AI Receptionists (Admin Center Voice Features)

AI Receptionists Usage and Advantages
- Beta Phase Note
Pricing
Creating New Phone Lines for AI Receptionists
Access and Existing AI Receptionists
Adding an AI Receptionist
AI Receptionists Settings
AI Receptionists Integrations

AI agents can be unpredictable and can make mistakes. Telebroad will not be responsible in any way for the behaviour of AI agents or related consequences.

AI Receptionists Usage and Advantages

AI Receptionists are virtual agents (often referred to as such) that can automatically answer and handle multiple incoming phone calls. They offer round-the-clock service 365 days a year on nights, weekends, and holidays.

AI Receptionists can almost eliminate the issue of missed calls by drastically improving calling responsiveness with consistent high-quality assistance, and also reducing the frustration of busy signals or long waits.

They can support large companies that need to address high call volume overflow, or smaller businesses that just can't always answer the phone or want better after-hours call handling.

Human-Like Conversation, Performing Actions, and Escalation Options

AI Receptionists can understand speech and respond in a human-like voice, conversing with callers in a natural professional manner. They can answer common questions about your business, and with the integration feature, can even access external software and servers to fetch data and perform parallel actions during a live call.

When they cannot provide certain details, they can offer to transfer calls to a human user, a relevant department, or a fallback extension. They can also offer a follow-up response by sending email notifications to staff.

Beta Phase Note

The AI Receptionists feature is currently in early beta phase as Telebroad continues to improve and refine it. While it is fully functional, you should thoroughly test AI agents before deploying them to live customers and continue to monitor calls handled by AI agents initially.

Pricing

The AI Receptionist feature is not included in your standard plan. Pricing is usage-based at 20 cents per minute. You only pay for the minutes your AI receptionist is actively on calls. You can set a maximum call duration limit to control costs.

There are no additional setup fees to configure a basic AI receptionist.

You can contact Telebroad sales for volume pricing or bundle plan options.

Creating New Phone Lines for AI Receptionists

AI Receptionists can be considered autonomous agents in the sense that they require their own phone lines to operate.

Technically, you can assign them to any existing user's line, but that will prevent the user from receiving calls, even if the AI agent is disabled, as calls will be routed to the fallback destination.

Therefore, you should create a new phone line for each new AI receptionist and route calls to this new line.

Since phone lines are attached to users, to keep things separate, you can optionally even go a step further and create a new user for the AI receptionist.

Adding new lines is explained here. Adding new users is explained here.

Access and Existing AI Receptionists

The AI Receptionists feature is accessed from the side navigation menu under Voice Features>AI Receptionists.

The page lists existing AI receptionists by Name, Description, Line (a dedicated phone line for the AI receptionist as explained above), and Status (Enabled or Disabled) columns. Clicking on the Starred column will add an AI receptionist to the home page Starred Features panel for fast access.

Check existing AI receptionists to delete them with the button, but be aware that this cannot be undone.

Adding an AI Receptionist

To create a new agent, click the ADD AI Receptionists button and continue to define the settings and prompts discussed in the following sections. You can also add a new agent from the Voice Features Quick menu.

Clicking Next or Previous will move you between the three settings tabs. Click Create to finish configuring the agent. You will see a pricing and disclaimer screen. Click Confirm and Enable to activate the agent.

AI Receptionists Settings

Select or create an AI agent to show its settings with four tabs – Agent Settings, Call Transfers, Department Emails, and Integrations.

Only the Agent Settings are mandatory. The agent can be activated without the other two tabs; however, they are important to allow the AI agent to deal with situations outside its scope of knowledge or simply to transfer calls to a human when the caller requests it.

Agent Settings

Internal Name / Internal Description: Self-explanatory. The description is optional.
Agent Voice: When you open the menu, you can click to hear how the agent sounds before selecting it.
Phone Line: Select a phone line for the AI agent. As explained above, you should create a new phone line for the AI agent.
Max Call Duration: Since AI agents' calls are billed per minute, you can limit the duration of calls in 5-minute increments to avoid runaway calls. When the limit is reached (timed out), the call is routed to the fallback destination extension. You can select No limit if you don't want calls to time out.
Monthly budget: This lets you control your expenses. Calls will be routed to the fallback destination when the budget limit is reached.
Fallback Destination: This is an importation option for ensuring callers are not stranded. It will route calls to the extension selected here if the AI agent is disabled, the AI system is temporarily unavailable, the monthly budget is reached, or for any other technical reason. It will also route the call if it has reached the maximum call duration and timed out.

You can assign the extension to any Admin Center destination or feature. Extensions are covered in this article.
Company Name (mandatory): You can reference the company name in any other prompts and instructions for the AI agent (for example, in the opening greeting) using double curly-brackets – {{CompanyName}}.

You don't need to type the full variable. Just type { and the system will display an option for you to select.
Opening Greeting (mandatory): This is how the agent introduces itself and greets the caller at the beginning of their conversation.

Company Info: Here you can inform the agent about the company and any specific details about it that it can use for conversing with callers.

You can also include important rules and behavior restrictions for the agent, including how it should handle call transfers or email requests.

