Most support volume is the same handful of questions. This answers those from your own approved sources, handles a short list of safe actions, and sends everything else to a person with the whole story already written up.
Anyone whose inbox fills with order status, policy and account questions — ecommerce, membership organisations, service businesses with a steady stream of the same five questions.
From your site, your inbox or a chat widget. It reads what was actually asked rather than matching a keyword to a canned reply.
The actual order, account or booking — not a guess. An answer about a specific order has to come from that order.
Your policy, your terms, your documented process. It does not invent an answer when it cannot find one; it says it does not know and hands over.
Reading is one thing. Changing an address or a booking is another, and nothing changes until the person is confirmed to be who they say they are.
Address changes, resends, appointment moves — the small things that make up most of the queue. The list is deliberately short and you decide what is on it.
Refunds, cancellations, complaints and anything that smells unusual go to a person — with a summary, the sources it used, and what it already did. Your support agent never starts from a blank screen.
We agree on these two before we build, and report them after.
It answers from sources you approved and nothing else. Where it cannot find an answer in those sources it escalates instead of filling the gap. That constraint is the difference between a support system and a chatbot.
Two numbers, agreed before we build: how many tickets close without a person, and how many of those get reopened. A high resolution rate means nothing if customers come back because the first answer was wrong.
Yes. It says so. Hiding it is both dishonest and, once a real problem appears, actively annoying for the customer.
That is the normal case. It drafts and acts inside the tool your team already uses rather than becoming a second place they have to check.
You see it, because the escalation carries the full trail of what it read and did. We tighten the sources or narrow the action list, and the case becomes a test we keep.
No. The first build runs on sample records. Live systems get connected once you have watched it work and agreed what it may do unattended.
One call, no charge. We’ll tell you whether it’s worth automating — including when it isn’t.