You can take nearly any black box model and add some interpretability into it, without modifying this model. No modifications to the model, so accuracy stays the same.