Call Transfers

Describe and set up extensions for the AI agent to transfer calls to when a caller requests human assistance or when there is a call situation the AI cannot or shouldn't handle.

Extension can be assigned to a user's line or to any Admin Center destination or feature (a queue, hunt group, department IVR, etc.). Extensions are covered in this article.

You can also set up parameters for when the AI should abort and hang up a call.

Extension: Select the extension the AI agent should transfer calls to.

You can add additional transfer extensions as needed, with the Add extension button or remove them with .

All the listed extensions are available to the AI agent to transfer calls to when the caller asks to escalate to a human or when it cannot handle some aspect of the call. The AI would select the most relevant extension based on its description...
Description: Unlike most feature description fields in the Admin Center, which are optional, the AI transfer extension description field is mandatory and you need to tell the AI agent the purpose of each extension and how to use it to transfer calls.

Department Emails

Describe and set up email addresses for use when the caller asks the AI for a later-time human follow-up (that is not an immediate transfer of the call), to open a support ticket, or just any other reason for emailing a summary of the call.

Email: Enter the address the AI Agent should email.

You can add additional addresses with the Add email button or remove them with .

All the listed addresses are available to the AI agent when the caller asks to email a human or open a support ticket. The AI would select the most relevant email based on its description...
Description: Unlike most feature description fields in the Admin Center, which are optional, the AI email address description field is mandatory, and you need to tell the AI agent the purpose of each address and how to use it to send emails.

AI Receptionists Integrations

The integration feature allows your AI receptionist to go beyond answering questions and perform parallel actions during a live call, like updating records, pulling up information, booking appointments, or triggering processes in external software.

You can create multiple integrations that can be attached to a specific AI agent or shared among several agents.

Technically, we refer to AI integrations as functions or tools since they can be called to action and reused.

The Integration Resource

Resource is a term that refers to any external software, service, database, files, records, APIs, or any other asset that the integration interacts with to perform an action or retrieve information. While the resource refers to the asset in general, the interaction would usually target specific data properties in it. For example, if a resource is a database of clients, the integration will update a specific client's records (like address) that the AI agent identified during the phone call.

It is recommended to keep ongoing backups of the resource if the AI is allowed to modify it!

Accessing and Managing Integrations

You can access and manage integrations in two ways:

Click the button on the top-right main page to the existing integrations page with all the integrations created for the account.

The Add Integration button on this page only lets you create new integrations. Creating, editing, or deleting integrations on this page will apply to all AI receptionists.
Select an AI receptionist and click the Integrations tab to see existing integrations attached to it.

The Add Integration button here gives you the options to attach existing integrations to the receptionist or create new ones. Creating new integration here will make them available to all other receptionists, but if you Remove an integration it will only be removed (detached) from the AI receptionist you selected.

Types of Integrations

There are two methods to integrate the AI receptionists:

MCP Server: Model Context Protocol is a widely adopted standardized architecture for AI tools to integrate and share data with external resources. It integrates the AI with the target system without requiring custom or specific code to connect the two.

Instead, the MCP server (configured and operated by the resource owner and running on its domain) acts as an intermediary and presents the MCP-compatible AI agents with a list of tools they can access and use to interact with the resource.

With MCP server integration, you can define tools in a natural language prompt, and only additional header details may be needed for authentication.
Custom Function: Before integration frameworks like MCP, developers would need to write tailored custom code to connect and configure the interactions between the AI and a resource.

With AI Receptionist custom functions, you don't need to deal with any coding, but you do need to define all the variables for each function and its communication method with the resource.

Which one should you use? When available, MCP integration is faster to set up, only requiring you to research and understand what tools are offered by the MCP server. Custom functions are handy when an MCP server integration is unsupported by the resource or if you need to create a tool for a specific task not covered by existing MCP tools. Of course, you can also combine the two and complement an MCP integration with custom functions.

MCP Server Integration

Server Label: A label for the MCP server that is also the name of the tool/function.

The name needs to be lowercase and include only letters and numbers. Use underscores instead of spaces!
Server URL: The web address of the MCP server.
Description: An optional summary that informs Admin Center users of what the integration does. It appears on the Integration tab. It has nothing to do with the actual performance of the AI (it is not a prompt).
Prompt Instructions: Directions telling the AI how it should process the integration tool, and giving it more details of what it should say or do in certain scenarios.

It is recommended to format the instruction block keeping the "Use when:" and "Requirements:" headers and bullet points of the sample text you will see when creating a new integration.
Headers: Values that the function sends to the server to provide additional metadata, such as format and content definitions, agent identification, tracking and proxies, protocol, and many more properties.

The most common use is providing authentication credentials to access the resource. For example, a bearer token value of "Bearer utitmdewe31631trdewgfi", and in which case you should enable the Secret value toggle to hide the value after saving it.
Defer loading until the receptionist needs this server: Enabling this to instruct the AI to use the tool on demand, accessing it (and the MCP Server) only when the interaction with the caller requires it. This offers an initial faster integration access to the resource (since the AI will not load all tools in advance), but the flipside is that there may be a momentary delay during a call when the AI needs to load the tool.

