpret existing object structures and detect instances of themselves in this existing data. Further, we introduce a planning algorithm which combines deducing classes to allow unanticipated interactions between applications. We discuss some of the implications of this approach and illustrate upcoming research challenges.