An agent that never admits uncertainty is a liability. The moment a customer realizes it is bluffing, every answer it ever gave is suspect. Airhop is built so the agent knows the edge of what it knows and hands off cleanly when it gets there, rather than guessing to keep the conversation going.
Escalation on low confidence
Because every answer is grounded in your docs and tickets, the agent can tell when a question falls outside what it can support. When confidence is low, when the docs do not cover it, or when the situation calls for human judgment, it does not improvise. It escalates. That restraint is what makes it safe to let the agent answer at all.
A human, with full context
Handover is only useful if the person picking it up does not have to start from scratch. Airhop routes the thread to a human in email or Slack with the whole history attached: the customer's question, what the agent already said, and the account context behind it. No one has to dig through a separate tool to reconstruct what happened. The person reads the story, not a fragment.
Reply once, and it relays
The handoff stays in the tools your team already lives in. You reply once, in email or Slack, and that reply relays straight back to the customer. The agent stays the connective tissue so the customer experiences one continuous conversation, not a jarring transfer to a different channel and a different system.
Every handover makes the agent better
A handover is not just a rescue, it is a lesson. The agent learns from how you answered, so a question that needed a person this time is more likely to be handled confidently next time. Over weeks, the share of threads that require escalation falls as the agent absorbs your team's judgment. Handover is the mechanism by which the agent grows into your knowledge rather than staying frozen at launch.
The same principle, everywhere
Handover on answers and the approval layer on outreach are the same idea in two places: the agent acts when it is sure and defers to a human when it matters. That consistent posture, autonomy bounded by judgment, is what makes Airhop a responsible take on customer success software rather than a black box that hopes for the best.
Common questions
When does Airhop hand a thread to a human?
When its confidence is low, when a question falls outside what your docs can ground, or when a situation needs judgment. Rather than guess, the agent escalates to a person.
What does the human receive?
The whole thread history and account context, in email or Slack, so the person picking it up does not have to reconstruct what happened. They reply once and it relays straight back to the customer.
Does the agent learn from handovers?
Yes. It learns from how you answered, so a question that needed a human once is more likely to be handled confidently next time.
Where does handover happen?
In the tools your team already uses: email and Slack. There is no separate console to monitor for escalations.