This presentation investigates whether large language models (LLMs) can replicate human behaviour in answering both Conjoint and MaxDiff choice tasks. The research comprising of more than 50 experiments, reviewed different LLM’s, parameter settings such as temperature, and whether different executions of the prompts would yield better results. More than 250,000 LLM choices were generated to see how close LLM’s could get to replicating the utility structure of real-world Conjoint and MaxDiff projects and the presentation will answer eight key hypotheses that were tested. The big question is whether machines can replace human respondents or is there still a role for human respondents and analysts!
