South KoreaRetail and e-commerce2019 – ongoingE-mart
SAIcast AI demand forecasting and automatic ordering
E-mart's AI forecasts beat manual ordering in 92 of 98 product groups in a pilot, and two stores switched to fully automatic ordering
Problem
E-mart runs about 140 hypermarkets, each selling more than 70,000 items on more than 340 days a year. Staff forecast sales a week ahead by hand, which led to waste, lost sales from stockouts and excess stock, especially for promotions whose demand swings with time-limited price cuts.
Approach
The group's IT affiliate Shinsegae I&C built SAIcast, a cloud forecasting engine on AWS trained on two years of sales data with about 40 variables. In an October 2019 pilot at two stores covering 10,000 products, it beat manual forecasts in 92 of 98 product groups, and after a month of review those stores moved to automatic ordering without human input. A second version for promotional items was tested across about 140 stores and some 50,000 items in April 2020; forecasts feed the ordering system, which places orders for each store and warehouse.
Results
Takeaway
E-mart compared AI with manual forecasts category by category before automating orders, and used a separate model for promotional items. The company has not disclosed waste or inventory savings.
Categories
- Size
- Large enterprise
- Industry
- Retail and e-commerce
- Business function
- Logistics and supply chain
- Innovation type
- Process innovation
- AI technology
- Predictive analytics and ML
- Adoption
- Co-developed with a partner
- Evidence
- Company's own claim
Sources
Source summaries
Byline Network describes how Shinsegae I&C built and tested SAIcast, drawing on a presentation by Shinsegae I&C and AWS. It reports the October 2019 two-store pilot (10,000 products, AI better in 92 of 98 groups, 18% lower error), the switch of those stores to automatic ordering and the April 2020 test of a promotional-item model across about 140 stores and 50,000 items, where AI was more accurate 82:18. It also notes doubts in the field about feasibility and cost-effectiveness.
Money Today describes SAIcast as a cloud platform predicting daily sales per product from hundreds of variables such as weather, price, promotions and season. Forecasts are passed to the ordering system, which orders the required quantities for each store and warehouse. It says the system reduced waste, lost sales from stockouts and excess inventory, without figures.
Full details are in the original sources.