Hand-off is not a fallback feature — it is the design that makes automation trustworthy. These are the triggers to write down before you switch anything on.
| Trigger | Why a human must own it |
|---|---|
| Money moves: refund, partial refund, warranty, fee waiver | Every one is a spending decision that belongs to the business owner |
| "I want to speak to a person" | Refusing this is the single most hated automation behaviour |
| Anger, distress, or a threat to escalate publicly or legally | Tone is the thing a restricted assistant cannot judge safely |
| Damaged, missing, wrong or unsafe product | Needs judgement, sometimes a photo, sometimes a recall conversation |
| Personal data requests, deletion, GDPR-style questions | Compliance answers have legal weight |
| Anything outside your written knowledge | The correct answer is "I do not know yet", which is a hand-off |
| Custom or bespoke pricing | A number quoted wrongly becomes the price you owe |
Say who will reply and where, never promise a time you cannot control. "A specialist will contact you within 24 hours" is a commitment; "I have sent this to the shop owner" is not.
Whatever the assistant could not answer is your roadmap: it is either a missing knowledge entry (add it) or a real exception (handle it). Review the unanswered list weekly for a month and the automated share climbs on its own — that is the compounding effect that makes this cheaper than a seat.
Seven days, 300 AI replies, no card. Your support page is live as soon as you write down the answers you already give.
Same day is a reasonable promise for a small shop. The mistake is not promising at all — set expectations in the hand-off message.
Yes. A short line that this is the shop assistant, with an option for a person, prevents the feeling of being tricked, which is what damages brands.
Turn auto-reply off and leave the page as a message box, or keep the assistant on for policy questions only. Do not let hand-offs pile up unanswered.