Imagine my flight is delayed and I need to move a tour. I tell the chatbot what happened, give it the booking reference, and ask about a new date. It sends me to a person. The first thing they ask is, “Can you tell me what happened?” I’m back where I started.
I liked the way design practitioner Minji Gwak described this problem on LinkedIn. The handoff can succeed technically while the customer still has to do the work again. That’s the moment I’d design around.
Give the next person enough to pick up the conversation
The bot can explain the usual policy and check what it’s allowed to check. If the change needs someone’s approval, I’d offer a handoff with a short summary: the booking reference, which date the customer asked for, and what’s already been checked. Let the customer correct it before sending. Give the support agent the information they need for this request, and tell the customer what to expect next.
Microsoft’s Human-AI Interaction guidelines and Google’s People + AI Guidebook both give teams ways to think about setting expectations, letting people correct a mistake, and helping them recover when the system gets stuck.
When I test this, I’d go all the way through to the person receiving the request. Can they pick it up without asking for the same details? Can the traveller see that someone has taken over? Those are the things I’d want to improve before celebrating how many chats the bot kept to itself.
