A new AI sales assistant called Javo was tested on real‑world websites. The study ran on 100 live sites across e‑commerce, travel, hospitality and high‑consideration services. Visitors were randomly shown either no chatbot, Javo, or a comparable frontier‑model chatbot. Over five months, Javo lifted revenue per visitor by 10% compared with no agent.
How Javo performed
The researchers measured four outcomes. First, on public sales benchmarks Javo beat ten leading general‑purpose models. Second, in the live trial it raised revenue per visitor by +10.0% (95% CI +5.1 to +14.9). Third, among visitors who interacted with the chat, Javo improved conversion by +21.6% over the frontier‑model agent (95% CI +16.9 to +26.3). Fourth, when compared with each site’s human sales staff on 20 sites, Javo’s conversion (11.26%) was statistically non‑inferior to humans (11.20%) while costing only 1.8% of the per‑conversation expense. Safety metrics such as return rates and unsupported claims stayed within preset limits.
Why it matters
For online retailers, a modest‑cost chatbot like Javo can add measurable revenue without needing a large human chat team. The study shows AI can match human sales performance on conversion while keeping operating costs low. It remains unclear how Javo would perform on smaller sites or with different product mixes.
