End to end supply chain visibility – the ability to track and monitor every stage of the supply chain, from raw material sourcing to final product delivery in real time – is the holy grail for supply chain managers. And it is easy to see why; comprehensive oversight enables businesses to optimize operations, improve supply chain resilience, and respond proactively to the inevitable supply chain disruptions. However, while clearly beneficial in theory, in practice it has been difficult to implement without top-down centralized control from large enterprises – and even then, challenges remain with the sharing of real-time data.
For many small and medium-sized companies, it remains a constant challenge to respond to the myriad of unforeseen supply chain disruptions. Perhaps the most infamous example of a global supply chain disruption was the Covid-19 pandemic, where supply chain management became a household term – however, smaller and more localized disruptions such as natural disasters, geopolitical conflicts, demand fluctuations remain a frequent occurrence in which effective mitigation plans are required.
Deep Learning has become famous in the consumer space with generative AI chatbots; however, it can also be implemented in supply chain applications to more effectively predict various supply chain events as part of an advanced analytics supply chain strategy. Still, like everything with AI, the biggest challenge is often not in the advanced models themselves but getting the relevant data to train models that can make accurate predictions.
There are two main challenges with getting the data to feed into AI-Powered advanced analytics: 1) Collecting, organizing, and storing internal data along with 2) coordinating the sharing of data with external parties. Each of these can be a massive undertaking for an organization. Simply getting a company’s own internal data infrastructure setup can be a complicated and drawn-out project. However, this challenge is relatively straightforward compared to the much more daunting task of figuring out how to efficiently and securely share supply chain data with partners throughout the supply network.
Data sharing limitations are in place for a multitude of reasons, including privacy concerns, security of intellectual property, and regulatory requirements. For these reasons, most organizations are often very reluctant to share their data with external entities. However, this practice hinders the ability to achieve real time supply chain visibility – to the detriment of all parties involved.
There are various attempts at solutions such as ERP implementations, RFID technology, and Blockchain applications – however, each of these have their own limitations. Legacy ERP systems are bulky, have difficulties integrating data from diverse sources, and process data in batches – slowing down the flow of real-time information. RFID technology, while it enables the tracking of assets across the network, still only leverages internal data and not network data. Further, RFID, including asset tags and the supporting infrastructure, can be cost prohibitive to small and medium sized businesses. Blockchain technology is promising for facilitating data sharing across parties, however it still runs into issues with energy efficiency, universal agreement among supply chain partners, and overall lack of scalability.
The fundamental challenge for AI and advanced analytics to solve for supply chain management is – how to facilitate the exchange or data across the supply chain without compromising privacy or security that accompanies sharing raw data? It turns out the solution is the result of the very nature of how AI models work.
Researchers Alexandra Brintrup and Ge Zhenga at the University of Cambridge recently completed a set of experiments that show how deep neural networks can do just this. Their research found that they can take sets of separate data with various information related to buyers, suppliers, products, locations, and certifications – make predictions about various linkages – and then, without sharing any of the raw data between entities, combine the model predictions across entities to enable the aggregated global model to make more accurate predictions than the isolated local models.
The dataset they utilized to run their experiments was from the automative industry and contained over 40,000 companies, 2,300 customers, 79 countries, 5 certificates, and 927 products. From this dataset, they constructed a supply chain knowledge graph, which is a data structure that holds information related to links and nodes of a supply chain network. Knowledge graphs are commonly used in social networks and biological networks, but can also be leveraged to analyze supply chain networks.
From this graph dataset, the objective was to then to train a deep learning AI model to make predictions about links between the nodes throughout the supply network to uncover various patterns and interdependencies. To do this, the researchers leveraged a graph convolutional network model, which has the distinct characteristics of being able to pick up interdependencies in a supply network. This is similar to what the more common convolutional neural networks do with image pixels in computer vision applications – except for a supply chain knowledge graph dataset. More information about Graph Convolutional Networks can be found here.
While the supply chain knowledge graph and link predictions made by the Graph Convolutional Network can be helpful – the point is to be able to leverage separate data sources in the model without sharing raw data. This is where Federated Learning comes into play.
The experimenters segregated their data into distinct entities to simulate how real-world data is normally siloed into each individual firm’s private database. Then on each of these siloed datasets, they trained a local Graph Convolution Network model. These local models learn on features and interdependencies that are unique to their specific scope of the supply network. The next step is the key to how federated learning can share information between entities without sharing the underlying raw data.
