Supply Chain AI Must Do More Than Inform; It Must Help Businesses Act

I’ve spent more than 25 years building and scaling supply chain operations before joining SAP to lead SAP Supply Chain Management. The decision came down to something I kept encountering throughout my career: the companies with the deepest understanding of their own operations are the ones best positioned to put AI to work.

And nobody sits closer to those operations than SAP. In every customer conversation I have, I hear a similar tension: leaders know that AI will reshape their supply chains. But they need software that understands their business processes well enough to move from experimentation to impact. That challenge, and SAP’s foundation to meet it, is what brought me here. 

What excites me most about this moment is what I call “the art of the possible,” looking beyond how things work today and imagining what best-in-class could look like if we took advantage of capabilities that simply were not available before. But expanding the art of the possible does not, by itself, create meaningful innovation.

For decades, companies have invested heavily in planning, manufacturing, procurement, and logistics systems that provide valuable visibility. But when a supplier misses a shipment, demand unexpectedly increases, or transportation capacity disappears, people remain essential to connect the dots across systems and functions. By the time they do, the business may already be reacting to the problem rather than getting ahead of it.

What becomes possible when the supply chain is a modern decision engine

Imagine an illustrative scenario: a supplier delay threatens production and several important customer orders. Traditionally, a planning system might generate an alert, leaving teams to determine the implications and coordinate a response. An intelligent system could go further by identifying which customer commitments are at risk, assessing available inventory across locations, evaluating whether production schedules should change, and comparing alternative transportation routes or suppliers. 

Each option involves tradeoffs. Expediting a shipment might protect a high-priority customer order but increase costs. Reallocating inventory could preserve one commitment while creating shortages elsewhere. Adjusting production might minimize disruption but affect other delivery schedules.

Humans evaluate these interconnected decisions against the outcomes that matter for the business, rather than optimizing each function independently.

Success is measured by whether the response reduced missed customer commitments, protected revenue, limited incremental logistics costs, and avoided unnecessary inventory. Comparing results against the original plan helps the system improve future recommendations.

That is a fundamentally different model. Competitive advantage increasingly depends on how quickly an organization can turn change into coordinated action.

Autonomy should remove waiting, not people

As AI assumes a greater role in these decisions, we also need to be precise about what we mean by autonomy. The art of the possible lies in the absence of unnecessary waiting—not removing humans from the loop. Constructing a false competition between humans and machines creates a direct obstacle to delivering autonomy. Instead, we should be laser-focused on eliminating unnecessary work and latency while allowing people to focus their judgment where it creates the most value.

That also means we need to have a more sophisticated conversation about AI accuracy. An agent making a low-risk, reversible decision does not require the same confidence threshold as one making a decision with significant financial, customer, safety, or compliance implications. The question is not simply whether an AI system is “accurate.” It is whether it is sufficiently reliable for the particular decision it is being trusted to make.

Enterprise AI must earn the right to act 

Before AI can be trusted to make operational decisions, it must progressively earn trust. A system might initially recommend responses while planners review its reasoning, compare alternatives, and decide whether to act. Organizations can then measure how those recommendations perform against actual outcomes, including whether they protect customer commitments, respect business constraints, and deliver expected improvements. 

As reliability becomes more tangible, businesses can then feel more empowered to authorize AI to execute defined, lower-risk actions within established limits. More consequential actions should continue to require human approval, with transparent reasoning, documented decisions, and the ability to intervene. 

This is how AI earns the right to act, and this is where SAP has an important role to play. Our applications connect critical business processes across planning and execution, providing the operational context needed to understand how one decision affects another. By combining that context with enterprise policies, permissions and auditable processes, we can help organizations move toward coordinated AI-driven action without sacrificing control.

We are already seeing the foundations of this approach emerge. For example, SAP’s Supply Chain Planning Assistant can help planners identify potential root causes of planning issues and recommend corrective actions. Takeda Pharmaceuticals is using the assistant to support this process alongside automated performance tracking, creating a feedback cycle that can improve planning over time.

As these capabilities mature, the role of supply chain practitioners will evolve as well. Rather than spending most of their time assembling information and manually resolving exceptions, planners can focus on defining objectives, evaluating tradeoffs, challenging assumptions, and making the strategic decisions where human expertise matters most.

The best AI strategy starts with the business problem

Perhaps the greatest risk in this period of rapid AI development is starting with technology and then searching for a problem to solve. 

The better approach is to start with the customer problem: understand the outcome that matters, determine how to measure it, and then ask how AI can achieve that outcome in a way that was not possible before. Better decisions can improve product availability, protect customer commitments, strengthen resilience, and unlock growth. 

As technology expands what businesses can accomplish, innovation must remain grounded in the outcomes customers need. They should hold us accountable not for how many AI agents we develop or how much automation we introduce, but for whether our technology helps them make better decisions, respond faster to change and deliver measurable business outcomes.

Devesh Mishra is general manager and chief product officer for SAP Supply Chain Management.