From firefighting to foresight: how automotive supply chains are learning to trust their data

Watch: ALSC Europe 2026- Welcome Drinks Reception Wrap up

Watch the ALSC Europe 2026 welcome drinks reception video

Every supply chain leader tells the same story: more data than ever, yet still stuck in firefighting mode - Marcin Mercik, head of supply chain management at ZF Lifetec stated at the ALSC Europe 2026 conference which kicked-off today. 

The panel, which included leaders from ZF Lifetec, Stellantic, Blu Yonder and Shippeo drew the consensus that AI and digital tools can transform automotive supply chains, and significantly. The question was what it actually takes to get there, why so many organisations are stalling, and what the honest path forward looks like.

Building from the ground up

Mercik's outlook on this was structured around what he called a hierarchy of needs: three layers that must be addressed in sequence, even if they are built in parallel.

The foundation is process and data standardisation. "You cannot cover the gap on process and data with artificial intelligence," he said. "This will not fly." At ZF Lifetec, that means cleaning master data, building an advanced planning system, centralising transport management and extending warehouse management.

The second layer is connectivity: getting real-time or near-real-time visibility across the supply network, from advanced shipping notifications and supplier portals to overseas material tracking integrated directly into SAP so that every planner can see how vessels and containers are moving.

The third and most ambitious layer is what he calls decision intelligence. The digitalisation and ultimately automation of decisions, with humans in the loop but supported by AI at every step.

The practical question of whether these layers must be tackled in sequence or simultaneously drew a response that captured the reality of running a complex business through transformation. "There is no other choice but to go in parallel," Mercik said, "but it is clear that in order to be successful with layer three, you need to address layer one."

For Benoit Gaucherand, senior manager of aftersales transport network design and cost control at Stellantis, that framework landed with resonance. Stellantis's aftersales supply chain operation - spanning multiple legacy systems inherited from its constituent companies - is at an earlier point in that journey than ZF Lifetec, and Gaucherand said so during the panel discussion at the event today.

"We are at the early beginning. Really at the early beginning. We are trying to understand what can we use, what kind of data we should gather. We have also internal constraints working with different systems from our past companies. So we need to merge together all the data," he pointed out.

Gabriel Werner, global vice president of end-to-end solution advisory at Blue Yonder, reinforced that the data problem runs deeper than most organisations acknowledge. In his experience working with automotive customers, data is distributed across systems numbering not in the dozens but easily in the hundreds.

"Forget AI for a second," he said. "What that means is people that are involved in a decision making process may look at different versions of the truth and that becomes a fundamental problem." His challenge was to start thinking of supply chain as a single system rather than a conglomerate of underlying systems.

Werner was equally direct about where ownership of the data problem must sit. "The people that rely on the quality of the output, they need to be the ones that are responsible for the quality of the input."

The pattern to generate a forecast with technology, then run a consensus process where every stakeholder adjusts the output is, in his view, precisely backwards. In the age of AI, managing the output must give way to managing the input. "We're all familiar with good old garbage in, garbage out. But in the age of AI it's going to be garbage in, very elegant garbage out with a lot of confidence," he said.

The trust problem - and the people behind it

Pierre Khoury, CEO and co-founder of Shippeo, brought an important counterweight to the foundation-first argument.

Two years ago, he was sceptical about AI agents himself. "I said, hey, that's a tool, that's a toy. What's in it for all users?" he told the panel. Working with Stellantis data, he has seen what AI agents can surface that no human analyst could find - patterns and relationships which no human can see. "It's impossible because the way the data is managed by the agents is crazy."

But, the technology is only as useful as the trust placed in it by the people who use it daily. Gaucherand and Mercik both identified the same human challenge: finding people who can bridge the worlds of data and operational process. "We have a similar problem to Marcin," Gaucherand said. "To find the people able to match the data with the processes. And that will be the challenge. If we want to develop AI, to develop real track and trace, we need to have somewhere this link between data and processes. And that's the biggest challenge today."

At ZF Lifetec, the AI competence centre sits in India, staffed by IT specialists who are strong on AI but less familiar with operational supply chain processes. "Probably this is the next step for us," Mercik acknowledged, "how to join the process expert with AI expert, or preferably to have such resources that one person understands AI and the process at the same time."

Mercik identified the specific consequence of getting this wrong - what he called the black box paradox. "We may fall into the black box paradox, that we have very sophisticated AI, a lot of data inside, then we have an outcome. But then this MRP controller gets some suggestion and then he is asking himself, should I trust? Maybe I shouldn't trust, because I have no idea from where it is coming."

Explainable AI (outputs that can be interrogated and understood) is not a technical detail but a prerequisite for adoption. Werner offered a practical map for how explainability can be achieved across AI types:

Predictive machine learning models can pinpoint precisely which input data drove which output change,

Generative AI can explain the reasoning of decision engines in language planners understand and,

Agentic AI, by its nature, can simply be asked to account for itself.

Khoury's recommendation for building that trust was grounded in operational reality rather than technology strategy. When he visited a plant in Zaragossa to understand how shop floor workers would respond to AI tools, the reaction surprised him. "They say, I want to start now. I use ChatGPT all the day long, so I exactly know how to use it."

His stated that the industry is over-complicating adoption from the top down when the appetite and capability to start already exists at the operational level. "Go in the shop floor, just go with the users, ask them what they need and build from there," Khoury said.

From visibility to decisions

The question on whether organisations wait for master data to reach a certain quality threshold before deploying AI agents arose to which Werner highlighted that waiting is one of the most fundamental misconceptions in the industry.

"AI is the single most accessible technology that mankind ever came up with. It is actually something that can help you climb the maturity curve. It's not a reward at the end of having climbed it." His recommended approach was to pick one decision in the supply chain where speed or quality is suffering, trace the data that supports it, and start there.

"The AI use case will almost present itself." Khoury added that AI agents can themselves become instruments for improving data quality adding that Shippeo deploys data quality agents that identify root causes of compliance gaps by lane, by day of the week, by supplier.

In practice, the Stellantis aftersales operation illustrates this step-by-step logic. "For the real time visibility, first we want to provide information to our customer, to provide visibility to our teams. That's the first step," Gaucherand explained. The immediate priority is basic but critical - giving customers and planners visibility of truck location in real time. "Where is the truck that we are sending to you?" From there, the plan is to layer AI to make that visibility faster and more actionable, not to leap immediately to full network optimisation. "Then the next step: use all the data that we have to optimise our network. But that will be far longer."

Werner drew a distinction that reframes how visibility tools should be evaluated entirely. "Visibility without being connected to a decision-making process is just a dashboard," he said. The value is not in seeing - it is in acting on what you see. He described how generative AI functions as an orchestrator in this model: surfacing insights, bringing problems to attention detected through visibility, and aligning stakeholders across the decision process. 

"There's three real people on the chat and two agents on the chat and they're discussing a business problem," he described of one current deployment, "and the agents have access to the same systems the users have."

Gaucherand brought it back to the most accessible entry point of all: take any question that feels routine and repetitive, put it to a generative AI tool, and learn from what comes back, including the limits of the answer. "The question is key," he said.

"We need also to teach our people to ask the right question to be precise enough to get the right answer." The payoff is already tangible: "In five seconds I have an answer that I would probably need 20 minutes on the internet to get."

Stay tuned for more insights from ALSC Europe 2026! 

Checkout the agenda for Day 2 below!

Automotive Logistics & Supply Chain Europe

17-19 March | Bonn, Germany