This is the part most AI customer success tools get wrong, and it is the part Airhop is built around. Letting an agent reach out to your customers is powerful and dangerous in equal measure. Done well, every at-risk account hears from you at the right moment. Done badly, an agent fires a tone-deaf message at the wrong account and you spend a week apologizing. The approval layer is how Airhop keeps the power without the risk.
What proactive outreach means here
Proactive means the outreach is triggered by a signal, not by a customer writing in first. The triggers come from the revenue-aware health read: a churn risk forming, a failed payment, an overdue invoice, a renewal approaching, or an onboarding that has stalled. When one fires, the agent does not blast a template. It drafts a message grounded in that specific account's context: who they are, what they use, what just happened, and what would actually help.
The approval layer, step by step
The default posture is that nothing reaches a customer without a human on the high-stakes moments. The flow is deliberate:
- The agent drafts. It writes the outreach grounded in the account, and shows you the message, the account, and the reason it was triggered.
- You decide. Edit a line, approve and send, or hold. You are reviewing a finished draft, not staring at a blank compose window.
- It learns. Your edits teach the agent the voice and judgment you want, so the next draft is closer to ready.
- You flip to autopilot. Once a play has earned your trust, you let the agent run it unattended, for that play, on its own.
Autopilot, but inside guardrails
Autopilot is not a blank check. Even when a play runs unattended, it stays inside guardrails you set: cooldowns so an account is not contacted again too soon, quiet hours so nothing sends overnight, and a frequency cap so the agent can never over-contact an account no matter how many signals fire. Autonomy is earned per play and bounded by rules, which is what makes it safe to turn on at all.
Why the approval layer is the whole point
It would be easy to ship an agent that emails everyone the moment a number moves. It would also be a mistake. Trust in an AI customer success agent is built one approved message at a time. By starting with drafts and earning autonomy gradually, Airhop lets you adopt automation at the pace of your own confidence rather than betting your customer relationships on day one. This is also why the agent hands off when it is unsure rather than pushing ahead, see human handover. The result is customer success automation you can actually let run, because you decided where to draw the line.
Common questions
Does Airhop send outreach to my customers automatically?
Not until you let it. By default the agent drafts the message and waits for your approval on high-stakes moments. You can edit, approve, or hold. Once you trust a specific play, you flip it to autopilot and it runs within guardrails.
What guardrails control autopilot?
Cooldowns so an account is not contacted again too soon, quiet hours so nothing sends overnight, and a frequency cap so the agent cannot over-contact an account. You set them, and they apply even on autopilot.
What triggers an outreach draft?
Signals from the health read: a churn risk, a failed payment, an overdue invoice, an upcoming renewal, or a stalled onboarding. The agent grounds the draft in that account's specific context.
Can I tune what the agent drafts?
Yes. You edit drafts before they send, and the agent learns from your edits. Approving and editing are how you teach it the voice and judgment you want before you ever turn on autopilot.