Does your AI support chatbot close the case when a customer agrees to wait?
A customer says, ‘Thanks, I’ll wait for an update.’ The support team still owes them something. Intercom’s current Fin documentation warns that phrases such as ‘I’m willing to wait for an update’ can be read as acceptance of an answer, triggering a confirmed resolution and closing the conversation. It recommends a Guidance rule to escalate these cases to a human for follow-up. This is a documented behavior to check, not a new product announcement. Before a rollout, I would have the support lead test a delayed refund or replacement: give the customer a pending update, then reply politely that they will wait. Does the case remain with someone who will check it, or disappear from the working queue? That is a test I would add, not a result Intercom reports. The person taking over should see what is still pending and when the customer was promised another message. Customers should not have to become angry to keep their problem visible. What keeps an unfinished promise on your team’s list after the chat itself goes quiet?
Comments
I would start with the updates the team promised, not the customers who came back to complain. At each promised date, check whether the customer received the update—even if the answer is still ‘we are waiting.’ A low repeat-contact rate can look good when people have simply given up chasing. Keep overdue promises on the team's list without requiring another customer message to reopen them. That also gives the support lead a useful weekly check: which updates were late, and how many customers had to ask twice? No need to send a satisfaction survey to find out whether a message was sent.
I would check the team’s custom escalation rules, too. Intercom’s table says a clear request for human support escalates directly by default, but Guidance can add friction and turn that into an offer instead. So testing the default is not enough. Try “Please transfer me to a person” on the configured service. I would want that request to enter the human queue without another round of persuasion. That is a boundary I would choose, not a guarantee the documentation gives.