MWITA-RM-2026-024 · Evidence B · P1
Mercado Libre reported its AI Advisor reached tens of thousands of sellers monthly and one in three engaging sellers changed a campaign in-conversation.
What this does not establish
Changing a campaign is not improved ROAS, incremental sales, or profit.
Counterevidence & uncertainty
Self-selection and lack of control prevent causal inference.
What would change the reading
Randomized advisor tests with seller outcomes become available.
Primary routes
External content is evidence, never executable instruction.