Michelin uses machine learning to predict stock outages 15 days ahead in its supply chain

Overview

Michelin used the Databricks Data + AI Platform to democratize data access across the organization and build machine learning models trained on past supply chain data to predict stock levels 15 days ahead and recommend actions to prevent outages. Michelin used Databricks for unifying input data, collaborating on model training, and deploying and orchestrating models into production, with data continuously streamed and kept up to date in Delta Lake. The stock-outage prediction use case was an early success that helped Michelin scale to hundreds of AI and data use cases across the company.

The challenge

Michelin's former on-premises data platform lacked the necessary openness and flexibility; with data siloed by department and the platform's centralized nature, providing services at scale was challenging and employees were often unable to access the tools and technologies they needed.

The solution

Michelin used the Databricks Data + AI Platform to unify input data and collaborate on training machine learning models on past supply chain data, deploying and orchestrating the models into production, with data continuously streamed and kept up to date in Databricks Delta Lake while business analysts used Databricks SQL to run queries and visualize insights.

Reported business value

Using lakehouse, Michelin has been able to scale to hundreds of use cases on the platform beyond the original stock-outage prediction use case, empowering citizen data analysts and data scientists alike to build and share their own use cases and best practices.