Exploratory Customer Discovery Through Simulation Using ChatGPT and Prompt Engineering

• Main Authors: Ilagan, Joseph Benjamin R, Alabastro, Zachary Matthew, Basallo, Claire Louisse, Ilagan, Jose Ramon
• Format: text
• Published: Archīum Ateneo 2024
• Subjects: Agent-based modeling; Agent-based simulation; ChatGPT; Computer simulation; Customer development; GPT; Large language models; Lean startup; LLM; Simulation; Startup; Artificial Intelligence and Robotics; Business; Computer Sciences; Physical Sciences and Mathematics
• Online Access: https://archium.ateneo.edu/qmit-faculty-pubs/23; https://doi.org/10.1007/978-981-97-4581-4_5
• Institution: Ateneo De Manila University

Entrepreneurship and the tech startup journey are complex, dynamic, risky, and uncertain. Risk-taking needs to consider the complexity and interconnection of different aspects of the entrepreneurial and startup context. A simulation is a model of an existing complex system and experimenting with the model to understand the whole system’s behavior. Computer simulations have been widely used to study complex environments such as entrepreneurship. Large language models (LLMs) such as ChatGPT, by nature of their training and design, are models of humans and likely possess latent social information. As such, LLMs could extend their usefulness from mostly being assistants to simulators of human behavior. Technology startups iteratively manage risks involving the uncertainty of their business models through Lean Startup Approaches (LSAs) combined with customer development. The first step in customer development involves customer discovery, where startup co-founders start with a vision and a set of assumptions (or “hypotheses”) about their business model and seek feedback from their prospective customers. This study explores using ChatGPT as a simulation tool for customer development for technology startups. The validation of the simulator involves coming up with baseline behavior and feedback without prompt preparations, followed by preparing synthetic prospective customers as agents in the virtual environment. This involves endowment of demographic characteristics and getting behavior and feedback again afterward.