Recent years have seen the emergence of a lot of research on the application of machine learning techniques in actuarial science. In particular, there has been a noticeable amount of papers regarding machine learning applied to P&C Loss Reserving. Overall, there is a common understanding that machine learning techniques provide better prediction accuracy of the outstanding liabilities compared to traditional methods. Nevertheless, the greater accuracy is offset by a difficult interpretations of results. This makes them in line of principle not suitable in a increasing regulated world, as it is the insurance business. Our objective is to show how we can introduce elements of machine learning into the traditional actuarial reserving methods in a gradual way. We strive to achieve a balance between predictive power and interpretability by introducing step-by-step new machine learning elements, with the possibility to simply start from the legacy paid/incurred datasets underlying the loss claim triangles without introducing any cumbersome data requirement or significant IT budget.