Create an AI voice assistant

Create an AI voice assistant to enable natural, conversational voice interactions between users and AI voice agents.

Before you begin

Role required: virtual_agent_admin or admin

Set up your preferred user identification and authentication methods to allow access to AI voice agents. See for more information.

About this task

An AI voice assistant enables natural, conversational voice interactions between users and AI voice agents. It uses speech-to-text (STT), large language model (LLM), and text-to-speech (TTS) to understand and respond to callers in real time. You can configure a voice assistant with personalized voice and welcome message, fallback options, and assign AI voice agents with specific AI instructions. The fallback options include live agent transfer and ticket creation, based on the origin of the call.

Procedure

1. Navigate to All > Conversational Interfaces > Assistant Designer > Assistants and select Create assistant.
2. Select Voice-only option in the Create an assistant window and select Continue.

   Figure 1. Voice-only option in Create assistant

   

3. Add basic details of the assistant.

   Figure 2. Basic details form

   

   1. On the form, fill in the fields.

      Table 1. Basic details form

      Field	Description
      Name	Name of the voice assistant. Provide a name according to the business outcome that the voice assistant targets. For example: HR Service Desk
      Description	Brief summary of the business outcomes that your voice assistant targets. Example for the HR Service Desk: HR service desk to help resolve employee requests and inquiries.
      Tags	Add tags to track analytics for the voice assistant. For example, HR. Adding or removing tags will apply the changes immediately. These will be tracked in analytics.

   2. Select Save and continue.
      You’re directed to the AI agents page.
4. Add one or more AI voice agents to the voice assistant by selecting Add from library and select Save and continue.
   Note:

   Adding AI agents is optional. If no AI agents are added, you can add them later by editing this assistant. The assistant will be inactive. Select Add from library to add an existing agent, or select Create to create a new one. See Create an AI voice agent for more information.
5. Select a voice personality.

   Figure 3. Voice personality selection

   

   1. Select the primary language your assistant will use for interacting with callers.
      You can configure a primary language and up to three secondary languages. You can select from the following languages:

      • English
      • German
      • Spanish
      • Japanese
      • Korean
      • Dutch
      • Brazilian Portuguese
      • Italian
      • Mandarin Simplified
      • French
      • Canadian French
      • British English
      • Mexican Spanish
      • Thai
      • Hindi

      See Multilingual support for voice assistants for more information.
   2. Add a personalized welcome message to greet the callers calling into the voice assistant.
   3. On the Voice persona tab, select a voice persona that best suits the conversational experience that you want to deliver through the voice assistant.
      Preview the voice samples to determine the appropriate voice and tone. All AI voice agents connected to the voice assistants share the same voice.
   4. Optional: Select + Add language to add secondary languages.
      For each secondary language, select the corresponding voice persona. The order of secondary languages reflects the caller experience during language selection.
   5. Select Save and continue.
      You’re directed to the Communication channels page.
6. Identify and authenticate the caller.

   Authentication settings apply only to telephony provider communication channel. If you have selected only mobile communication channel, skip this step.

   Figure 4. Caller identification and authentication method selection

   

   Identification and authentication factors must be configured at the platform level, where you define which tables and columns the system should use for both identification and authentication. After the factors are defined, they appear here as selectable options for your voice agent configuration. For more information, see .

   1. Select the method used to identify the caller when the call begins.
      Caller identification determines who the caller is before any authentication occurs. The information provided by the caller is matched with system records to identify the caller. If you select phone number as the primary identification method, enable the checkbox to allow the system to automatically match the caller’s phone number to the incoming caller Id received from the telephony system.
   2. Optional: Select the fallback method to identify the caller.
      If the caller can't be uniquely identified through the primary method, the system automatically retries using the fallback method.
   3. Enable caller access to AI voice agents by selecting the authentication type, First factor, and Second factor methods.
      Caller authentication verifies the caller’s identity before allowing access to AI voice agents that require authentication.

      Select from the following authentication types.

