Hello everyone, Let’s jump into the ChatGPT boat and explore if this new machine learning model would disrupt supply chain management. Quick intro: OpenAI trained an artificial intelligence called ChatGPT, which is an LLM (large language model) with the key word here being “language” as it processes words. LLMs have a new form of artificial intelligence called a transformer neural network. This new form of AI is a lot more capable than its predecessors because of its capacity to stay in context of the topic being used for, with a mechanism called self-attention. Models in Supply Chain In supply chain management, various types of models are used to support management executives in making different decisions. These mathematical models can be classified into the following categories:
Source: MIT – PhD Chris Caplice Supply Chain Management MicroMasters (mit.edu) – Lesson 2: Analytics Basics | Week 1: Introduction to Supply Chains & Basic Analysis | Supply Chain Analytics | edX
It’s essential to understand the applications of these models to comprehend what you are trying to achieve and how to go about it. Machine learning and artificial intelligence fall under the category of predictive models. They are built for predicting outcomes, such as forecasting sales in supply chain terms. However, it’s important to note that you cannot ask a Machine Learning model to provide prescriptive guidance, like minimizing costs or maximizing sales, as that is not within its design. The answers generated by these models do not come magically but are based on patterns and correlations learned from historical data. Forecasting Traditionally, forecasting was accomplished using time series statistical models to capture components such as base, trend, and seasonality. However, with the advancement of machine learning, ML-based models, particularly those based on regressions, have shown superior performance in forecasting tasks. These ML models can incorporate a wide range of variables into the equation, including promotions, price, weather, advertising, and more. The machine then learns from the data and determines the impact of each variable on the outcome, such as sales. ML/AI Regression models employ mathematical formulas to create models that explain how sales or other outcomes can be predicted based on a set of input variables. One of the key advantages of using machine learning for creating forecasting models is that they can generate models of tremendous complexity. These models can be so intricate that even a PhD mathematician might struggle to understand the mathematical equations describing sales. By leveraging the computational power and pattern recognition capabilities of machine learning algorithms, these models can capture non-linear relationships and interactions among multiple variables, leading to more accurate and sophisticated predictions. As a result, organizations can make more informed decisions based on these forecasts, considering various influencingfactors and improving their overall supply chain management and decision-making processes.
How can Large Language Models (LLMs) like GPT be used for forecasting?
LLMs are great at predicting the next word and staying in a certain context. These models are built to handle languages or words not numbers, their primary strength lies in natural language processing tasks. While ML/AI Regression models excel at forecasting numerical outputs such as demand and sales, there are ways to leverage LLMs to improve forecasting, particularly in capturing customer sentiment and other qualitative factors. The most challenging aspects of demand behaviors to model and forecast are qualitative factors such as customer sentiment news, social media trends, in other words things going viral. It’s very likely that in the near future LLMs models would feed ML/AI Regression models with data capturing qualitative data: LLMs could be trained to analyze customer reviews, social media trends, and news articles to gauge customer sentiment towards products or brands A binary variable [1,0] or [true,false] could be easily be generated by LLMs based on the presence or absence of certain trends or sentiments ML/AI Regression models will receive such binary variable and consider it as part of one of the demand influencing factors While ML/AI Regression models are more adept at handling numeric data and creating mathematical equations for forecasting, LLMs can complement these models by providing valuable qualitative insights. By combining the strengths of both LLMs and ML/AI Regression models, businesses can create more robust and accurate demand forecasting models, which consider both numerical and contextual factors for making predictions. As LLMs continue to evolve, they may become increasingly valuable in enhancing forecasting accuracy and capturing complex market dynamics.
How are LLMs like GPT used in Supply chain right now?
The best of breed supply chain planning software and our technology partner, Relex, has recently launched RELEX-GPT. The tool helps users and consultants solve queries or find product and functionality information curated in an understandable and applicable way. Given that it could serve as a support tool it will save valuable time and money both for live customers and implementation projects. RELEX Solutions enhances workforce productivity with RELEX-GPT | RELEX Solutions Summary Before finishing I want to thank you for reading our article. In Demandtex we support companies with complex supply chain problems such as improving forecasting. Typically, we see the most benefits and ROI by adopting new technologies into the planning operative models. We help from advice on strategy and initial assessment, to selecting a technology, with main specialty being implementation of new systems that support supply chain best practices.
