Reviewing Actions

Step-by-step guide to reviewing, editing, approving, rejecting, and snoozing proposed actions from proactive agents.

What This Page Covers

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.

Opening a Proposed Action

From the inbox list:

1. Navigate to Operations → AI Assistant → AI Inbox.
2. Click anywhere on the action row you want to review.
3. The detail panel opens, showing Context & History on the left and the Message Draft on the right.

What you see at a glance:

Element	Where	What it tells you
Priority indicator	Top of the panel	How urgent the action is — Low, Medium, High, or Critical
Status badge	Next to priority	Current state — Pending, Approved, Rejected, Snoozed, Deferred, Awaiting reply, Expired, or Auto-executed
Internal alert badge	Next to status (alerts only)	Marks actions that raise an internal alert instead of contacting the customer
Confidence score	Context panel	The AI’s confidence level (0–100%) with a progress bar
Expiration	Inbox list	For pending items, a countdown badge (less than an hour, hours, or days left) appears under the subject when less than 48 hours remain

Understanding the Context

The left panel gives you everything you need to decide whether the proposed action makes sense.

Customer Information

• Full name and company name (if applicable)
• Customer type — Member (has active contract), Contact (no contract), or Lead
• Quick link to open the full customer profile

Why This Was Proposed

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.”

Confidence Score

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.

Related Activity

When the action is linked to a conversation with the customer, the context panel shows:

• Sent messages — Messages already delivered to the customer as part of this exchange.
• Escalated tickets — Any help desk tickets raised from the same conversation.
• Linked conversation — A shortcut to open the full AI conversation for this exchange.

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.

Review Details

Once an action has been processed, the context panel also shows who reviewed it (Reviewed by) and when (Reviewed on), so your team can trace decisions.

The Message Draft

The right panel shows the message the AI has composed. You can review it as-is or edit it before sending.

The Draft

The draft consists of a Subject line and a Message body. Both are editable directly in the panel while the action is Pending or Snoozed — there is no separate edit mode. Once the action has been processed, the fields are locked for viewing.

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.

Delivery Channel

Below the draft you choose the Delivery channel — how the message reaches the customer:

• AI Email — Standard email to the customer’s address.
• Email Message — Delivered through the member portal. You can optionally pick a connected email account to send from, which enables reply tracking.
• WhatsApp — Requires the customer to have messaged your location within the last 24 hours; otherwise a pre-approved opt-in template is sent first and the action waits for the customer’s reply.
• Voice — An AI voice call. Only sent during business hours; if approved outside those hours, the action is automatically deferred until the next available window.
• Help Desk — Creates a support ticket instead of contacting the customer. You can assign a department, set a priority, and add internal notes visible to your team only.

Warnings you may see:

• Voice not enabled — The Voice AI agent must be enabled in your location’s settings before voice messages can be sent.
• WhatsApp not configured — A WhatsApp sender number must be configured in your location’s settings first.
• Invalid or missing phone number — Voice and WhatsApp require the customer to have a valid mobile number in international format (+XX…). Sending is blocked until it’s fixed.
• Deferred to business hours — Voice messages approved outside business hours are queued automatically.

Taking Action: Three Options

Every pending or snoozed action gives you three buttons at the bottom of the detail panel. Here’s when and how to use each one.

Approve & Send

Use when the message is accurate, the tone is right, and you’re happy with it as-is.

1. Review the customer context and the message draft.
2. (Optional) Add review notes in the text area at the bottom — these are internal notes for your team, not sent to the customer.
3. Click Approve & Send (green button with a check icon).
4. Confirm in the modal that appears.

What happens next:

• The message is sent immediately through the selected channel.
• The action status changes to Approved.
• The customer receives the email, WhatsApp message, or voice call within seconds.
• Any reply from the customer starts a two-way conversation in AI Conversations.
• The action is logged in the customer’s history and tracked in your metrics.

Once approved, the message is sent immediately and cannot be unsent. If you need changes, edit the message before approving.

Snooze

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.

1. Click Snooze (orange button with a clock icon).
2. 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.

3. 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.

Reject

Use when the action isn’t appropriate — the issue is already resolved, the timing is wrong, or you’d prefer to handle it personally.

1. Click Reject (red button with a cross icon).
2. 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.

Review History

Once an action has been reviewed, you can see who reviewed it and when:

Field	Description
Reviewed by	Name of the team member who approved or rejected the action.
Reviewed on	Date and time of the review.
Notes	Any internal review notes added at the time of review.

This review history appears in the Context section of the detail panel and is visible in the Approved and Rejected segments.

Status Reference

Understanding the different statuses helps you know where each action stands:

Status	Meaning	Can Edit?	Can Approve?
Pending	Awaiting your review. This is the default state.	Yes	Yes
Approved	You reviewed and sent the action.	No	N/A
Rejected	You dismissed the action.	No	N/A
Snoozed	Deferred for later — will return to pending after the snooze period.	Yes	Yes
Deferred	Held until a specific time or condition (e.g., business hours).	No (until executed)	No
Expired	The time window to act on this has passed.	No	N/A
Auto-executed	Sent automatically without manual review (agent has auto-execute enabled).	No	N/A

Best Practices for Reviewing

Daily Review Routine

For the best results, check your inbox 2–3 times per day:

1. Morning — Review overnight actions first thing. Prioritize Critical and High priority items.
2. Midday — Catch up on anything that came in since your morning review.
3. End of day — Clear any remaining pending items before logging off.

Prioritizing by Priority Level

Priority	How to handle
Critical	Review immediately — these are time-sensitive (e.g., invoices 15+ days overdue, cancellation risk).
High	Review within a few hours — important but not urgent.
Medium	Review during your regular daily check.
Low	Batch-process when you have spare time.

Building Confidence Over Time

As you use proactive agents more, your approval patterns will help the system learn:

• Consistent approvals for a specific agent → higher confidence scores over time.
• Frequent rejections → the agent may lower its confidence or flag actions for closer review.
• Review notes (when available) → provide feedback that trains future proposals.

AI Inbox

Understand the inbox interface, filtering, and segments.

Configuring Agents

Adjust agent settings based on your review patterns.