When a human takes over

Escalation is a feature, not a failure

Panelbot is built to know its limits and to step back when it reaches them. When it hands a conversation to a real person on your team, that is the system working as intended. A tool that presses ahead when it is unsure is far more dangerous than one that stops and asks. So Panelbot stops and asks.

This page explains exactly when that happens and what it looks like from every side.

The four reasons Panelbot hands off

There are four situations that cause Panelbot to bring in a human. They cover the cases where continuing on its own would be a bad idea.

1. The assistant is not confident. If Panelbot cannot tell with confidence what the problem is, or the customer’s situation does not clearly match something it understands, it does not guess. It escalates rather than risk acting on a wrong reading of the issue.

2. The change is risky and needs approval. Some fixes are straightforward and reversible; others carry real weight. When the right next step is a higher-impact change, Panelbot does not make that change on its own. It prepares the fix and requests a human’s approval first. Nothing impactful happens until one of your people says yes.

3. The issue is out of scope. Panelbot targets common, well understood hosting problems. When something falls outside that, an unusual failure, a request that needs judgment only your team should make, or a problem that is simply beyond what the assistant handles, it hands the conversation to a person instead of forcing a solution.

4. The customer asks for a person. If someone simply wants to talk to a human, that is reason enough. Panelbot does not argue or loop them back through the same questions; it brings your team in.

Alongside these, a conversation that runs unusually long or gets stuck is treated the same way: rather than spinning in circles, Panelbot escalates so a person can pick up where it left off.

What the handoff looks like

Escalation takes one of two shapes depending on why it happened.

An approval request. When Panelbot has diagnosed a problem and has a fix ready that needs sign-off, it creates an approval request for your team. Your staff see what the assistant found and what it proposes to do, and they approve or decline. Meanwhile the customer is told that their request is being reviewed, so they are never left wondering. If the change is approved, it proceeds; if not, it does not.

A ticket and live handoff. When the situation needs a person to take the conversation, Panelbot hands it to your support team through your normal channels, as a ticket or a live chat takeover. A member of your team picks it up with the context Panelbot already gathered, so they are not starting from a blank page. The assistant steps back and the human takes the lead.

You decide where these escalations go by setting an escalation contact for your account, so handoffs reach the right people on your team.

The customer is always told

From the customer’s side, an escalation should feel like being taken seriously, not being dropped. When Panelbot escalates, the customer is told plainly that a human on the team has been notified and will take over. They are never left staring at a silent chat or a raw error message. If something goes wrong on the technical side, the customer sees a simple, reassuring note that a technician has been alerted, never a confusing stack trace.

The goal is that a handoff reads as “a real person is now on this,” which for many customers is exactly the outcome they wanted.

Nothing risky happens unattended

This is the promise that ties it together: Panelbot never pushes through a risky change on its own initiative. Higher-impact operations wait for explicit human approval, by default a person on your team, and that requirement is not something the assistant can set aside or work around. If you opt in, a narrow, clearly scoped class of actions can instead be completed in chat once the customer passes a strong confirmation, and the most dangerous operations stay off limits no matter what. When confidence is low or the stakes are high, the default is always to involve a person.

That is what lets you point Panelbot at real customers on real servers. It handles the routine work at speed, and it knows the exact moment to bring in your team, so the hard calls and the risky changes stay firmly in human hands.

Where to go next

• Safety and permissions explains how changes are graded by impact and gated.
• What Panelbot can and cannot do sets the honest boundaries of its reach.
• How Panelbot works end to end shows where escalation fits in the overall flow.