Harnessing ChatGPT for business model validation via AI-simulated interviews
Potekhin, Andrei (2024)
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2024060320168
https://urn.fi/URN:NBN:fi:amk-2024060320168
Tiivistelmä
This thesis examines the potential of simulated customer interviews conducted with ChatGPT, a generative artificial intelligence (AI) model, in startup business model validation. The objective of the project is to explore the use of such interviews and the extent to which simulated responses align with those from real interviews. This study includes a literature review covering topics such as business model validation, Lean Startup methods, Lean Canvas, and the role of AI in startups. Concepts such as persona and avatar are examined, along with their applicability in AI-simulated interviews.
A comparative case study design was employed to contrast three real interviews and three respective AI-simulated interviews conducted with ChatGPT. Each case study focuses on obtaining preliminary validation for a novel web application idea for designers. The results demonstrate a notable alignment in responses between the real and the AI-simulated interviews, particularly in terms of the facts specified in the model instructions. However, some discrepancies and an inability of the approach to produce accurate predictions were also observed. The findings indicate that interviewing AI-simulated customers can indeed aid in business model validation. However, a more effective simulation design is needed. This thesis enhances the understanding of the role of AI in business model validation and suggests areas for future research in this field.
A comparative case study design was employed to contrast three real interviews and three respective AI-simulated interviews conducted with ChatGPT. Each case study focuses on obtaining preliminary validation for a novel web application idea for designers. The results demonstrate a notable alignment in responses between the real and the AI-simulated interviews, particularly in terms of the facts specified in the model instructions. However, some discrepancies and an inability of the approach to produce accurate predictions were also observed. The findings indicate that interviewing AI-simulated customers can indeed aid in business model validation. However, a more effective simulation design is needed. This thesis enhances the understanding of the role of AI in business model validation and suggests areas for future research in this field.