Taking control of the supply chain

Building flexibility and resiliency into the supply chain relies on data transparency, exchange between stakeholders based on trust, and the integration of that data across the supply chain. Delegates at this year’s Automotive Logistics and Supply Chain Global conference were looking at ways to centralise data and establish control towers to better manage operations and anticipate disruption, as well as make faster decisions. That now includes the prospect of using generative and agentic AI to automate complicated processes, such as assessing the impact of cross-border tariffs.

Put simply, a control tower is a way to compile and centralise a lot of information in one place, providing end-to-end visibility of the supply chain and enabling quick decisions to be made with data on parts and materials shipments. Given the number of information sources and applications feeding into it, a user either has enough data to make decisions themselves or automates those actions, including through the use of artificial intelligence (AI).

“Either way, you have to be careful that the data you’re generating or consolidating is actionable and is going to the right audience,” said Skotti Fietsam, senior vice-president of supply chain and chief information officer (CIO) at Accuride, a supplier of forged aluminium and steel wheels for commercial vehicles.

Fietsam explained that one use Accuride made of its control tower is to get visibility on ocean freight shipments from the moment they leave the supply location, through the distribution centre, and onto the customer destination. “Having that information at our fingertips is clearly helpful to give the delivery dates to our customers,” said Fietsam.

It is also of benefit to Accuride’s management, who can see exactly how much every container is costing thanks to the transparency on each step in the shipment process. Furthermore, Accuride is using AI in its freight audit and payment system. 

“We know if we’re getting charged or invoiced for the right price, for the rates that we’ve locked in,” explained Fietsam. “We build those into the system and we have categories of payments. Our freight audit will only pay those categories up to the amounts that we set.”

Dorothy Ashford, vice-president of enterprise accounts for automotive at ITS Logistics, said that a control tower is essential to aggregate the data from several systems accrued through its strong growth in automotive over the last four years. “The task is to get all of the systems in one place, [make the data] visible and connect everybody so they can see it,” she said.

“As a carrier and 3PL, we also need to support our customers and add value for them,” added Ashford. “That includes analysing trends and anomalies, and bringing innovative ideas to the table.”

Chris Cutshaw, vice-president of business development and market solutions at CH Robinson Managed Solutions, said a good control tower is one that’s automating transactions and connecting various data feeds. According to Cutshaw, building a control tower that is good for the customer means business growth for CH Robinson.

“Where we see amazing scale is when we can tie a unified technology platform that is globally distributed,” he said. The company’s Navisphere purchase order management tool uses automation and increased visibility to streamline orders. It makes it easier to control the entire process for international shipping activities, from order creation to delivery. 

What is as important as the data feeding into the control tower, is the local expertise of the people operating it and acting on the data gathered. Those domiciled teams are connected globally at CH Robinson. Cutshaw said that the orchestration layer the company has been working on to manage the data in its control towers using AI is not about replacing people but “unlocking” them.

System integration options

AI-enabled control towers promise greater tracking and delivery accuracy by quickly amalgamating a large number of data sources, but a first principle, according to Fietsam, is properly defining the problem it is you want to address. Is the problem delivery accuracy or is it that inventory is too high? Is quality an issue? She said once those things are defined, then it is time to look at speeding up the process.

“We have all these integrative tools and we don’t have to design them in-house, but we do have to know which ones work with our systems the best,” said Fietsam. “My advice is research the systems that integrate the best. Make sure you know what the problem you’re trying to solve is, define it, and then go after the solution that fits that problem.”

Dr Oleg Gusikhin, senior director of supply chain analytics at Ford, pointed to the use of AI in analysing the impact on the automotive supply chain of changes to master production schedules. That impact might occur in different areas and Gusikhin identified the use of premium freight, the capacity pressure on the suppliers and the impact on inventory as three possibilities. AI agents can be used to analyse these areas through a digital twin of the processes and an orchestration layer that takes all of the different agents and provides a final range of potential impacts.

“The agents can analyse and optimise, and [then indicate that] you could distribute this change over a couple of months, and you might minimise any negative impact and increase profitability of the vehicle,” observed Gusikhin. “I think many of us experience these type of changes in demand quite often and it’s one of the big challenges in the industry.”

Gusikhin said that the development of digital twins (or virtual replicas of a system) was driving supply chain analytics at Ford. “We have evolved from being simply data-driven to being digital twin-driven,” he said. “It’s not only the technology, we view it also as organisational principle for how we develop everything.”

Gusikhin went on to explain that Ford was looking beyond data aggregation to the creation of a connected, dynamic and predictive representation of the supply chain. That fits with Ford’s digital transformation and the integration of AI technology. 

“Right now, we can see the development of the supply chain digital twin as one of the core drivers behind the technology innovation and digitisation of Ford’s supply chain,” he said.

Large language models

Cutshaw pointed to CH Robinson’s use of large language models (LLMs) to manage the magnitude of queries that it receives. An LLM is an AI program that understands and generates human-like text by learning patterns and context from large amounts of text data.

“We get about 4m inbound questions or queries that come to us from customers, from suppliers, from carriers, so we trained an LLM to listen to these requests and we register AI agents against that; it’s a higher level LLM,” said Cutshaw.

An AI agent is a software system that can use tools, plan multi-step actions, remember past interactions, and adapt its behaviour over time.

“Our goal in the next 18 months is to eliminate people responding to about 50% of those requests and take it through agentic AI; it is a major step change for us,” said Cutshaw.

Ashford said ITS Logistics is building AI in to support track and trace, and to support analytics and tools to benefit the customer. “It is for digesting large amounts of data, and then in addition to that we are focusing on integration and internal processes, looking at how we can take unstructured data and make it work for us, and improve productivity,” she said.

Securing the data

It is also vitally important to consider cybersecurity when using AI in control tower data gathering and automating processes. Cutshaw said the first six months of its AI project was focused on data security and privacy. “You absolutely have to get that right… and we don’t send that outside of our four walls because it’s proprietary information.”

John Rich, manager of data analytics and AI Programs for Mazda North America Operations, said security and governance was one of the biggest gaps to fill as AI solutions proliferate, and it is crucial to ensure that data is not leaking out and being exposed to external models.

“When we look at the compliance area of AI right now, there are so many different options. There really is a war going on right now in the AI space,” said Rich. “You’ve got all these companies investing hundreds of billions of dollars to get ahead – Meta, OpenAI, Anthropic, and overseas with DeepSeek from China – so how do you evaluate all those different things?” 

Rich said it is very important that companies using AI ensure they have people focused on security, and that they are working with cybersecurity and legal advisors to be aware of risks. Incorporating security and governance into the data pipelines is essential. 

“How do we make sure that the AI agent is using the right user principal ID to interact with the data so that I’m not giving it access to data that it shouldn’t be based on?” asked Rich. “Who is using the agent? There are a lot of different complexities.” 

Rich said users need to stay up to date and make sure that there is governance not only of data but of any solutions that a company is building on top of that data.

Identifying demand gaps in minutes

In a live demonstration at the conference, John Rich demonstrated how AI can be integrated into automotive logistics and supply chain problem solving. Prior to the conference, Rich ran some Python code to install the demonstration, which took three seconds to process and one minute to import. With that he created a dashboard that provided demand and forecast shipment recommendations.

“In the average user’s hands, it’s important for them to understand what they’re looking at, and one of the greatest things that AI can do is allow those users to communicate with the data directly in natural language,” said Rich. “So not only can I look at this report, but I can have an AI agent explain the data sent to me and tell me what’s in here.”

In the live demonstration there were six tables that showed shipment recommendations, bill of materials, raw materials, supply demand, as well as product demand both historical and forecast. “That’s great and everything, but I really want to see which distribution centre has the highest chance of being a bottleneck,” said Rich. “So not only is [the AI] going to go and search for that information, it’s going to write the code behind the scenes to identify the distribution centre that is most likely to be a bottleneck, and tell me what the biggest difference is.” 

The AI-empowered dashboard showed that Distribution Center 4 had a 38,700 part gap. Going one step further, however, Rich then demonstrated live that he is able to expand on that information by altering the SQL code the AI agent wrote behind the dashboard which was limited to one result. He then reran the request to look at all of the possible demand gaps.

“Distribution Center 4 is the most obvious problem [but] it looks like I’ve got a demand shortage across all my distribution centres, and not only that, it’s created a nice little visual for me as well,” Rich revealed. “So, in five minutes I can do that, and it is all done using a Databricks free account that costs zero dollars. Imagine the possibilities of what you can do with all your enterprise data and enterprise level computing and processing in five minutes on an iPad.”

Tackling the tariffs

Under the Trump administration in the US, import taxes have had a disruptive impact on the automotive supply chain, increasing costs and complicating the rules of origin on parts and materials being moved cross border. Shippers and logistics providers need to know about the part number, the percentage of USMCA content, the percentage of steel and aluminium affected by the tariffs, and all of the destination points. Fietsam made clear that tariffs are not just about from where parts are coming but also where they are going and how many times they cross a border. This is an area that AI technology promises to provide much needed assistance in, but more data is needed, according to Fietsam. 

“We’d like to use generative AI, but there’s not enough data yet to support an AI model,” she said. “What you first have to do, and I’m sure everyone in automotive is going through a similar exercise, is make sure that your own part number data is accurate. What we can’t do yet is get all the information from the government websites.”

Manually matching parts content, origin and destination, and tying that to the US tariff harmonisation schedule introduced in April is a labour-intensive process that changes every month. 

“If there’s a way to know when that change occurs, and instantly get alerted from the governmental databases that hold that information to tie it into ours, I think it will be awesome,” said Fietsam.