The project focused on solving the situation that US Sales leaders in Nuqleous may struggle with fluctuating demand and ineffective perishable goods allocation due to unknown future sales under different purchase channels.
To optimize perishable goods allocation for Nuqleous, we developed several models to better understand product sales across 4 purchase channels and stores in different locations by analyzing sales trends and forecasting next month’s weekly demand.
The data source we used is from Nuqleous’s daily historical sales. It comprises 1.8 million records with roughly two years of data. In this project, we focused on the sales prediction of the top-selling item, SKU number 1765845, across the 4 purchase channels and among different stores in the US over the past two years.
Our data-driven forecasting solution is briefly described below.
Since our goal is to forecast the demand for the top-selling item (SKU number: 1765845), we first clean the data we needed, including combining this and last year's data, transforming daily data into weekly data, filling the gaps with no sales data, looking deep into unusual peaks, and then use reasonable ways to make the data tidy and able to run forecasting. Next, we applied 8 time-series models to each of the 4 purchase channels while considering the sale trend and seasonality at the same time. Then, we found the one with the best forecasting results under each purchase channel. Lastly, we applied those best models to forecast the weekly sales of each purchase channel for the next 4 weeks, which is our main goal of this project!
After finding the best forecasting tools and results, here are our recommendations.
Since the “In-store purchase channel” and the “Store-pickup purchase channel” have an overall decreasing trend from the forecasting plot, Nuqleous sales leaders have to be very careful if they would like to place more goods than the previous period in the future. Besides, it is necessary to evaluate these stores’ sales, especially for the stores with dramatically dropped sales these years. If the ability to earn money for some physical stores were lower due to the change in customers’ purchasing habits, Nuqleous can also think of gradually switching some investment to purchase channels like “store delivery channels”, which had a growing sales trend these years.
In contrast, the “store delivery channel” has a clear upward total sales trend, which tells us this purchasing channel has the potential to earn more money for Nuqleous. However, the company still has to prevent ordering too many goods at a time because the overall growing trend does not promise the same increasing sales each week, instead, the sales always fluctuate.
At last, the “Online (Shipped from Store) channel” has an apparent fluctuation in sales trend and also the lowest total sales amount among the 4 purchase channels. We’ve found that the forecasting sales had a high correlation with the sales exactly a month (4 weeks) before that day, which can be taken into consideration when making decisions.
Overall, this project aims to assist Nuqleous in grasping the sales trend more accurately in the future in a data-driven way. This kind of tool can help when making ordering decisions and product allocation.