Demand-Driven Manufacturing (DDM) is the solution that most companies are heading to in our days. Although this strategy consists of producing goods based on what consumers demand, companies should also rely on accurate forecasting systems to prepare their production chain for such an operation by supplying enough raw material, increasing production capacity to fit the desired demand, etc.… However, due to the fact that most companies have been relying on massive production, most sales forecasting systems usually used rely on sales data of previous years that, not only contain the actual demand, but takes into consideration the marketing strategy effects like massive promotions. Hence, the resulting forecasts do not mainly reflect consumers' demand. For this reason, a switch to demand forecasting, instead of sales forecasting, is essential to ensure a good transition to DDM. This paper proposes an artificial intelligence based demand forecasting system that aims to determine "potential sales", mainly reflecting consumers' demand, by correcting historical sales data from external variables' effects. A comparison with other sales forecasting models is performed and validated on real data of a French fashion retailer. Results show that the proposed system is both robust and accurate, and it outperforms all the other models in terms of forecasting errors.