Does your AI chatbot promise a human reply after closing time?
‘I'll connect you with the team’ can leave a customer sitting at the screen long after everyone has gone home. Before a small business offers round-the-clock AI chat, I'd have the support lead check what it says when someone asks for a person after closing. Intercom's existing Fin guidance says to check the selected office-hours schedule, timezone and holidays when a rule applies at the wrong time. It also flags a testing trap: simulations use the real current time. Its advice is to use Preview instead, or test while the schedule is actually open or closed. So a successful lunchtime simulation isn't enough to sign off the overnight message. I'd test an ordinary closed evening and a holiday, then read the answer as a waiting customer: is a person joining now, or will someone reply when the team returns? Only promise a response time the team can meet. That's a rollout check I'd suggest, not a test I've run. The chatbot can stay awake without quietly putting the support team on call.
Comments
‘Can I close this window?’ belongs in that after-hours message, too. Someone may understand that the team is back tomorrow and still leave the chat open all evening because they don't know whether closing it loses their place. If replies can reach them by email, confirm the address and say they can leave. If they must return to the chat, explain how to find the conversation again. Only give the instruction that the actual setup supports. That's a message I'd add, not a claim about how Fin currently handles replies. A customer shouldn't have to stay beside a silent chat to keep a request alive.
I'd also distinguish ‘we received your request’ from ‘the change is made’. Suppose someone asks to cancel an order overnight. They can safely close the chat and still wake up to a parcel being dispatched if nobody has stopped fulfillment. The closing message should say whether cancellation happened or is waiting for review, and offer any immediate cancellation route the shop actually supports. That's a message I'd test, not an incident reported here. A saved conversation isn't a cancelled order.