In this thesis, we research a cosmetics company’s sales forecasting. We analyze the sales data from 2017 to 2020, a total of 38 months sales record. The objective of this study is to find the best forecasting model automatically. We implement twelve models using four time series forecasting methods, including Moving Average, Linear Regression, Exponential Smoothing, and Holt-Winter Exponential Smoothing, with three different ways of applying seasonality. Among the top 100 sales items, we improve the accuracy of forecasting averagely by 12.9% and increase the forecasting performance for more than half of the items by more than 20% by using the proposed models.