Forecasting systems usually use historical data and specific methods (typically statistical models) to derive decisional information. To be accurate, the volume of historical data is often very large and the calculation and the exploration processes are complex. Anticipeo is one of the forecasting systems which is devoted to the prediction of sales based on collected sales over a \long" period of time. Even if it provides quite a reliable prediction, the query evaluation stays a costly process with high latency. So far, we have investigated the latency provenance and we have proposed some solutions to improve query response time for the exploration process. Our design principles can be reused in any application context that displays similar requirements.