Bayesian Statistics and Supply Chain
Supply Chain can be thought of as a set of procedures that are coordinated to combine manufacturers, suppliers, warehouses, and stores in order to ensure proper production and distribution of material of right quantities at the right location and in right time. This, in turn, ensures that the total supply chain cost is reduced while providing an appropriate service level. Improving supply chain is one of the critical issues that enterprises are focusing on to gain competitive advantage. In fact I saw first hand as an intern at Amazon just how significant this can be.
Application of Bayesian Statistics in Supply Chain
Bayesian Network is a statistical model that is capable of calculating the posterior probability distribution for any unobserved stochastic variables, provided the observation of complementary subset are available. This approach has been recommended by experts as a comprehensive way to determine the relationships and influences which might exist among different variables.
It is a directed acyclic graph consisting of nodes that represent variables and arcs which represents conditional dependencies among the variables. Some of the advantages of Bayesian Statistics compared to other approaches to model supply chain disruptions includes compact representation, ability to operate with different variable types, robustness to small alterations, ability to handle incomplete datasets and facilitation of prior knowledge.
The Bayesian Statistics Network allows managers to monitor the performance of their supply chain, and or a particular section of their supply chain cycle. Using the Bayesian Network framework, managers can also try out different scenarios in order to make sure the sufficiency of their outcomes, before implementing them in reality.
The prerequisite for preparing a Bayesian Network for supply chain management is to define what different types of entities must be considered. As Bayesian Statistics is based on probability theory, the entities which are represented by nodes have a more probabilistic nature. For example, the frequency of production planning which might be conducted once or twice a week, will not be thought of as an entity. Whereas, the efficiency of the same production planning that has a probabilistic presentation will be included in the network.
In Bayesian Network, connections among the nodes are defined on the basis of the influence among them that is dependent on the context. For example, the efficiency of production planning will be influenced by the availability of raw material in the supply chain process. Thus, as this example illustrate, the accuracy of the Bayesian Network is dependent on the overall understanding of the supply chain.
In today’s business environment, Bayesian Networks are used as an alternative to artificial intelligence, map learning, heuristic search, and language understanding. In the supply chain management network, many companies confront various uncertainties on a daily basis. With this situation in mind and various useful applications of Bayesian Network, it can prove to be an effective means of forecasting the reliability of system given disruption in the supply chain network.
Want to learn more?
I’ve spent 3 years teaching over 400 professional data scientists how to build Bayesian models. I’ve put this together and included examples from marketing, policy analysis, sports analytics, and e-commerce into a course called Probabilistic Programming Primer. If you’d like to sign up to a free email course – you can sign up here. These methods have been applied at world leading organisations like Vandervilt University and Harvard and at companies such as Uber, Hotels.com, Zopa, Facebook and Quantopian.