What if the AI gives a customer wrong information?
This is why scope and testing matter. Assistants are limited to what they actually know, escalate anything outside it, and are reviewed monthly against real conversation logs. We also keep a manual fallback in place during the first weeks so nothing depends solely on the automation.