- Home
- /
- Article
Webex AI Agent Studio Administration guide
This article is available as part of an early release. The content is a work in progress and may change as we enhance our product. We welcome your feedback to help improve the documentation.
Get started with Webex AI Agent Studio
Webex AI Agent Studio is a sophisticated platform that is designed to create, manage, and deploy automated AI agents to fulfill customer service and support needs. Using artificial intelligence, AI agents provide automated assistance to customers before they interact with human agents. These agents support voice interactions with intonation, language understanding, and contextual awareness within conversations. Also, AI agents seamlessly and informatively handle digital channel interactions through text and online chat. Customers benefit from a concierge-like experience, receiving assistance with questions, information retrieval, and minimizing wait times.
Key benefits for businesses
-
Efficiency & productivity
-
Automates repetitive tasks, freeing human employees for strategic work.
-
Provides 24/7 availability, handling higher volumes and improving response times.
-
Improves accuracy in data processing, analysis, and reporting.
-
- Cost savings
-
Reduces labor costs through automation.
-
Optimizes resource allocation for greater efficiency.
-
Lowers operational costs by streamlining processes and preventing errors.
-
- Enhanced decision-making
-
Provides data-driven insights by analyzing large datasets.
-
Enables predictive analytics for anticipating future outcomes.
-
Improves risk management through identification and assessment.
-
- Improved customer experience
-
Personalizes interactions based on customer data.
-
Offers faster response times through AI-powered chatbots and virtual assistants.
-
Provides 24/7 customer support.
-
- Scalability & flexibility
-
Easily scales up or down to meet changing business needs.
-
Learns and adapts to new situations and information.
-
- Competitive advantage
-
Drives innovation and development of new products/services.
-
Increases efficiency for a competitive edge.
-
- Employee Empowerment
-
Frees human potential for more creative and strategic work.
-
Enhances collaboration by providing data, insights, and support.
-
Understand AI agent types and examples
The following table provides a glimpse of AI agent types and their capabilities:
AI agent type | Purpose | How to set up? |
---|---|---|
Autonomous |
Autonomous agents work independently to fulfill the defined goals, reducing the
need for continuous human intervention.
|
Set up autonomous AI agent |
Scripted |
Scripted AI agents are programmed to follow a set of predefined rules and
instructions.
|
Set up scripted AI agent |
Examples
Both autonomous and scripted AI agents apply to various use cases, depending on the specific requirements and desired capabilities. Some examples include:
-
Customer service—Both autonomous and scripted agents can provide customer support, with autonomous agents offering more flexibility and understanding of natural language.
-
Virtual assistants—Autonomous agents are well suited for virtual assistant roles as they can manage a wide range of tasks and offer more personalized interactions.
The choice between autonomous and scripted AI agents depend on the complexity of the tasks, the required level of autonomy, and the availability of training data.
Prerequisites
- If an organization doesn't have an active contact center trial or subscription, partners can set up a contact center trial with the Webex AI agent feature. For more information, see Webex Contact Center Self-Service Trials article.
- To sign up for subscription, customers need to purchase the AI Agent add-on for the Flex-3.0 Contact Center license. Based on entitlements, Webex Contact Center provisions AI Agent as one of its services for an organization. For more information about provisioning the Webex Contact Center, see Get Started with Webex Contact Center.
Access Webex AI Agent Studio
To create your AI agents, you must sign in to the Webex AI Agent Studio application. You can sign in using the following ways:
Sign-in from Control Hub
- Sign in to Control Hub.
-
Select Services > Contact Center.
- From the Contact Center navigation pane, select Customer experience > AI Agent.
- Click Webex AI Agent Studio to access the application.
This opens the Webex AI Agent Studio application in another browser tab, and you’re ready to configure your AI Agents.
Sign-in from Webex Connect
To access the Webex AI Agent Studio application, you should have access to Webex Connect.
- Sign in to Webex Connect application using the tenant URL provided for your enterprise and credentials.
By default, the Services page appears as a home page.
- From the App Tray menu of the left navigation pane, click Webex AI Agent Studio to access the application.
The system opens the Webex AI Agent Studio application in another browser tab and you’re automatically signed-in to the application.
