Customer-facing AI

This is AI that talks to your customers with nobody reading first.

Intercom Fin, Tidio and Chatbase answer questions, resolve tickets and send replies on their own. Every common use here touches personal information, and most run without a person reviewing each answer. We rate most of these uses at the highest consequence level on the site and ask for a monthly look.

Chatbots are not bad at their job. The trouble is that three things arrive together: personal information, a public audience and no pre-check. A wrong answer is already with the customer. A manipulated conversation can steer the bot into saying or doing something you never intended. Volume is the point, which means volume can outrun anyone's ability to review it.

Governing customer-facing AI moves the human from before the answer to around it. Sample conversations on a rhythm and look for wrong, unfair or off-brand answers. Keep a log, so you can trace a complaint to what the bot actually said. Make the hand-off to a person real and easy, and decide which requests the bot must never handle alone. Where the bot can act on an account as well as talk, treat it as an agent: scope its access, know how to stop it, and verify that what it reports doing happened.

Open a tool page to see what each bot may do on its own and what we checked last.