Grounded answers, or no answer
Every reply the system writes is built from sources you have approved and from live data — the order status, the policy, the help article — and each answer shows where it came from. If the answer is not in the sources, the AI does not improvise; it drafts a reply for a person and flags the knowledge gap.
Rolling out safely
Week one is observation: the system classifies and drafts, and your team sends. The weekly quality report shows accuracy per category. Auto-send is enabled one category at a time — order status first, returns policy next — and anything involving money or a frustrated customer stays with people.
Beyond email
The same engine answers website chat, WhatsApp and social messages from a single queue, so a shopper gets the same correct answer no matter where they ask.