Agent Behavior

The Behavior tab decides what happens when your agent reaches the edge of what it can do - when it can't answer, or when a message clearly needs a human. You set where those moments go, what you promise the customer, and a few automatic rules for common situations.

The Behavior tab showing escalation settings, the escalation routing toggles, and behavior rules

The Behavior tab showing escalation settings, the escalation routing toggles, and behavior rules

Escalation Settings

These set the tone and timing of every escalation:

  • Support email - where escalations and notifications go (defaults to your account email).
  • Response window - how soon your team usually replies. The agent tells the customer this when it hands off ("our team will respond within 4 hours").
  • Custom escalation message - the exact words the agent uses when it escalates. Leave it blank for a sensible default that points to your support email and response window.
  • Fallback message - shown only if the AI itself is unavailable (a model error or outage). Blank uses a default that points the customer at your support email.

Escalation Routing

When the agent escalates, you choose what actually happens - and you can turn on more than one:

1

Create a DeskClone ticket

The default. The conversation becomes a ticket in your Tickets list for your team to pick up.

2

Send an email notification

Email an address you specify - handy if your team lives in their inbox.

3

Invoke a custom API tool

Call one of the agent's own custom API tools with the escalation details - for example, open a Zendesk ticket or post to Slack. Pick the tool from the dropdown.

Keep Create a DeskClone ticket on even if you also email or call a tool - it keeps a record you can track and resolve in one place.

Behavior Rules

Rules act on what the customer types, before the agent even reasons about it - useful for sensitive or repetitive cases. Add a rule and pick its type:

  • Escalate to team - if the message contains certain keywords (like "refund" or "cancel"), skip the AI and escalate straight away.
  • Respond with canned text - reply with a fixed message for specific keywords (like business hours or your return policy).
  • Politely redirect - steer clearly off-topic messages back to what your agent is for.

Keep rules few and specific - they are a scalpel for known cases, not a replacement for a good knowledge base.

What's Next

Give your agent the tools to actually resolve those escalations in Tools and Integrations, or fine-tune how it reasons in Advanced Agent Settings.

Was this page helpful?