With thousands of applications serving around 70 countries globally, the Platform Team at Delivery Hero cannot just hit the upgrade button after grabbing coffee in the morning. Upgrades should not affect our uptime - but what if they do? We need to make sure that any issues that do occur have the least possible impact on our global operations. In this session, discover how we built a data-driven scheduling algorithm that understands the context of each environment to upgrade, eliminating the need for manual intervention. Learn how this system powers thousands of monthly EKS cluster upgrades, maintaining high availability and reliability. We'll share practical insights on how you can adapt this concept to your own environment, regardless of its complexity.