Seeing the Unseen: AI-Driven Discrepancy Detection to Foster Improved Decision-Making and Supply Chain Resilience - A Case Study of PipeChain SCM

Sammanfattning

Background Supply Chain Management is becoming increasingly more important to organizations in the post Covid-19 era. Managing Supply Chains is also moving towards finding data-oriented ways of working, and companies are making substantial investments in getting digital information systems to integrate with their ways of working. Simultaneously, Artificial Intelligence has become a well-discussed topic over the past few years, with the introduction of large language models and chatbots to the public. There is a massive interest across many industries to find ways in which generative AI can be used to create value, and supply chain companies are no different. Problem Definition and Purpose When managing their supply chains, actors often have access to a number of different systems to monitor operations, e.g. ERP-, customer relationship management- or supplier relationship management-systems. These systems displays data to support system users in their decision making. However, it takes time and effort to reliably be able to learn and navigate from different systems. According to research, one of AI's main strengths is its ability to perform analysis on big data sets. This thesis aims to explore if generative AI can be used to interpret, analyze and present its found insights in supply chain processes to a system user, and thereby making it easier to gather insights from available data. Method The thesis deploys a case study, using an abductive research approach. Initially a literature review and a mapping of a case-specific process to establish a foundation of what generative AI will have to interpret and present. Thereafter, current frameworks for AI-prompting are looked into, and then turned into a new one fitting for the case. Finally, the framework is applied to the case, and generated output is evaluated to identify strengths and weaknesses of generative AI's analysis. Conclusions The study finds that generative AI can be able to interpret, structure and visualize data in a reliable way in a supply chain setting. The quality and procedure of the analysis is highly dependent on which generative AI model is used. Between the evaluated models, Claude was the best performing, closely followed by ChatGPT, while Gemini and Copilot showed weaker results. Furthermore, all models achieved worse overall output in a test environment where two metrics was to be analyzed compared to only looking at one.