The objective of machine learning models is to train a set of parameters, also known as weights, to be able to predict an output given a certain input. What they aim to predict could be anything; weather patterns, energy consumption, equipment failure, credit risk, or in this case, linkages between supply chain nodes. The key then, once each of these disparate local models have been trained on their subsection of the supply network data, is to share these parameters into a combined global model. The global model in federated learning can aggregate the parameters from all the local models, without sharing any of the raw data.
The critical finding from Brintrup & Zhenga was that this global model, aggregated from the trained parameters of the local models, has more predictive power than any of the individual local models. This is the case because AI models need to be trained on rich, diverse datasets to be most effective. The datasets at individual firms are not going to be as rich and diverse as the combined dataset of the entire supply network.
This novel federated learning approach to supply chain knowledge graphs can help solve the age-old problem of leveraging the datasets of the entire supply chain without the privacy or security risks. This can be especially useful for small and medium-sized businesses that have much smaller datasets than large companies. Further, each company in the supply chain is going to have separate systems with various database architectures and schemas. Trying to integrate raw data across disparate systems can require a massive effort and costly changes to an organization’s data infrastructure. With this sharing of model parameters approach, each entity can keep their current systems and schemas intact, but still share the information through the trained model weights. More information on Federated Learning can be found here.
While this represents a great experimental breakthrough, the most important part is what this means in practice for procurement managers. Using this combination of a Supply Chain Knowledge graph and AI-powered advanced analytics, there are several important areas of interest for managers:
Proactively Identify Disruptions
GCN and FL with supply chain knowledge graphs can give procurement managers the tools to proactively identify potential disruptions, from wherever they may emerge including natural disasters, geopolitical conflicts, demand fluctuations, and other supply shocks.
Mitigate Risks
Mitigating risks is about having enough time to formulate a plan to either completely alleviate the potential danger or at least reduce the impact. Proactively gaining improved supply chain visibility can provide improved resilience and result in significant savings. For example, during Covid, Walmart was able to leverage their advanced analytics across their extensive logistics network to continue sourcing products for consumers. While some items were briefly out of stock, they stood out for their ability to quickly identify a problem and devise an alternate solution.
Enforce Compliance and Quality Standards
Ensuring compliance and quality standards are met across the upstream supply network can be a daunting task. However, being able to leverage information across entities related to real-time supplier certifications can provide the necessary insights to improve compliance and quality.
Identify New Opportunities
New opportunities can take many shapes. Identifying new sources or strategies to improve cost, quality, and lead time can provide enormous benefits. Supply chain demand and inventories can change rapidly and being able to proactively anticipate these changes can lead to unbounded new opportunities.
Manage Supplier Dependencies
Oftentimes, throughout a supply network, there can be various dependencies that are not apparent to downstream firms. But nevertheless, they are still exposed to the risk. For example, an OEM may source metal fabricated components, but these fabricators in turn may all utilize the same raw material vendor. If this raw material vendor suddenly ceases operations, for any reason, this could be detrimental to the OEM, even though the risk may not have been immediately apparent.
Spot Purchasing Patterns
This goes along with why it is critical for organizations to regularly conduct spend analyses to understand their own purchasing patterns and where efficiency lies. This can be taken to the next level of insight by understanding purchasing patterns of the broader supply network.
Devise Sourcing Strategies
Improving real-time supply chain visibility through AI-powered tools can help organizations stay more in tune with real-time market dynamics and proactively adjust their procurement operations accordingly. For example, organizations can get more efficient visibility into whether a linchpin source is potentially going to turn into a bottleneck and adjust their strategies accordingly by formulating ways to diversify their supply base.
To leverage these state-of-the-art AI methods, it is essential for an organization to be as digital as possible – however, as mentioned above, GCN and FL do not require the systems to directly integrate. This means organizations are not constrained by having to worry about direct raw data integrations and all the security concerns that come with it. Rather, they can simply choose the procurement software system that best fits their needs. The systems don’t need to be the same or directly integrate raw data, but they must be digital procurement systems.
While it might seem that these kinds of solutions are mostly beneficial to large entities, the opposite is true. The machine learning methods discussed here are well suited for smaller datasets. These smaller datasets can then be aggregated across the supply network into a larger dataset that can provide considerable predictive accuracy. So small and medium sized companies can use AI to leverage data from across the supply network to get much better real time visibility into their supply chain than they otherwise could have.
Procurement software like Lasso can help small and medium sized businesses digitize their procurement and sourcing operations and get data organized and stored so that it is ready to be utilized by AI models to deliver cutting-edge, real-time insights.
support@lassoprocurement.com
© 2025 Lasso Supply Chain Software LLC
This website uses cookies to ensure you get the best experience on our website.
