What are Voice AI Custom Actions?

Voice AI Custom Actions empower AI agents to initiate custom POST webhook calls to external APIs during a live conversation. These actions can incorporate authentication, headers, and dynamic parameters gathered in real-time from the call. This functionality allows agents to instantly retrieve or transmit information based on the caller's input.

Key Benefits of Custom Actions

Voice AI Custom Actions facilitate smooth integrations with external systems directly within a live call, improving the agent's capacity to personalize interactions and resolve issues more quickly.

• Execute real-time API calls prompted by conversational cues.
• Automate data lookups or submissions during the call. For instance, if a customer asks, “What’s the status of my recent order?”, the AI can immediately access your order management system to retrieve the current status, eliminating the need to put the caller on hold.
• Set up POST requests complete with authentication and headers.
• Dynamically transfer call data, such as phone numbers or order IDs.
• Validate webhook responses before implementation.
• Minimize follow-up tasks by addressing needs during the initial call.

How to Set Up Custom Actions in Voice AI

You can configure Voice AI Custom Actions from the Voice AI Labs interface in a few simple steps.

Here’s how to begin:

1. Go to Labs > Voice AI > Custom Actions.
2. Click Create Action to launch the configuration window.
3. Provide a name and description for the action.
4. Enter your Webhook URL and choose POST as the method.
5. Input any necessary authentication details, like a Bearer token.
6. Add any required custom headers.
7. Define parameters that will be dynamically extracted from the conversation.
8. Establish conversation trigger conditions. For example, if your AI assistant needs to fetch a user’s account balance when they say “check my balance,” you can create a trigger for that phrase and set a parameter for their phone number to be passed to your API.
9. Utilize the Test Webhook feature to confirm your setup.
10. Click Save to enable the custom action.

Webhook Integration

Webhook integration forms the foundation of Voice AI Custom Actions, enabling your agents to connect with any external system that supports APIs, such as CRMs, scheduling tools, and databases.

Each Custom Action is built around a POST request, which can contain:

• A webhook endpoint URL.
• Headers, including API keys or tokens.
• A request body containing dynamic parameters.
• Authentication methods like Bearer token or Basic Auth.

Example: When integrating with a CRM system, your webhook URL could target an endpoint like /api/v1/lookupContact and pass parameters such as the contact’s email or phone number.

Conversation Triggers

Triggers specify the conditions that activate a Custom Action during a call. You can set up simple triggers based on phrases or design more sophisticated logic.

Example triggers include:

• When a user says, “I want to check my appointment.”
• When an email address is mentioned.
• When a sequence of digits, like an order number, is spoken.

You can also add conditions to triggers, such as “only execute if parameter X is available.”

Dynamic Parameter Collection

Voice AI can identify and tag important data in real time throughout a conversation. This information is automatically allocated to parameters for use in the webhook request. This feature is particularly useful when the AI must capture multiple details, like an order number and an email address, before initiating a shipping status inquiry. These extracted values can be assigned to your webhook payload, making the integration both context-aware and personalized.

Supported data types include:

• Text (String)
• Number (Numeric)
• Email
• Phone Number
• Date (coming soon)

You can map these values to any part of your webhook payload to tailor the integration.

Real-time Testing

Before finalizing a Custom Action, you can utilize the integrated Test Webhook tool. This function lets you simulate a call, send test data, and review the live response from your external system. For instance, you can use the tool to simulate a customer request to “reschedule an appointment” and verify that the webhook accurately captures and transmits the date and time to your calendar system.

This tool allows you to:

• Review the complete request, including headers and body.
• Examine the raw response, such as 200 OK or 404 Not Found. For example, if the webhook response provides an estimated delivery date, your AI agent can instantly tell the caller, “Your package is scheduled to arrive by Thursday.”
• Detect and correct any setup errors before saving.

Frequently Asked Questions

Q: Can I use GET or other request methods?

A: No, only POST requests are currently supported for Custom Actions.

Q: Where can I find Voice AI Custom Actions?

A: You can access this feature under Labs > Voice AI > Custom Actions once Labs has been enabled in your account.

Q: Is authentication supported for webhooks?

A: Yes, you can use Bearer tokens, Basic Auth, or include keys in the headers.

Q: What happens if my webhook fails during a call?

A: The system will log the failure. You can also define a fallback response if no data is returned or the webhook times out.

Q: Is it possible to trigger multiple webhooks in a single call?

A: Yes, each Custom Action can be activated independently based on its specific conditions.

Q: Do I need a developer to configure these actions?

A: Not necessarily. If you have access to the API documentation for the external system, you can configure these actions with basic technical knowledge.