Event-B offers a rigorous state-based framework for designing critical systems. Models describe state changes (transitions), and invariant preservation is ensured by inductive proofs over execution traces. In a correct model, such changes transform safe states into safe states, effectively defining a partial function, whose domain prevents ill-defined state changes. Moreover, a state can be formalised as a complex data type, and as such it is accompanied by operators whose correct use is ensured by well-definedness (WD) conditions (partial functions).This paper proposes to define transitions explicitly as partial functions in an Event-B theory. WD conditions associated to these functions prevent ill-defined transitions in a more effective way than usual Event-B events. We advocate that these WD conditions are sufficient to define transitions that preserve (inductive) invariants and safety properties, thus providing easier and reusable proof methods for model invariant preservation. We rely on the finite automata example to illustrate our approach.