From global models to local solutions: The impact of generative AI on entrepreneurial search in underrepresented markets

Generative artificial intelligence (genAI) has the potential to revolutionize entrepreneurial idea generation by producing high-quality ideas more quickly and cheaply than humans (Meincke et al., 2024; Girotra et al., 2023, Dell’Acqua et al. 2023, Csaszar et al., 2024). This potential may be particularly impactful in lower-income regions, which have some of the highest proportions of entrepreneurs (Gindling and Newhouse, 2014). However, these regions are underrepresented in the training data that powers the most widely used genAI models — models predominantly developed in the United States, reflecting training text from wealthy Anglophone countries (Tao et al., 2023). This data imbalance raises the possibility that the benefits of genAI — the rapid production of cheap, high-quality ideas — may not be realized if the ideas are infeasible or inappropriate for different cultural and economic realities (Cornwall 2024, Birch 2023, Han 2023). Yet, the use of generative AI in emerging market contexts is understudied and has produced mixed results to date (Otis et al., 2024). We explore this tension in a randomized experiment with aspiring entrepreneurs in Zimbabwe. With a GDP per capita 45 times smaller than that of the United States, Zimbabwe is a nation of entrepreneurs by necessity (Mazikana, 2023), and its business context is vastly underrepresented in the training data of foundational models compared to entrepreneurs in wealthy Anglophone countries. We conduct an entrepreneurial workshop with 750 participants, where entrants are randomized to search and refine their entrepreneurial business idea with either a peer or a generative AI tool. To evaluate whether investing in local data may help overcome potential limitations in the model’s knowledge base, we test two versions of the genAI tool: one standard “global” model and another fine-tuned to be more localized by providing additional codified knowledge about the local environment and entrepreneurial opportunities. We document how the ideas received from the AI tools shift the entrepreneurial search process relative to the baseline: do they find the ideas feasible? High-quality? Do they change which business idea they ultimately decide to pursue? We also follow up with entrepreneurs three months after the completion of the program to examine outcomes in practice: what percentage of entrepreneurs initiated business start-up, and whether there were any differences in business performance.