Business Intelligence systems provide an effective solution for multidimensional online computing and analysis from large volumes of data. In the decision making process, analyzed data are typically stored in a set of cubes often heterogeneous. Most of the time, the structure of these cubes is unknown by decision makers. Our goal is to enable decision makers to express their need via a query in natural language. In this paper, we deal with the problem of data cube design and construction according to a user query where the expressed need is dispersed on more than one cube. We propose a cubes design and construction approach based on fusion of cubes containing a part of the user's need. We validate our approach via a tool, called "Design-Cubes-Query", that implements our approach and we show its use through a case study.