Demand Forecasting
About the Emart Demand Forecasting
A system where the store manager reviews the sales records and current inventory to place orders.
E-Mart, currently Korea's leading mega store, has always looked into maximizing revenue and minimizing inventory loss for each product and each store.
Each store of E-Mart reflected the promotional events and sales records of the past, as well as the knowhow of the store manager before placing product orders but the timing was not always optimal. Stock-outs led to customer dissatisfaction and accumulation of unsold inventory were ongoing problems.
The Challenge
Minimization of inventory loss and optimization of store operation.
Shinsegae I&C established a demand/sales volume forecasting platform based on AI to address the concerns of E-Mart (in October,
2019). This platform is currently in operation at two E-Mart stores.
Demand forecasting is done by collecting and analyzing the sales information and inventory data. AI-based features such as pre-processing and model learning are used to forecast the sales and based on the forecast figures automatic orders are placed and inventory is managed.
I&C Solution
Shinegae I&C focuses on the execution pattern and data flow (process) of the forecasting platform to design and establish an Infrastructure optimized for AWS resource and functions. Security services are appropriately used to minimize any security issues with the legacy system or during data transmission.
First, sufficient resources are used for data analysis, pre-processing and model learning but AWS Batch and ECR services are used to ensure there is no increase in unnecessary costs and that assets are allocated to only when workload is carried out. Since the machine learning server for data analysis and model learning is not always operated, cost can be saved.
Secondly, with the VPN Connect setting using AWS Transit Gateway services, security issues between the cloud and the legacy system during data transmission are minimized. This is because for sales forecasting, the sales and inventory data must be collected from the legacy system and the forecast figures must be sent to the legacy ordering system.
Used AWS Services
The Benefits
By developing an AWS-based demand forecasting platform, a high specification machine learning server is converted into AWS Batch+ECS(Fargate). offering a 50% cost-saving and efficiency in the operation environment.
3rd Party Applications or solutions used
Other solutions were not used in this project.