Challenge
A logistics organisation were losing over £5M per year in wasted stock. The organisation imported fruit, vegetables and flowers from across the world and were often wildly over or understocked. When understocked goods had to be purchased from other local providers at a premium.
Additionally, they employed a team to produce demand forecasts and also purchased a forecasting tool costing over £350K per year.
Their existing ERP platform provided the necessary data, however the existing data warehouse was on-premise, preventing use of modern cloud based data processing and machine learning tools.
Solution
An initial Proof of Concept (PoC) was conducted to explore the use of traditional machine learning on a data export. Results were positive with 8 out of 10 items testing showing improvements in forecast accuracy.
The existing data warehouse was migrated into the Azure Cloud and hosted in an Azure SQL database. Azure Machine Learning was then used to test a range of machine learning algorithms on the data. External data sources (weather, holidays…) were stored in Azure Data Lake and used to augment the models.
The resulting forecasting and supporting ERP data were built into an interactive Power BI dashboard for the demand forecasting team.
Project Details
Project Duration - 1 week PoC, 4 week delivery
Project Team - 1 delivery lead, 1 data scientist, 1 data analyst
Benefits
The machine learning based solution was in some cases over 30% more accurate than the existing forecasting process, enabling potential cost savings of over 500K per year.
The solutions cost was under £7000 per year, a fraction of the existing technology and people cost.
With machine learning foundations built in Azure, the organisation could begin to expand their analytics capabilities more efficiently into other use cases.