Docs › AI sales agent
Review AI actions (Approvals)
Review AI-proposed actions before they are carried out, and approve, edit, or reject sensitive actions.
Review AI Actions (Approvals)
Approvals lets you review actions proposed by your AI agent before they are carried out. You can inspect the proposed action and conversation context, then approve, edit, or reject it so important customer-facing actions stay under your control.
What is Approvals?
Approvals is the review step for AI actions that may need human oversight. It gives your team an opportunity to check what the AI intends to do before the action is completed.
Approvals can be used for actions such as customer replies, quotations, payment links, and lead-stage changes.
Why use Approvals?
Use Approvals to:
- Review sensitive AI-generated actions before they are sent or applied.
- Prevent incorrect discounts, quotations, or payment requests.
- Keep important lead-stage changes under human control.
- Identify areas where the AI needs better knowledge or guidance.
- Improve confidence in AI-assisted workflows.
Real-Life Use Case
The situation
For example, imagine UrbanNest Realty uses an AI agent to communicate with potential customers and manage sales conversations.
The problem
A customer asks whether a discount is available. The AI may propose a reply offering a discount, but the business may require a team member to review discounts before they are communicated.
What happens without the feature?
The AI could complete the action immediately, even when the action requires additional review or business approval.
How Vaanexa helps
With Approvals, the proposed action can be reviewed before it is carried out. A team member can approve the action, make changes before approving it, or reject it.
The new workflow
- The AI proposes an action.
- The action appears in Approvals.
- A team member reviews the proposed action and conversation context.
- The team member chooses Approve, Edit and approve, or Reject.
- Rejected actions remain logged for review and improvement.
Case Study
Business
UrbanNest Realty
Challenge
The business uses an AI agent to handle customer conversations, but some actions require human review before they are completed.
Objective
Give the team control over sensitive AI actions while allowing the AI agent to continue handling routine conversations.
Vaanexa Solution
UrbanNest uses Approvals to review proposed replies and actions. Team members can inspect the conversation context and decide whether to approve, edit, or reject each action.
Outcome
The team can maintain human oversight over important AI actions while using rejected items to identify where the agent may need additional knowledge.
How Approvals works
When an AI agent proposes an action that requires approval, the action is added to Approvals.
The reviewer can:
- Open the proposed action.
- Review what the AI wants to do.
- Read the related conversation context.
- Choose whether to approve, edit and approve, or reject the action.
- Review rejected items later to identify areas where the agent needs better knowledge.
Before you begin
Before reviewing AI actions:
- Make sure your AI agent is configured and active.
- Make sure the relevant approval requirements are configured for your workflow.
- Make sure your team members know which AI actions require human review.
- Review the agent's knowledge regularly so repeated incorrect proposals can be reduced.
How to use Approvals
- Open Approvals.
- Select an AI action that requires review.
- Read the proposed action.
- Review the conversation context to understand why the AI proposed it.
- Choose one of the available actions:
- Approve to allow the proposed action.
- Edit and approve to make changes before allowing it.
- Reject to prevent the proposed action.
- Review rejected items later to identify areas where the AI agent needs additional knowledge.
Understanding Approvals
| Approval type | Example | What to review |
|---|---|---|
| Customer reply | A reply about a discount | Accuracy, wording, and whether the offer is appropriate |
| Quotation or payment link | Sending a quotation or payment link | Customer context and correctness of the information |
| Lead stage change | Moving a lead to won or lost | Whether the conversation supports the proposed stage |
What can require approval
Approvals can be required for:
- Replies on specific topics, such as discounts.
- Sending quotations or payment links.
- Moving a lead to won or lost.
Rejected actions
Rejected items are logged so your team can review them later. This can help identify situations where the AI agent needs additional knowledge or clearer guidance.
Best practices
- Review the conversation context before approving an action.
- Use Edit and approve when the proposed action is useful but needs a correction.
- Reject incorrect actions instead of approving them just to clear the queue.
- Review rejected items regularly to identify knowledge gaps.
- Keep AI knowledge updated when policies, pricing, or business rules change.
- Apply human review to sensitive actions where an incorrect response could affect a customer or sales process.
Common scenarios
Reviewing a discount reply
A customer asks for a discount and the AI proposes a response. Review the conversation before approving the reply.
Reviewing a quotation
The AI proposes sending a quotation to a customer. Check the proposed action and context before allowing it.
Reviewing a payment link
A payment link requires approval before it is sent. Verify that the action is appropriate for the customer and conversation.
Reviewing a lead-stage change
The AI proposes moving a lead to won or lost. Review the conversation before approving the change.
Common mistakes
Approving without reading the conversation
The proposed action may look correct on its own but may not match the customer's actual request.
Rejecting without addressing the knowledge gap
If the AI repeatedly proposes incorrect actions, simply rejecting them does not solve the underlying problem. Review the relevant knowledge and update it when necessary.
Editing the action without checking the context
Make sure any changes you make still match the customer's conversation and business rules.
Ignoring rejected actions
Rejected actions can reveal recurring areas where the AI needs better information. Regularly reviewing them helps improve the agent's future responses.
Frequently asked questions
What is Approvals?
Approvals is where you review AI actions that require human approval before they are carried out.
What actions can require approval?
Examples include replies about specific topics such as discounts, sending quotations or payment links, and moving leads to won or lost.
Can I change an AI action before approving it?
Yes. Use Edit and approve to modify the proposed action before approving it.
What happens when I reject an action?
The action is rejected and logged so you can review it later and identify areas where the AI agent may need more knowledge.
Should every AI action require approval?
Not necessarily. Use approvals for actions where human oversight is important, while allowing routine actions to proceed according to your configured workflow.
How can rejected actions help improve the AI agent?
Rejected actions can show where the agent is missing information or needs better knowledge. Reviewing these patterns helps you identify what should be added or updated.