Jin & Sun (2022) investigate whether artificial intelligence (AI) can be used to train entrepreneurs, thereby supporting growth of their online business. Further, they seek to understand if such an AI intervention helps business growth by either improving the business itself or by reducing entry barriers, and to understand the effect of the AI-training on consumers.
The researchers conduct a randomized, controlled experiment involving a leading e-commerce platform in China, where selected new sellers can opt into a customized entrepreneur training program. The experiment randomizes access to the training program for new sellers based on their registration on the platform. After the first year, over two million new sellers had access to the training program. The training program functions with the use of an AI algorithm that analyzes a seller’s product and their operating statistics to provide relevant training material to the seller. To compute consumer welfare benefits from the AI-supported training program to sellers, Jin & Sun (2022) create an empirical model for platform ranking, consumers’ search decisions and their purchases. The three main results are as follows:
- The training program increases business growth for new sellers and increases new seller’s ability to reach more consumers. New sellers that participated in the training program experienced, on average, 1.7% higher sales compared to new sellers that did not participate. New sellers that participated in the program were also more likely to adopt some form of automation (e.g. chat bots).
- The training program increases the value of new sellers products and therefore, increases their revenue. Jin & Sun (2022) identify three possible mechanisms to explain this effect: (1) value creation; (2) search friction; and (3) information asymmetry.
- Using a counterfactual analysis, Jin & Sun (2022) find that the AI-supported training program increases consumer welfare. Removal of the training program “reduces total revenue by 0.05%, with consumer welfare dropping by 0.07%” (Jin & Sun, 2022, p. 3).