If the core value of AI is the ability to sift through oceans of data for insights that lead to better, faster decisions, then the supply chain is close to an ideal use case.
“You have data coming in from ERPs, transportation management systems, warehouse management systems, and manufacturing execution systems,” says Peter Budweiser, general manager for supply chain at Celonis. “In sectors like retail and manufacturing, the supply chain is what enables most of the revenue, and it’s also where most of the cost comes from. So, making AI work in these industries really means making it work in the supply chain.”
But AI tools need more than just massive volumes of data to produce actionable insights, Budweiser warns. They also need a way to understand what this data represents and how different data sources relate to one another.
“It’s not only about having data,” he says. “You can’t just put your AI agents on top of a huge data lake and let them run. That doesn’t work, because the data alone does not provide enough context.”
To provide that context, organizations need a real-time digital representation of their supply chains. Tools like the Celonis Context Model, Budweiser says, provide both a semantic layer for the data (which translates raw system data into business meaning) and an ontology (which maps the relationships among supply chain objects, events, and processes).
Without a semantic layer to make sense of various types of information, Budweiser explains, data points like stock keeping unit (SKU) numbers and shipment details float around data lakes as essentially “meaningless” figures. Just as a human would likely struggle to even identify a particular string of eight to 12 characters as a SKU, AI tools like large language models (LLMs) need to be told what type of data they are looking at before they can make sense of it.
Similarly, Budweiser says, an ontology is needed to show how these data points relate to one another across the supply chain. A SKU, for example, becomes far more useful when an AI system can connect it to the customer who ordered it, the supplier that provided it, the inventory location where it is stored, and the business event that requires action. By mapping these relationships, an ontology helps AI tools understand the supply chain at a systems level.
Budweiser stresses that the value of an AI-enabled supply chain comes not from generating another dashboard or a list of insights, but rather from delivering just-in-time support for practical decision-making. For instance, if a manufacturer is missing a critical spare part, an AI agent could evaluate whether the best option is to pull the component from another warehouse, source it from a supplier, or expedite a shipment by air based on the economic tradeoffs of each option.
“You need to make sure the decision support ends up in the place where it matters,” Budweiser says. “In the end, AI is only worth it if it gets operationalized.”
To learn more, click here.