Leveraging Large Language Models in Analytics Practice: Opportunities, Challenges and Best Practices

Abstract

Large Language Models (LLMs) such as ChatGPT have become widespread and are positioned to impact many jobs and roles. A notable study by the Harvard Business School in collaboration with Boston Consulting Group found that consultants using an LLM completed 12.5% more tasks, completed these tasks 25% quicker and produced 40% higher quality results than consultants which did not have access to an LLM. Conversely, this study also sheds light on some potential dangers of the usage of LLMs. In our presentation, we will delve into the practical implications of LLMs for analytics practitioners. Following work led by the INFORMS Practice Section Board and using a real-world use case, we will show how an LLM can be used throughout the lifecycle of designing, implementing, and deploying a data driven optimization solution. Key takeaways of this presentation include: • A technical overview of LLMs, their capabilities and usage. • Awareness of how an LLM could be used throughout the lifecycle of creating an analytics application, including precise problem formulation, optimization and data analysis code generation, and user interface and documentation creation. •Guidelines for more effective LLM usage in analytics projects. • Potential pitfalls and dangers of LLM usage for analytic solution creation, including ethical and intellectual property aspects.