Randomized test of an AI sales agent on 100 sites (ArXiv) (arxiv.org)

A recent study titled "Selling in the Wild" introduces Javo, an AI sales agent tailored from 4.6 million commercial conversations, and evaluates its performance against both general-purpose models and human sales associates. Conducted across 100 live websites in various industries, the study involved randomizing 14.8 million visitors to assess the impact of Javo on sales performance. Notably, Javo increased revenue per visitor by 10% compared to no agent and showed a significant conversion improvement of 21.6% over frontier-model agents. Importantly, the agent performed comparably to human associates, achieving an engaged conversion rate of 11.26% versus 11.20%, while operating at just 1.8% of the cost per conversation. This research is significant for the AI/ML community as it showcases the practical application and effectiveness of a deep learning-based sales agent in real-world scenarios. Javo's performance indicates that AI can compete with human associates in sales environments, prompting deeper discussions on the deployment of AI in customer-facing roles. The study highlights key implications for the future of sales, suggesting enhanced efficiency and potential cost savings for businesses integrating advanced AI solutions in their sales processes.