Customizing your AI receptionist's training guide
This helpdesk article is designed to guide you through the process of refining your voice AI agent's training guide to achieve optimal performance. The training guide is the central component that dictates how your AI agent interacts with callers, serving as its core intelligence and operational instructions.
The training guide functions as the system prompt for your AI agent. It is a free text area where all the information necessary for the AI agent's performance, including things it should say, its roles, and its responsibilities, is entered. It can be considered the equivalent of the knowledge a human receptionist holds about your business. The system supports a significant amount of information within the training guide, up to ~10,000 words.
Upon initial setup, you are guided through a process that asks for details like company name, target customers, and what the AI agent should know and do. Based on this input, a draft of the training guide is automatically generated, which aims to get the user approximately 90% of the way to a functional agent.
The remaining 10% is crucial for transforming a functional agent into one that can handle each phone call effectively and perfectly. This transformation is achieved through refinement, an iterative process. It involves designing the prompt (or starting with the auto-generated one), testing its performance, analyzing the results (often by reviewing call logs and transcripts), refining the prompt based on the analysis, and repeating this cycle.
Utilizing structured formatting, such as Markdown, is highly beneficial for organizing information within the training guide, as it helps the AI understand and process the instructions effectively. Key elements typically defined and refined within the training guide include:
- Identity: Defining the AI agent's persona and role (e.g., virtual assistant, sales agent).
- Style: Establishing the communication tone (e.g., informative, polite, conversational). Personality can be added to make the agent sound more human-like.
- Task & Goals: Outlining the specific objectives the agent needs to achieve during a call (e.g., gather information, book appointments, answer questions, take messages).
- Greeting: Customizing the agent's opening statement when answering the phone.
- Specific Scenarios/Logic: Providing detailed instructions on how the agent should handle various caller intents or questions. Instructions for transferring calls to a live person, including when and where to transfer, can also be specified.
- Information Gathering: Clearly stating what information the agent needs to collect from callers (e.g., name, phone number, reason for the call).
It is important to differentiate the training guide from the knowledge base. The training guide contains the prompt that is loaded at the beginning of a phone call and represents the agent's core set of instructions and initial information used to navigate the conversation.
The knowledge base serves as a separate repository of supplementary information that the AI agent can reference during a call if needed. While scripts or documentation can be included in the knowledge base, information that the AI needs to actively use to drive the conversation or must follow strictly is best placed directly into the training guide.
Refining the training guide also allows for enhancing the AI agent's voice realism and conversational tone. This can be achieved by:
- Adding personality and providing a specific tone (e.g., conversational, less corporate jargon).
- Using emotional prompting to shape the agent's tone with expressive language.
- Incorporating natural speech elements like hesitations ("uh," "um"), pauses (indicated by ellipses or multiple periods), stuttering (repeated letters or sounds), and emotional emphasis (capital letters, exclamation marks).
- Instructing the agent to spell out numbers instead of typing them numerically.
- Optionally, training the agent to mimic the speaking style of the caller.
Thorough testing is essential to ensure the training guide is performing as expected. The system provides three primary methods for testing:
Simulated text chat: This allows users to see how the agent responds to various inputs based on the training guide.
Call agent in browser: This allows you to have a conversation with your agent through your web browser.
Calling the AI agent's unique phone number: This is the most effective way to test before going live, allowing users to role-play different scenarios and experience the conversation flow from the caller's perspective.
Refinements made to the training guide are immediately live once updated.
Users have full control over prompting to define the agent's role and behavior. Specific instructions can be given for handling various situations. Negative constraints can also be included, such as:
- Instructing the agent never to say certain words or phrases.
- Preventing the agent from moving too quickly or rushing the caller.
- Updating the prompt to handle repeated questions effectively.
Becoming an expert in prompt engineering takes time. While users have tools to refine the prompt and support is available, additional assistance is also provided.
- Using External AI Tools: Users can copy their training guide into large language models like ChatGPT and instruct the AI to make changes or reformat the prompt in markdown.
- White-Glove Setup Service: For those who require more hands-on assistance, a white-glove setup service is available. This service includes working directly with users to build and refine the training guide.
By dedicating time to structure and refine the training guide through iterative testing and utilizing available resources, you can significantly enhance their voice AI agent's performance and ensure it meets their specific operational needs.
