The Viability of Large Language Models for Conjoint and MaxDiff Analysis in Market Research

The advent of Large Language Models (LLMs) such as GPT-4 has sparked a significant shift in the landscape of data analytics within the market research industry. These advanced AI tools have the potential to emulate complex human decision-making processes, offering new avenues for understanding consumer behaviour and preferences. Despite early explorations into the capabilities and limitations of LLMs in executing more sophisticated tasks such as Conjoint, the rapidly evolving nature of these models necessitates continuous and comprehensive research to fully comprehend their potential impact.

The potential of LLMs to accurately predict consumer choices and quantify trade-offs presents an opportunity to streamline market research practices, offering insights without the need for exhaustive surveys. However, the emergent nature of Generative AI demands a rigorous examination of its predictive reliability, biases, and limitations in capturing the nuanced aspects of human cognition. 

Ipsos has undertaken one of the largest research exercises in this field, eliciting over 250,000 AI generated responses to evaluate a range of LLMs across a diverse set of scenarios, comparing their performance against real-world data. This research offers a comprehensive insight into the transformative potential of LLMs on choice experiments and the strategic implications for businesses.

For more details and access to the research presentation with the complete set of results please contact: [email protected]

Sources:

• James Brand, Ayelet Israeli, Donald Ngwe, “Using GPT for Market research”, 2023, Harvard Business School Marketing Unit Working Paper No. 23-062
• Mohammad Atari, Mona J. Xue, Peter S. Park, Damián E. Blasi, Joseph Henrich, 2023, “Which Humans?”, Department of Human Evolutionary Biology, Harvard University