Kraft Heinz has connected more than 8,000 machines across its North American network, which speak over 50 communication protocols, and has built over 3,000 use cases on top of that data foundation, according to a Microsoft Azure and AI MVP's write-up of the company's work. Its process digital twin runs in two plants, with expansion planned to all ketchup facilities across the network. The company generates $26 billion in annual revenue, employs 36,000 people, operates more than 70 factories and manages over 200 brands and roughly 5,000 SKUs. The manufacturing challenges map to three OEE domains, machine performance, yield and utilities consumption, which are the KPIs the digital twin is designed to optimize. The partnership with Microsoft goes back to 2022 and the manufacturing AI journey accelerated in 2023. Microsoft's Industry Solutions Engineering team helped identify a process digital twin as the highest-impact option. Kraft Heinz began with a Most Viable Experiment at one of its oldest cream cheese plants to test whether connecting a machine to a machine via data models could produce meaningful results. It validated the approach, particularly around the overfill problem on the filling line, and evolved into a Most Viable Product. On the edge, Azure IoT Operations runs on Azure Arc-enabled Kubernetes clusters and handles protocol translation, data normalization, MQTT brokering and routing. Shop floor data flows into Azure Data Lake Storage through the integration partner ST1 and is processed by Apache Spark jobs; machine learning predictions are stitched back together with real-time machine settings, and the resulting recommendation is pushed back to the shop floor. Two Philadelphia Cream Cheese use cases are described. Mixing and separating optimization sets machine set points for each production run and adapts to seasonal changes in milk fat and protein, with models retrained throughout the year. Fill point optimization runs distinct optimizations on each of the eight lines of the filling conveyor to avoid overfilling and underfilling. Kraft Heinz follows a four-step model: Paperless Operations, Recommendations, Predictive and Prescriptive Analytics, and Touchless Manufacturing, the step it is actively pushing toward. It operates primarily at Level 3, Intelligent, of a four-level digital twin maturity model, with Level 4, Autonomous, the desired future state. The cloud pipeline delivers a recommendation in approximately five minutes, and a migration to edge deployment targets sub-one-minute latency. The company is also building agents that connect into manufacturing systems to execute recommendations as soon as they are generated.