Understanding the Training Guide
The training guide functions as the system prompt for your AI agent. It is a free text area where all the information necessary for the AI agent's performance, including things it should say, its roles, and its responsibilities, is entered. It can be considered the equivalent of the knowledge a human receptionist holds about your business. The system supports a significant amount of information within the training guide, up to ~10,000 words.
The Initial Setup and the Need for Refinement
Upon initial setup, you are guided through a process that asks for details like company name, target customers, and what the AI agent should know and do. Based on this input, a draft of the training guide is automatically generated, which aims to get the user approximately 90% of the way to a functional agent.
The remaining 10% is crucial for transforming a functional agent into one that can handle each phone call effectively and perfectly. This transformation is achieved through refinement, an iterative process. It involves designing the prompt (or starting with the auto-generated one), testing its performance, analyzing the results (often by reviewing call logs and transcripts), refining the prompt based on the analysis, and repeating this cycle.
Structuring the Training Guide
Utilizing structured formatting, such as Markdown, is highly beneficial for organizing information within the training guide, as it helps the AI understand and process the instructions effectively. Key elements typically defined and refined within the training guide include:
- Identity: Defining the AI agent's persona and role (e.g., virtual assistant, sales agent).
- Style: Establishing the communication tone (e.g., informative, polite, conversational). Personality can be added to make the agent sound more human-like.
- Task & Goals: Outlining the specific objectives the agent needs to achieve during a call (e.g., gather information, book appointments, answer questions, take messages).
- Greeting: Customizing the agent's opening statement when answering the phone.
- Specific Scenarios/Logic: Providing detailed instructions on how the agent should handle various caller intents or questions. Instructions for transferring calls to a live person, including when and where to transfer, can also be specified.
- Information Gathering: Clearly stating what information the agent needs to collect from callers (e.g., name, phone number, reason for the call).
Training Guide vs. Knowledge Base: A Key Distinction
It is important to differentiate the training guide from the knowledge base. The training guide contains the prompt that is loaded at the beginning of a phone call and represents the agent's core set of instructions and initial information used to navigate the conversation.
The knowledge base serves as a separate repository of supplementary information that the AI agent can reference during a call if needed. While scripts or documentation can be included in the knowledge base, information that the AI needs to actively use to drive the conversation or must follow strictly is best placed directly into the training guide.
Enhancing Voice Realism and Tone
Refining the training guide also allows for enhancing the AI agent's voice realism and conversational tone. This can be achieved by:
- Adding personality and providing a specific tone (e.g., conversational, less corporate jargon).
- Using emotional prompting to shape the agent's tone with expressive language.
- Incorporating natural speech elements like hesitations ("uh," "um"), pauses (indicated by ellipses or multiple periods), stuttering (repeated letters or sounds), and emotional emphasis (capital letters, exclamation marks).
- Instructing the agent to spell out numbers instead of typing them numerically.
- Optionally, training the agent to mimic the speaking style of the caller.
Testing Your Refinements
Thorough testing is essential to ensure the training guide is performing as expected. The system provides three primary methods for testing:
Simulated text chat: This allows users to see how the agent responds to various inputs based on the training guide.
Call agent in browser: This allows you to have a conversation with your agent through your web browser.
Calling the AI agent's unique phone number: This is the most effective way to test before going live, allowing users to role-play different scenarios and experience the conversation flow from the caller's perspective.
Refinements made to the training guide are immediately live once updated.
Addressing Common Refinement Challenges
Users have full control over prompting to define the agent's role and behavior. Specific instructions can be given for handling various situations. Negative constraints can also be included, such as:
- Instructing the agent never to say certain words or phrases.
- Preventing the agent from moving too quickly or rushing the caller.
- Updating the prompt to handle repeated questions effectively.
Seeking Assistance with Refinement
Becoming an expert in prompt engineering takes time. While users have tools to refine the prompt and support is available, additional assistance is also provided.
- Using External AI Tools: Users can copy their training guide into large language models like ChatGPT and instruct the AI to make changes or reformat the prompt in markdown.
- White-Glove Setup Service: For those who require more hands-on assistance, a white-glove setup service is available. This service includes working directly with users to build and refine the training guide.
By dedicating time to structure and refine the training guide through iterative testing and utilizing available resources, you can significantly enhance their voice AI agent's performance and ensure it meets their specific operational needs.
Updated on: 23/04/2025
Thank you!