      • Multi-factor authentication (MFA) requires callers to successfully complete more than one verification method configured before the system grants access. MFA is required by default.
      • Single factor authentication requires the user to complete the configured verification method. To enable single factor authentication, you must set the glide.voice.authenticate.mfa_mandatory system property to false.

      Select from the following First factor authentication methods.

      • Knowledge-based authentication (KBA)
      • Okta Verify push notification
      • SMS verification code
      • Authenticator app time-based One Time Password (TOTP)
      • Soft PIN

      Select from the following Second factor authentication methods.

      Numeric authentication factors such as SMS verification code, Authenticator app time-based One Time Password (TOTP), and Soft PIN support voice input. Callers can respond verbally instead of using the keypad. Voice input for each factor can be configured at platform level and scoped per voice service. See for more information.

      Note:

      The option selected as the First factor is not available in the Second factor dropdown. KBA authentication, for example, employee security questions, requires questions to be configured at platform level with the Channel field set to Voice and the Type field set to Identification, Authentication, or both. Questions configured this way automatically appear in Assistant Designer for selection. Explicit service mapping is not required. For secure and consistent verification, KBA authentication factor must use numeric data only, for example, date of birth, Social Security Number, or employee Id. Additionally, the source table used must reference the sys_user table so that caller identity can be validated reliably across the platform. See for more information.

      KBA now also supports external authentication. The caller's system ID and their spoken response are passed to a verification script, which checks the answer against an external system and returns a true or false result. Optionally, context from earlier responses in the same session can be passed to the script. Configure external authentication scripts at platform level.

   4. Select Save and continue.
      You’re directed to the Safeguards page.
7. Set up safeguards to create a secure and seamless experience for users interacting with the assistant.

   Figure 5. Safeguards selection

   

   1. Set fallback options to route the call to a live agent or create a ticket.
      When a voice agent cannot complete a user's request, the system determines the appropriate fallback behavior based on the call origin, for example, telephony provider or mobile channel.

      • Telephony provider requires either connect to live agent or a record producer as fallback option.
      • Mobile channel requires a record producer as fallback option.

      • Connect to live agent option. When selected, this option redirects the caller to a live agent. You must set up live agent transfer for your telephony provider separately.
        Note:

        You can enable the Capture details before live agent handoff option, in which the voice agent prompts the caller to provide details in order to triage the call to the appropriate live agent.
      • Generate a ticket with record producer option. When selected, this option creates a ticket for further tracking.
        Note:

        If you choose to use generating a ticket with record producer as the fallback option, you must keep the fields in the record producer simple and short to optimize the user experience for both the communication channels. For example, a short description, description, and an optional field for the callback number should suffice. You can also enable the Require authentication option, which restricts ticket creation to authenticated employees.
   2. Set the time limits for call duration and reprompting users after inactivity.
      • Set Max call duration to trigger fallback behavior when the call reaches this limit. You can set up to 10 minutes.
      • Set the duration of inactivity after which the user is reprompted for a response. If there's still no response, the call is disconnected. You can set up to 300 seconds.
   3. Select Save and continue.
      You're directed to the Advanced settings page.
8. Review your voice assistant configuration.

   You can change the configuration later.

9. Select Save and activate to complete the configuration steps or review a previous step by selecting Back.
   Activating the assistant also activates any inactive AI agents associated with it.The following conditions must be met in order to activate the assistant. If any of these conditions are not met, select Save and close to return later.

   • At least one communication channel is selected (telephony provider or mobile channel)
   • At least one AI agent is associated with the assistant
   • If a telephony provider is enabled, authentication is set up and fallback is configured to either connect to a live agent or create a record producer.
   • If a mobile channel is enabled, fallback is configured to create a record producer.

What to do next

Test the execution of your AI voice agent by manually calling in the telephony number to see if the AI voice agent functions the way you defined it. Review the transcript and logs for troubleshooting and improving the conversational experience of users. See AI voice agent transcript and logs tables for information on the tables containing transcript and logs.