Operational Risk Rises When Supply Chains Deploy Generative AI Over Deterministic Models

AI isn’t failing supply chains. Model selection is. As adoption accelerates, a growing number of high-profile misfires trace back to a basic mistake: applying generative, non-deterministic AI to problems that require repeatability, precision, and financial certainty. The result is a widening gap between AI hype and operational reality, where the wrong models introduce new risk instead of delivering real efficiency.

Stefan Groschupf is the CEO of Centrum AI, a deterministic AI platform for supply chains, and has spent decades working on systems where precision is non-negotiable. A serial AI and big data entrepreneur, he helped launch Hadoop, founded Datameer and Automation Hero, and brings that experience to a blunt assessment of why today’s supply chain AI is so often misapplied.

"Demand planning, logistics decisions, and inventory allocation are deterministic problems by nature, which is exactly why generative, non-deterministic AI fails in supply chains," says Groschupf. According to him, the core issue is a failure to distinguish between different branches of the AI "technology tree."

• A tale of two AIs: Groschupf says leaders are drawn to generative AI’s flexibility and deploy it in core functions where uncertainty becomes a liability. In supply chains, he draws a hard line. "For real math on real dollars and real physical goods, we need deterministic AI," he insists. "We cannot use anything generative. That includes agentic AI, which is just a way of using generative models." The reason, he adds, is basic operational reality. "With generative AI, you can ask the same question twice and get two different answers. That’s completely unacceptable for deterministic problems."

That distinction is more important than ever, as supply chains face increasing volatility from global uncertainty. With disruptions from cyberattacks on suppliers and trade disputes contributing to significant financial losses measured in the hundreds of millions of dollars, Groschupf believes companies should view investments in supply chain security and resilience with a similar priority to cybersecurity.

Groschupf also dismantles the idea that companies must wait for pristine data before seeing AI value. Asked about the minimum data maturity threshold, his answer is blunt: "None." He explains that deterministic AI can be used to fix data problems as they occur, rather than after the fact.

• Data debunked: "If someone wearing gloves accidentally logs 11 boxes instead of 10, deterministic AI can catch and correct that error at the source," he says. That approach delivers immediate impact, even in small environments. "We saw a 60-person manufacturing company achieve 10 to 15 percent inventory optimization without a data warehouse or formal database," Groschupf says. "Waiting for perfect data is the wrong starting point when AI itself can be the tool that makes the data usable."

Groschupf suggests the path forward involves education. He recommends that generative AI be reserved for peripheral tasks, like summarizing documents. But he cautions that applying the same technology to core inventory optimization is not innovation—it's a choice that introduces unnecessary risk. As global pressures mount, the consequences of misusing these tools are not abstract. "It all comes back to the same thing," he concludes. "Use the right tool for the right job."