Sales forecasting is one of the most critical applications that can significantly affect various management levels in the retail industry. Traditionally, retailers offer thousands of products in several locations resulting in sales records with extremely high variance. Given the high variance, traditional forecasting methods may struggle to accurately predict future demand. In this paper, a systematic experimental study has been conducted to investigate the impact of various time series and ANN-based models on retail demand forecasting. First, data has been prepared using extract, transform, load, and data cleaning techniques. After that, the dataset has been investigated to explore its nature and detect outliers. Then, nine time-series techniques are applied to select the best fit. Finally, the results indicate that the ANN model outperforms the time-series one considering high-variance data. In addition, it shows that including both the product stock and price features can improve the ANN's prediction accuracy by 35%.