Forecasting customer demand for packages in SIG Indonesia market for better customer sales promotion

Project Details


Fall 2018


Shang-Chi Tu, Beverly Lin, Ken Wu, Jimmy Wang






Introduction of Our Client: SIG
In this project, our client is SIG Combibloc. SIG is a leading systems & solutions provider of carton packaging and flexible filling machines for beverages and food, helping bring food products to consumers in a safe, sustainable and affordable way.

Business Problem
The main business goal for this project is to offer next-12-month forecasts for future monthly sales volume of packages on a customer and product type level at the beginning of each month by us. With the forecasts, the salesperson of SIG Indonesia can use it as a reference during their monthly meeting to revise their month marketing strategy and design tailor-made promotions for customer sales in advanced.

Data Description
We obtained data from SIG which included fields such as Customer,
Product hierarchy, Month and Plan qty, etc. The time period of the series
are from January 2009 to December 2018 and it recorded every sale for
45 different customer and product type level.

Forecasting Solution
Before forecasting, we did data preprocessing and customer segmentation. Then, we applied different models to our 3 types of customer. For the inactive customers, we forecast zero in the next 12 months. For the new customers, we used Naive to forecast their sales
demand for packaging. For the repeat customer, because of some customers’ extreme behaviors, we used linear regression and modeled these months separately. For more information, please refer to the detailed report below.

According to the results, we observe orders from some customers are steadily growing. Hence, the salesperson of SIG Indonesia market can pay more attention to those customers with potential to purchase more in the future while developing a sales promotion strategy. Besides, SIG can try more external data such as customer satisfaction scale can help
discover customers’ ordering behavior and improve the forecast accuracy.

Application Area: