AI agents are taking over our supply chains
Generative artificial intelligence (GenAI) enables a radically different approach to the supply chain. Users no longer have to fiddle with complex systems themselves, but can talk to their own ‘agent’ who generates schedules, analyses scenarios, makes recommendations and executes decisions on their behalf. During Webinar Wednesday, Martijn Lofvers of Supply Chain Media spoke to Gabriel Werner of Blue Yonder about the implications of this development. ‘The question is whether it is still necessary for those AI agents to ask for permission to execute decisions.’
By Marcel te Lindert
Global turbulence has put the discipline of supply chain management in the spotlight. Many companies have secured a place at their boardroom table for the chief supply chain officer. ‘Perhaps the attention is fading a bit now that companies have become accustomed to all the disruptions, but those disruptions will not go away,’ argues Martijn Lofvers, Supply Chain Media’s chief trend watcher who is hosting the webinar.
He is joined by interlocutor Gabriel Werner, chief technology officer at Blue Yonder. ‘The companies I speak to are wondering how to keep their supply chain running in these turbulent times. They often have a fragmented IT landscape, which limits their options. While companies need to bridge the gap between supply chain planning and supply chain execution right now. Companies will have to design their decision-making processes to transcend both these domains. That calls for supply chain orchestration.’
Point solutions
In supply chain orchestration, Werner distinguishes five maturity levels. At the lowest level operate the companies that use point solutions for such things as demand planning, inventory management, warehouse management, transportation management and other disciplines in the supply chain. If those point solutions are connected at all, it is through interfaces that allow data traffic only at set times. ‘That means there is a delay in the flow of information through the business,’ Werner argues. ‘As a result, there is also a delay in decision-making processes, which hinders an agile supply chain. Because that decision-making is based on data from different systems, the quality of decisions is also under pressure.’
Take a consumer products manufacturer who has run a successful promotion in Germany and wants to extend it. If he indicates this in his promotion management system now, it might take a day for the supply chain planning system to pick it up. This system’s calculations show that extending the promotion requires moving stock, which leads to transport movements. ‘In this way, it only becomes clear after a few days that transport capacity is lacking because of a road transport strike. And that the distribution centres know of nothing and therefore have not scheduled enough employees to process all the extra order lines.’
Generative AI
Companies with higher maturity levels have found a solution in creating a data warehouse or data lake. This makes data from different systems readily available to the entire company. ‘But data integration is not enough. Companies also need interoperable workflows: processes that enable cross-functional collaboration,’ Werner argues. ‘That allows promotion and supply chain management to be integrated into one process, eliminating delays in data exchange. Decision-making is faster and better, which means that that road transport strike no longer has to be a problem. And giving distribution centres time to adjust their staff planning.’
Once companies have reached this maturity level and have an integrated system for supply chain planning and supply chain execution, all the conditions are in place to take the step towards deploying generative artificial intelligence. With GenAI, it is possible to create a ‘conversational user experience’: an environment where users can talk to the system via a kind of chatbot to generate a schedule, analyse scenarios or interpret data. Lofvers: ‘That means users no longer need lengthy, in-depth training to use supply chain software. Sales people can also start a conversation with the system themselves to learn more about promotions, stock availability or other issues.’
Role-dependent agents
Blue Yonder deploys GenAI to create ‘agents’ that help users deploy supply chain software. Werner compares these AI agents to Copilot, Microsoft’s AI assistant. ‘Only we soon discovered that supply chain is too complex for a ”one size fits all” Copilot. So we are developing 15 different agents for various roles in the supply chain. For instance, we have agents for warehousing, transport, supply planning and demand planning,’ says Werner, showing how an agent creates a briefing for the warehouse manager every morning. ‘Normally, five to seven people need half a day to cough up that information, so you lose half a day. The agent has that information ready immediately.’
In the future, the capabilities of such AI agents will only increase. Now it is all about interpreting and presenting information, but soon they will also be able to analyse the cause of problems and make recommendations based on that. ‘The next phase is for the agent to implement those recommendations itself. At some point, the question arises whether that agent still needs to ask for permission to implement a decision or can just go about its business independently without human intervention,’ Werner argues.
Robotic process automation
Werner emphasises that AI agents do not generate a schedule or forecast themselves, but use the software available for that purpose on behalf of the user. This reminds Lofvers of robotic process automation (RPA), in which digital robots are used to automate simple tasks, such as transferring order data from an e-mail into the ERP system. ‘Five years ago, this concept attracted a lot of attention. So actually an agent is a more advanced form of that digital robot,’ the chief trendwatcher argues. Werner agrees. ‘With the difference that that agent has intelligence and actually understands business operations.’
The final maturity level in supply chain orchestration is reached when the deployment of agents is not limited to a company’s internal supply chain, but extends across the entire network of customers, suppliers and other supply chain partners. Then all partners have a single version of the truth, allowing them to make decisions in full alignment with each other. Lofvers recalls Professor Hau Lee, who has been advocating such networks for years. ‘If you want to avoid hang-ups in the supply chain, you need to share data directly with all supply chain partners. So you need a platform and a network for that.’
