Thanks to graph theory and Machine Learning, it is possible not only to increase visibility on one’s supplier network, to model the flows between the different actors, but also to automate the detection of potential risks throughout the supply chain. This would allow SC managers to identify threats, redefine strategies, manage contingencies and steer supplier networks in a few clicks.
As defined by Professor Martin Christopher in his book “Logistics & Supply Chain Management”, the supply chain is not exactly a chain but rather “a network of connected and interdependent organizations mutually and cooperatively working together to control, manage and improve the flow of materials and information from suppliers to end users”. Today, organisations have links with other partners: suppliers, subcontractors, co-contractors, distributors, and so on. The interdependencies and connections are increasingly numerous and complex, notably due to the globalisation of the economy and the tendency to outsource certain tasks. Consequently, supplier networks are flexible, open and hyper-connected, which therefore makes the system more fragile and means organisations have to operate in turbulent environments. In certain sectors, such as the aeronautics, automotive or railway industries, there can be tens of thousands of different players within the same supply chain, spread over several tiers, depending on the number of intermediaries involved with the end customer.
No visibility beyond the second tier
Experience in the field shows us that Supply Chain managers generally have an extremely local vision of the Supply Chain, essentially confined to tier-one suppliers. In such conditions, it is obviously difficult to anticipate risks, react quickly and minimise any potential impact, because information concerning the failure of a tier-three supplier (or beyond) may not be reported to the end customer before stock or delivery problems arise. However, it is actually possible to obtain a vision of the entire supplier network, which includes all the players, flows and their status, using mathematical methods and tools. This involves understanding and formally mapping the supplier network with the help of mathematical structures that are then used to model it. The model can then be exploited and studied using different tools and theories.
The support of graph theory
One of the mathematical theories used to study networks is graph theory, which creates simple network models with links between the different objects. These models consist of points called vertices (or nodes) and links between the points called edges. They can also be represented as binary matrices. A supplier network can be represented by a graph in which several edges are allowed between vertices, as well as with the colouring of nodes (different types of business activities) and edges (different types of flows). These graphs can, in turn, be represented by a generalisation of binary matrices, called tensors. The tensor is the input element that enables functions based on graphics libraries to transform it into a graph, allowing the data to be visually exploited. Consequently, we are now able to represent all or part of a supplier network. The approach proposed enables us to model all the activities of the network’s players, as well as the material and non-material flows governing the chain.
Determining network links
To build the supplier-network graph, we use data from the Client-Supplier ERP systems, which are increasingly connected, in order to establish a database of information about the network, consolidated by conventional audit phases carried out by field experts. Another interesting possibility would be to implement an approach based on big data to determine the links between the players in the network, like recommendation engines on e-commerce sites. Admittedly, there are still several limitations, notably linked to the unwillingness of suppliers to open up to the rest of the network, with issues concerning trust often used as a pretext. This kind of representation will already highlight the operational complexity of supply network management, but it is only the first step towards better end-to-end management of the Supply Chain.
Automating risk detection
What is the best way to rapidly identify potential supplier failures that could have an impact on all or part of your network? Business activities face many different types of threats (weather events, natural disasters, social, geopolitical or financial crises, etc.) that are difficult to monitor “manually” owing to the significant number of players involved in the supply chain. The quantity of data generated by a growing number of sources is too large – and therefore too costly – to be tracked. One possible solution might be to automate the identification of potential risks, particularly climate-related threats and natural disasters. Machine Learning (ML) would seem highly suited to this type of task, by retrieving unstructured data available on the web, preparing them, then processing them through specific models in order to rank the data according to different risk categories. If this approach provides the expected results, we could then move on to the other types of risks by creating specific models by type.
