This is a thing I keep noticing in real customer calls that doesn’t really show up in voice AI demos.
People interrupt constantly.
They start answering before the question is finished, correct themselves halfway through a sentence, say 'wait actually…' and completely change what they were asking about.
That’s normal when two people are talking but it seems like a pretty difficult problem for an AI voice agent because it has to know whether the customer is adding context, correcting something or trying to stop the current response entirely.
We’re looking at enterprise voice AI for longer customer service conversations and I’m beginning to wonder if turn taking is as important as natural voice.
For anyone testing conversational AI over the phone, how are you testing interruptions? Is this still something customers notice pretty quickly?
[link] [comments]