Use case in production
Using Transformer Models to Predict Demand at Scale in Production
Senior Data Scientist, Tchibo GmbH
Sören has been working as a data barista in the IT department at Tchibo for 3.5 years and develops innovative data products for his internal customers from supply chain, sales and controlling departments. At the same time, he tries to support his customers in keeping clean data in a structured manner, promotes data literacy and modern machine learning solutions.
We implemented a machine learning system using Google’s Temporal Fusion Transformer to forecast customer demand in the online shop of a large omnichannel retailer. The ML system ensures that customer demand can always be served from the primary warehouse, even if articles may be in high demand. The training data combines categorical and real-valued time-series data from multiple internal sources. The ML system is built on the Google Cloud Platform, using various cloud-native services.