Data analytics in FVL
Carmakers embrace AI agents to improve vehicle logistics processes
Experts revealed how technologies such as agentic mesh systems, JEPA models and data-sharing protocols are transforming route optimisation, ETA accuracy and OEM-carrier negotiations.
Carmakers and tier suppliers are working with technology providers on where the latest digital tools can optimise operations and speed decision making, as well as mitigate cost pressures in finished vehicle logistics. They are also looking at AI systems to support negotiations with transport providers to avoid disruption and unnecessary cost. At this year’s Automotive Logistics and Supply Chain Global conference in Michigan, speakers from Ford, Mazda and Cognosos, outlined some of the latest technology, where it is being applied and why.
John Rich, manager of data analytics and AI Programs for Mazda North America Operations, talked about some of the latest cutting-edge technology starting to have an impact on the finished vehicle supply chain.
“Agentic mesh is really the future,” said Rich, pointing to the creation of teams of AI agents dedicated to specific tasks. Using the example of a chatbot, Rich explained that while the user interacts with one interface in natural language, behind the scenes there is a team of AI agents finding context and gathering references.
Creating a team that has an orchestrator function and focusing it on specific parts of the finished vehicle supply chain, such as visibility on inventory levels or demand forecasting, promises greater efficiency in operations. It is important, however, to keep the human in the loop and use the technology to propose actions for the user in the decision-making process.
Advertisement
Rich said that rather than working for hours or even days to gather all the information necessary to make a decision, agentic mesh technology reduces that to minutes or hours.
“Having a team of AI agents or an agentic mesh solution to deliver all the information I need [means] I can click a button to make a decision and fundamentally change my supply chain over the course of weeks or months; it's really dynamic. I would say agentic mesh is the future.”
Rich also pointed to joint embedding predictive architecture (Jepa) and its potential to improve the vehicle logistics supply chain in the next five to ten years.
“What Jepa does is take the gaps in strict knowledge-based large language models (LLMs) and combines it with data to understand trends,” said Rich. “It's almost like an AI specifically made for machine learning. So when you look at logistics and supply chain, there are actually a ton of applications for this.”
Strong data foundations
Effective digitalisation of logistics and supply chain functions relies on a solid data foundation on which to build an architecture that supports the interconnections. Rich pointed to data mirroring and tools such as Delta Sharing Protocol (developed by Databricks) for secure data sharing between departments or organisations regardless of individual computing platforms.
“I don't have to move the data or copy it to give access to a supplier or between locations, I can leave it in one place and use Delta Sharing Protocol or data mirroring to pull it in via federated queries, building a sort of agentic mesh layer on top of that data without the need for duplication,” said Rich.
Rich pointed to technology players including Snowflake, Microsoft and Databricks which are looking at the need for greater interoperability.
“When you think about the true gamut of what the automotive space covers, all the way from the dealer back to your component suppliers, there's so many different areas,” said Rich. “To continue moving forward, there has to be that seamless sharing of information so that you can truly optimise from end to end in your supply chain.”
Leaking legacies
Anthony Butler, senior director of product at technology provider Cognosos, said that a robust application programming interface (API) was important to avoid getting locked into legacy systems that lacked interoperability and made data extraction and sharing difficult.
He also noted a consistent gap in implementation by failing to connect the teams in technology and operations, again pointing to the importance of keeping the humans in the loop. “When you bring those people together can really come up with some meaningful cost savings and optimisation improvements for your business,” said Butler.
Good reliable data is essential before you can do anything more technologically advanced, according to Butler. “AI can't do anything without good reliable data and I think that the industry has been focused on the right things and has made a lot of strides with to get consistently better, more reliable data that you can do things with,” he said.
Ford is looking at better route optimisation using tools that support closer partnerships with logistics providers given the unique challenges in the finished vehicle logistics space, according to Michael Arnold, manager of North American Vehicle Logistics Strategy and Planning at the carmaker.
Ford is working on a number of different scenarios that the finished vehicle sector is faced with and assessing whether it has the right partners, the right systems and the right network in place. That includes having the right route optimisation tools and Ford is adapting technology to deal with the pressures on outbound operations and support teams by automating certain tasks where it can, said Arnold.
“We need partners that we can rely on that can be flexible from both an operations and an IT perspective, able to make decisions and make movements of vehicles in very little time,” he said.
ETA accuracy
Getting better accuracy on delivery times remains a challenge in the finished vehicle supply chain and data analytics is increasingly being used to improve the order to delivery cycle.
Butler said that there are two particular applications Cognosos is working with to support its customers. One is on high-level supply chain visibility to identify which ports are processing customer vehicles the quickest and which routes are most efficient to send vehicle volumes through.
The other is in driving efficiency into outbound shipments by identifying where vehicles are in dwell the longest and working to make those sites, be it yard, terminal or port, more efficient. “A lot of the focus is trying to get more efficient sites and trying to get the sites processing vehicles faster and getting them out,” said Butler. “That helps their ETA out because they have fewer barriers and less variability in their overall supply chain and can have better predictions because of that.”
Identifying and preventing very costly misdirected shipments is also an important area where digital technology is being applied. Butler said the cost to a logistics provider of putting a finished vehicle on the wrong rail car and having to get it back to where it needs to be is between $15,000-$20,000, a significant dent in a working margin of 5%. Cognosos has technology to detect where the lost vehicles are and creates alerts if a vehicle is on a rail car when it should be on a truck or ready to ship in the wrong location on the wrong transport mode.
Rich highlighted the role of the process analysis concept known as Poka-Yoke, or mistake proofing, and how important data and analytics teams are in identifying potential areas of error using machine learning in avoiding cost of recovery.
“There are ways using data driven mathematical equations to determine what a true outlier is and how to account for it in your models,” said Rich, adding that the costs of running the models had reduced significantly. Rich pointed to the open-source model MLFlow from Databricks that teaches users how to create a continuously iterative approach to machine learning models that includes labelling and classifying outliers so that models are always kept up to date.
Clear consultation
Digital tools are also helping to predict pricing trends and capacity constraints to support OEM-carrier negotiations in finished vehicle logistics. Better, more reliable data is definitely helping in those negotiations by providing better visibility.
“So many of the tensions that you see between the carriers and the OEMs comes down to a lack of clarity on what actually happened,” said Butler. “One of the things that we're supporting clients on right now is detecting rail impact for vehicles that causes damage. That is a major source of contention.”
Rich pointed to translytical data flows, which automate end-user actions such as updating records, adding annotations, or creating workflows that trigger actions in other systems. Beyond simply filtering the data in a particular report, it provides the ability to update the report and feedback the information to the database.
OEM negotiations with carriers on pricing for example can be conducted by an AI agent that is able to collect all of the information from an email exchange or portal and put the information into a model to provide price quotes and optimise the price margin.
Looking ahead Rich said that the most effective application of digital technology is to create a very specific training model for AI. “You can create a very hyper-specific use case for AI because you don't need your models to be able to do everything,” he said. “You want it to be very focused and specific on executing a specific task or set of tasks so that it can learn from itself and do things more efficiently over time.”
