Once a proactive agent detects a pattern worth acting on, a proposed action appears in your AI Inbox. This page walks you through exactly what to do when one shows up — how to understand the context, edit the message, and choose the right action.
This section explains the trigger that created the action:
Agent name — Which proactive agent generated this (e.g., “Due Invoice Reminders”).
Criteria — A short label for the trigger type.
Reason — A plain-language explanation of what the AI detected. For example: “Invoice #INV-2024-0847 is 5 days overdue. Customer has a strong payment history but missed this payment, which is unusual.”
The confidence score reflects how strongly the detected situation matches the agent’s criteria. It’s shown as a percentage with a color-coded progress bar:
Green (80–100%) — High confidence. The AI is fairly certain this action is appropriate.
Yellow (50–79%) — Moderate confidence. There are some signals, but they may not be definitive.
Red (0–49%) — Low confidence. The AI is uncertain — review carefully before approving.
Click the Customer history tab at the top of the detail panel to see:
Recent activity — Check-ins, bookings, last visit date.
Communication history — Recent AI conversations, emails, and previous proactive actions sent to this customer.
Payment history — Past invoices and payment patterns.
Support tickets — Recent help desk messages or unresolved issues.
This context helps you avoid sending messages that conflict with recent activity — for example, a payment reminder when the customer just submitted a dispute ticket.
When the action status is Pending or Snoozed, the subject and body fields are editable:
Type directly into the subject or message body fields.
Use the rich text toolbar to apply bold, italic, bullet points, or links.
The message body supports HTML formatting for professional-looking emails.
Tip: Keep the core message intact. Variables like {{customer_name}} and {{invoice_number}} are auto-replaced with real data when the action is sent — editing them manually is possible but usually unnecessary.
You can change how the message is delivered by selecting a different channel from the Delivery channel dropdown:
Email — Text-based messages sent to the customer’s email address.
Voice — AI-powered phone call using the customer’s mobile number.
Important warnings for Voice messages:
If the customer doesn’t have a valid phone number, a warning appears and blocks sending.
If business hours enforcement is enabled and it’s outside your location’s business hours, approving a voice message will automatically defer it until the next business hours window.
Use when you want to revisit the action later — perhaps after following up with the customer directly, waiting for more information, or simply because you’re busy right now.
Click Snooze (orange button with a clock icon).
Choose a duration from the menu:
Duration
When to use
1 hour
You’ll review it shortly — after a meeting, for example.
4 hours
You want to finish other tasks first and come back this afternoon.
24 hours
You’d like to review it tomorrow.
3 days
You want a few days before following up.
1 week
You want to revisit it next week.
The action moves to the Snoozed segment and returns to pending after the selected period.
Tip: Snoozing is a great way to batch-process actions when your inbox gets overwhelming. Snooze everything, then review the batch later when you have more time.
Use when the action isn’t appropriate — the issue is already resolved, the timing is wrong, or you’d prefer to handle it personally.
Click Reject (red button with a cross icon).
Confirm in the modal that appears.
What happens next:
The message is not sent to the customer.
The action status changes to Rejected.
The action is removed from the pending queue.
The rejection is logged for analytics — this helps the system improve future proposals.
Rejection reason tracking is planned for a future release. For now, rejections are logged internally to help improve agent criteria over time.
When to adjust an agent: If you’re rejecting many actions from the same agent, consider adjusting its configuration — for example, increasing the cooldown period, raising the confidence threshold, or reducing the max actions per day.