Depending on the resource and bandwidth, the difference between the two options may be negligible, but you should test and see which one is better for your integration.

Read only: Enable this to make the AI only be able to read data from the resource, preventing it from accidentally deleting or modifying properties.

Allowed Tool Names: Use this to limit which MCP tools the AI is able to access and use. Press the Enter or Comma keys to add each tool name. Leave the box empty to allow the AI to use all available tools.

Custom Function Integration

Function Name: Needs to be lowercase and include only letters and numbers. Use underscores instead of spaces!
Description/Instructions: This is a general description (base prompt) passed to the AI model when it loads the function, telling it what it does and how to use it.

Unlike the MCP server description or other description fields in the Admin Center, this is not optional and must be included with the tool.
Prompt Instructions: Additional instructions added to the base prompt of how the AI should process the function, giving it more details of what it should say or do in certain scenarios.

It is recommended to format the instruction block keeping the "Use when:" and "Requirements:" headers and bullet points with the sample text you will see when creating a new integration.
HTTP Method: The communication method the AI uses to interact and (possibly) modify properties and resources on the endpoint.
- GET: Retrieve information from the resource without making any data modifications to it.
- POST: Creates new properties on the resource.
- PUT: Updates an entire data set of a resource (like a client's entire profile), replacing it with all the data sent in the body of the request. In other words, this is a complete overwrite of the relevant data, and if the request sent by the AI lacks some fields, it may result in the external server replacing them with blank fields.
- PATCH: Updates only the necessary fields of a resource (like a client's address) with the changed data sent in the body of the request, without affecting the rest of the resource. This is more data economical and prevents the blank field problem of the PUT method.
- DELETE: Permanently removes properties from the resource.
Endpoint: A URL pointing to the resources the AI interacts with to perform the actions of the integration.

You can include any variables (explained next) in the URL by enclosing them in curly brackets and using the & symbol for multiple entries. For example, https://resouce-url.com/process-pdir/{name}&{date_of_birth}.

Custom Function Variables

Custom function can have three kinds of variables that can be attached to the Endpoint URL:

Body: This is any data that the AI process, updates, or creates on the resource. For example, we can have a "Topic" variable that the AI will fill in with the reason a client called about.
Header: Header variables can provide additional metadata to the function, such as format and content definitions, agent identification, tracking and proxies, protocol, and many more properties.

However, the most common use is providing authentication credentials to access the resource. For example, a bearer token "Bearer utitmdewe31631trdewgfi", and in which case you should enable the Secret value toggle to hide the value after saving it.
Query: Query variables are used to when the AI needs to obtain additional information from the resource to answer a caller's questions.

For example, if a client calls to ask for a discount on subscription renewal, we can have query variables for the AI to check their current plan, remaining months on it, and if they have any outstanding balance. The tool's prompt instructions will have several lines to calculate the discount accordingly.

Static and Agent Values

The Source option determines how values are assigned to variables.

Static: This is a predefined value that you provide, for example an authentication token.
- Key Name: A label that identifies the variable and the value it contains. It can be referenced in prompt instructions and, using curly brackets, can be included in the Endpoint URL.
  
  While not required for static values, descriptive names would help the AI understand better how to process the variable.
- Value Type: Specify the data type of the variable as either a String, a Number, an Integer, or a Boolean.
- Array of values: Checking this box defines the variable as an array containing multiple values. Press Enter or Comma to add each value in the Static Value box.
- Static Value: The value contained by the variable that is delivered to or fetched from the resource. Make sure it matches the value type, for example a Boolean variable can have a "true" or "false" static value.
- Secret Value: Toggle this on to hide the value in the Static Value box, especially relevant to passwords and authentication data.

Agent: This is a value that the AI fills in dynamically based on the conversation, for example a client's location or area code.
- Key Name and Argument Name: The key name is a label that identifies the variable and the value it contains, while the Argument Name is a label for the Argument Description.
- Argument Type: Specify the data type of the argument as either a String, a Number, an Integer, or a Boolean
- Argument Description: Instructions for the AI about the meaning of the variable and how to assign value(s) to it.
- Array of values: Checking this box lets the AI define the variable as an array containing multiple values.
- Max values: (optional) Limits how many values the AI can include in the array.
- Required argument: Marking an argument as required tells the AI it must obtain a value for the variable before it can process the integration function.
- Missing / Invalid Behavior: Tell the AI what to do when it can't properly process and populate the variable or if the variable contains an invalid value.
  
  You can tell the AI to Send Empty Value instead or Omit Feild to make the AI ignore the variable when processing the function.
  
  Selecting Reject Request tells the AI not to proceed with processing the function. You should include directions in the Prompt Instructions field to tell the AI what to do in such a case.
  
  You can also select Use Default Value to replace the value with the content of the Default Static Value field.
- Enum values: Let you define a list of allowed values that the AI needs to select from. It is applicable when a variable can contain some distinct values and can be useful for preventing erroneous value assignment.
  
  For example, if we have a Subscription_type variable, we can limit the AI to make a selection only from these values –

Articles in this folder