PT XYZ is a company engaged in the retail minimarket business in Indonesia. In running its business, one of the key activities involved is opening a new minimarket store. The audit team will analyze proposals to predict sales of would-be new stores, however, the results of the predictions are often not following reality, so research is needed to predict sales more accurately. This study aims to analyze the prediction of minimarket stores sales using deep learning technique to determine new minimarket stores. The model can predict 53.18% stores that achieve the target prediction and 28.32% stores that have the potential to achieve the target in the future, which indicates an increase in predicting stores on target compared to the branch office method which only predicts 31.2% of stores that achieve the target and 31.62% of potential stores. Thus, the approval decision of new minimarket stores which predicted achieve its target can be more accurate than the branch office method. The audit team will use the model to predict the store sales and consider the result for approving the proposal. Factors that had a significant influence on sales were rack size, store age, distance between competitors, domain location, and store type.