Home page layout
Welcome to the Webex AI Agent Studio application. When you sign in, the home page displays the following layout:
-
Navigation bar
The navigation bar that appears on the left provides access to the following menus:
- Dashboard—Displays a list of AI agents the user has access to, as granted by the enterprise administrator.
- Knowledge—Shows the central knowledge repository or knowledge base, which serves as the brain for autonomous AI agents to respond to customer queries.
- Help—Provides access to the Webex AI Agent Studio user guide on the Webex Help Center.
-
User profile
The user profile menu allows you to view your profile information and sign out of the application.
The Enterprise Profile page contains information about the AI agent tenant, accessible only to administrators with full-admin access.
-
The Overview tab contains the following information:
- Enterprise identifiers—Includes Webex Org ID, CPaaS Org ID, Subscription ID for the enterprise. This is available for enterprises with Webex Contact Center integration for the corresponding Webex Connect tenant.
- Profile settings—Contains enterprise name, enterprise unique name, and the Logo URL.
- Global Agent settings—Allows selection of the default agent for voice channel to handle fallback scenarios.
- Data retention summary—Provides a summary of data retention periods for this enterprise.
-
In the Teammates tab, you can view and manage the list of teammates who have access to the application. Each user is assigned a role, which determines the actions they can perform based on granted permissions.
-
Know your Dashboard
On the dashboard, the AI agents are represented by cards. Each card displays basic information, including the AI agent name, last updated by, last updated on, and the engine used for training the agent.
Tasks on AI agent card
Hover over an AI agent card to view the following options:
- Preview—Click Preview to open the AI agent preview widget.
- Ellipsis icon—Click this icon to perform the following tasks:
-
Copy Preview link—Copy the preview link to paste in a new tab and preview the AI agent on the chat widget.
-
Copy Access token—Copy the AI agent's access token for invoking the agent through APIs.
-
Export—Export the AI agent details (in JSON format) to your local folder.
-
Delete—Permanently delete the AI agent from the system.
-
Pin—Pin the AI agent to the first position on the dashboard, or unpin to move it back to its previous position.
-
Create a new AI agent
You can create a new AI agent by using the + Create agent option on the top-right corner of the dashboard. You can choose to use a predefined template or create an agent from scratch.
To know how to create scripted and autonomous AI agents, see the following sections:
Import AI agent
You can import an AI agent in JSON format from a list of available AI agents. First, ensure you’ve exported the AI agent in JSON format to your local folder. Follow these steps to import it:
- Click Import agent.
- Click Upload to upload the AI agent file (in JSON format) exported from the platform.
- In the Agent name field, enter the AI agent name.
- (Optional) In the System ID, edit the system-generated unique identifier.
- Click Import.
Your AI agent is now successfully imported to the Webex AI Agent Studio platform and is available on the dashboard.
Keyword search
The platform provides robust search capabilities to help you easily locate and manage AI agents. You can perform keyword search using the agent name. Enter the agent name or a portion of the name in the search bar. The system displays a list of AI agents that match your search criteria.
Filter by agent type
In addition to keyword search, you can refine your search results by filtering based on the type of AI agent. Choose one of the agent type filters from the drop-down list—Scripted, Autonomous, and All.
Manage knowledge bases
A knowledge base is a central repository of information for the Large Language Model (LLM)-powered AI agents. The AI agents use advanced AI and machine learning technologies to understand, process, and generate human-like text. The AI engine trains the AI agents using vast amount of data, enabling them to provide detailed and contextually relevant responses. Knowledge bases store the data necessary for the functioning of autonomous AI agents.
Create the required knowledge base for the AI agent that you're configuring.
Create knowledge base for AI agent
1 |
Log in to the Webex AI Agent Studio platform. | ||||||||||||||
2 |
Click the Knowledge icon on the left navigation pane. | ||||||||||||||
3 |
On the Knowledge bases page, click +Create Knowledge base on the upper right corner. | ||||||||||||||
4 |
On the Create knowledge base page, enter the following details: | ||||||||||||||
5 |
Click Create. The system creates a knowledge base with the specified name. | ||||||||||||||
6 |
You can either upload files to the knowledge base or create documents from scratch.
