SAP Says Supply Chain AI Should Start With the Decision

Supply chain teams should choose the business decision and its metric first, and the AI use case last, Devesh Mishra told the audience at SAP Connect’s Supply Chain keynote.

Mishra, general manager and chief product officer for SAP Supply Chain Management, joined SAP three months ago. He said most companies make sound choices within each function that don’t add up to the best result for the whole network.

“You don’t have a bad decision problem,” he said. As networks grow, decisions multiply faster than people can coordinate them, and isolated AI projects deliver isolated value.

SAP now offers six specialized supply chain assistants, and Mishra said the number of agents behind them has doubled in three months. SAP plans to standardize a warehouse operator agent it built with Bosch and roll it out at scale, and Döhler has gone live on SAP Logistics Management at one of its plants.

What Starting With the Decision Looks Like

Mishra laid out a simple order. Identify the customer problem, define the decision and the metric that need to improve, and then ask how AI can help and how to measure the result. Starting from an AI use case, he said, is one reason so many pilots never reach production.

He used a planner facing a supply disruption as an example. The planner doesn’t need another alert. They need to know which customers are at risk, what their options are, and what to do next.

Hagen Heubach, chief marketing officer for SAP Supply Chain Management and Industry AI, explained why getting the decision right matters so much. He repeated the main keynote’s point that 80% accuracy isn’t good enough in supply chain, where an agent’s mistake can stop a production line or block a shipment. He described a customer whose self-built agent moved 2,000 components between warehouses, and the customer couldn’t explain why.

How Bosch, Döhler, and Microsoft Chose Where AI Fits

Each panelist traced their company’s AI work back to a specific business problem.

Anna Fleming, head of corporate supply chain management at Bosch, said she was skeptical when generative AI arrived, after trends such as blockchain failed to deliver. She changed her mind when pilots showed AI could handle complex tasks that rules-based software never could. Bosch and SAP have screened supply chain roles together for AI potential, and Fleming named workforce planning as one of the biggest opportunities. The result is the warehouse operator agent, which spots bottlenecks and recommends where to assign staff.

Döhler’s problem is volume. The ingredients maker runs one supply chain set by harvests and another set by customer demand, and climate and geopolitical disruptions produce more signals than its people can work through. Stefan, who leads supply chain in-house consulting at Döhler, said AI’s job is to organize that information so people can act on it. Döhler built a sales order agent with SAP, and every AI project there has to earn its place.

Microsoft pairs process cleanup with AI. Jeevan Potdar, a software engineering leader for Microsoft’s cloud supply chain, said the company’s Lean Plus AI initiative, started about three years ago, maps processes, removes waste, and uses AI for the routine work that remains. Microsoft now runs its own agents alongside SAP IBP.

The same pattern came up in SAP’s Finance and SuccessFactors keynotes. PwC and ITOCHU described moving off heavily customized systems first, and SAP’s own HR leader advised redesigning work before designing agents.

How SAP Plans to Connect Decisions Across Functions

SAP’s answer to isolated decisions is its supply chain suite. Mishra described it as the shared base of processes, rules, and data for planning and execution. SAP pointed to Blue Diamond Growers, which cut days of inventory from 75 to 60 after digitizing its supply chain on SAP with partners XpertMinds and project44.

The assistants are built to work across that base. The logistics assistant, for example, balances inventory, labor, and transportation together. In the live demo, a request to speed up a product launch moved from design through planning, finance, and manufacturing. The design assistant caught a 13-week component delay at the design stage, and a person signed off at each step.

Customers that want their own agents can build them in Joule Studio and connect them to SAP’s. None of the three panelists expects a fully autonomous supply chain, and Mishra agreed. Autonomy, he said, means taking waiting and waste out of operations, and people remain part of it.

What This Means for SAPinsiders

• Write the decision and metric into each business case. Name what the project should improve and how you’ll measure it. That gives each pilot a clear bar for moving into production.
• Look for work rules-based automation couldn’t handle. Bosch found its opportunity in complex tasks such as staff allocation. Those tasks often hold value that earlier automation left behind.
• Clean up the process alongside adding agents. Microsoft pairs lean process mapping with AI. Agents then take on routine work in a process that already runs well.