Classical architectures proposed so far for data warehouses show some drawbacks when adopted to work over large numbers of heterogeneous operational sources. In this paper we propose a variant of a three-level architecture for data warehouses that overcomes these drawbacks. However, in the application context under consideration, having a suitable architecture may be not enough for the design purposes. Indeed, data warehouse design in very large operational environments can be a quite hard problem to attack with traditional manual methodologies. In this paper, automatic techniques are also illustrated that are capable to produce the data warehouse design according to the proposed architecture, with a limited human intervention.