Loading
Because it is confident and unsourced. It answers from a general model, invents a refund policy that sounds plausible, and the customer screenshots it.
The damage is not the wrong answer. It is that nobody can tell which of the previous thousand answers were also wrong.
The customer-facing agent answers only from your knowledge base and the customer's own record, and cites the article behind each answer. Where it has no source, it says so and hands to a person with the conversation intact.
The agent-assist side drafts replies for a human to send, and never sends on its own.
The customer can check. So can you: every AI answer links to the article it used, so a wrong answer is a wrong article, which is fixable once for everybody.
Handover stops being a restart. The person picks up with the history, the diagnosis and what has already been tried.
They will not make commitments. Refunds, credits, dates and exceptions are outside what the agent may state, and it hands those to a person by rule rather than by judgement.
They also will not pretend to be human. Customers are told what they are talking to.
Nothing here is an integration. Each line is two products reading the same record from different sides.
The customer-facing agent answers only from your knowledge base and the customer's own record, and cites the article behind each answer. Where it has no source, it says so and hands to a person with the conversation intact.
The customer can check. So can you: every AI answer links to the article it used, so a wrong answer is a wrong article, which is fixable once for everybody.
They will not make commitments. Refunds, credits, dates and exceptions are outside what the agent may state, and it hands those to a person by rule rather than by judgement.
Fourteen days, every module, no card. Or half an hour with someone who will run it on your own records and tell you where it does not help.