AI sales advice raises revenue compared with unavailable support
Lu Fang, Zhe Yuan, Kaifu Zhang, Dante Donati et al. · arXiv · Fang et al. · 14 Oct 2025
Source finding
In a randomized experiment, an AI chatbot increased sales per shopper by 16.3 percent compared with a message that support was unavailable.
Our interpretation
The baseline determines value. This does not establish better performance than human advice.
Method and evidence
44,614 shoppers requesting support, September–October 2023. One of seven randomized field experiments on a cross-border retail platform. Strong evidence for this comparison.
Limitations and interests
One platform, short horizon, weak baseline. Not a general return or labour-productivity rate. Two authors consulted for the company; one was an employee.
What businesses can do
Test AI against your actual service. Measure revenue, quality and costs separately; outperforming unavailable support does not establish an advantage over human advice.
Where to start
Section 2.2, Table 4 and Appendix C1. First posted 14 October 2025; reviewed arXiv version 29 June 2026. Original study