The perfect representation for data manipulation is based on collections, lists or maps but most applications are designed based on an object domain model.
We solved this issue by implementing a KVM (Key Value Mapping) at the core level of our application. Data manipulation became very easy after implementing the bi-directional mapping of our entities to a key/value model. It took significant time before being able to leverage this new ecosystem and it started with a wide refactoring of the entities in order to fix our model key space.