Two features that make the assistant feel less like a machine: it remembers people who come back, and it notices when a conversation is going badly.
Returning-visitor memory
Pro feature.
When someone who has chatted before comes back, the assistant gets their history as context: what they asked about, how it was resolved, how they felt about it.
In practice that means it can open with warmth, skip questions it already knows the answer to, and refer to a past issue naturally — "Did the replacement part sort out the noise you mentioned last week?" — instead of starting from zero every time.
There's nothing to configure. It builds on the conversation insights that are tagged when chats end, so it needs a bit of history before it has anything to remember.
Real-time frustration detection
Paid feature — Basic and Pro.
The assistant watches each conversation for signs it isn't going well: negative phrasing, dismissive one-word replies, too many back-and-forths without a resolution.
When frustration crosses a threshold, two things happen:
- The visitor is offered a way out — a "Talk to a person?" chip appears in the widget, before they give up and close the tab.
- You see it — the chat is badged Frustrated · needs human in your AI-handled list, so an operator can prioritise a takeover.
What should I do when I see the frustration badge?
Take over. That's the whole point of the signal — it's telling you there's a human-sized problem in a chat the AI can't close. A takeover at that moment is worth far more than a perfect AI answer two messages later.
If the same topic keeps producing frustration, it's a knowledge problem: check Analytics → Answer gaps and write the missing article or note. See Answer gaps and AI corrections.