To delete all the uploaded files, click the Delete icon next to the Processed files section. To delete a single file, click the Delete file icon next to the file name.
| ||||||||||||||
7 |
Navigate to the Information tab to view the following details:
|
What to do next
Configure the knowledge base for the AI agents.
Access knowledge base
Before you begin
Create the knowledge base required for the AI agent.
1 |
Log in to the Webex AI Agent Studio platform. |
2 |
Click the Knowledge icon on the left navigation pane. The knowledge bases appear as cards on the Knowledge bases page. The cards display the number of files uploaded and the number of documents created in the knowledge base. |
3 |
Click on a card to navigate to the specific knowledge base. You can also access the knowledge base from the Knowledge tab of the AI agent configuration page. For more information, see Configure knowledge base section. You can search for the required knowledge base using the following criteria:
Click Reset all to reset the search criteria. |
Set up autonomous AI agent
Autonomous AI agents operate independently without direct human intervention. Autonomous agents can access and use a knowledge repository to provide informative and accurate answers to user queries. These agents use advanced algorithms and machine learning techniques to analyze data, learn from their environment, and adapt their actions to achieve specific goals.
You can use the autonomous AI agents in various scenarios, including:
- Provide customer support—Answer FAQs, troubleshoot issues, and guide customers through processes.
- Offer technical assistance—Provide expert advice on specific topics or domains.
- Natural Language Processing (NLP)—Understand and respond to human language in a natural and conversational manner.
- Decision making—Make informed choices based on available information and predefined rules.
- Automation—Automate repetitive or time-consuming tasks.
Guardrails configured in the autonomous AI agents ensure that the AI agent doesn't respond with unethical and harmful content.
Create an autonomous AI agent
Before you begin
Ensure to create the knowledge base for the autonomous AI agent. For more information, see Create knowledge base for AI agent.
1 |
Log in to the Webex AI Agent Studio platform. |
2 |
On the Dashboard, click +Create agent. |
3 |
On the Create an AI agent screen, choose Start from scratch and click Next.
You can also choose a predefined template to create your AI agent quickly. You can filter the AI agent type as Autonomous. In this case, the fields on the Profile page autopopulate. |
4 |
Choose Autonomous agent type. |
5 |
Specify the following required details: |
6 |
Click Create. You've now successfully created the autonomous AI agent which is now available on the Dashboard. On the AI agent header, you can perform the following tasks:
You can also import the prebuilt AI agents. For more information, see Import AI agent. |
What to do next
Configure the autonomous AI agent.
Configure autonomous AI agent
The following sections guide you on how to configure autonomous AI agent for your specific needs:
Update autonomous AI agent profile
Before you begin
Create an autonomous AI agent.
1 |
On the Dashboard, click the AI agent that you've created. |
2 |
Navigate to the tab and configure the following details: |
3 |
Click Save changes. |
4 |
Click Publish to make the AI agent live. For more information, see the Publish your autonomous AI agent section. |
What to do next
Add the required actions to the AI agent.
Add actions to autonomous AI agent
Autonomous AI agents are designed to comprehend user intents and act accordingly. For example, consider a restaurant with the need to automate online food order intake. To accomplish the task, create an autonomous AI agent that performs the following actions:
-
Get the required information from the customer.
-
Transfer the information to the required flow.
-
Deliver the customer requirement.
The autonomous AI agent works on the following three building blocks:
-
Action—A task that an AI agent performs by understanding user intents and completes by connecting to external systems.
-
Entity or slot—Represents a step in fulfilling the user's intent. Slot filling involves asking specific questions to the customer to fulfill the customer's intent based on utterances. It's the trigger for an AI agent to start performing an action.
-
Fulfillment—Determines how the AI agent completes the action by connecting with external systems.
1 |
On the Dashboard, click the AI agent that you've created. |
2 |
Navigate to tab.The Actions page displays the Agent handover action. The Agent handover action is enabled by default allowing the AI agent to escalate the conversation to a human agent. Use the toggle option to disable it. |
3 |
Click +New Action to add a new action for an AI agent. |
4 |
On the Add a new action page, specify the following details: |
What to do next
You can either configure slots only or slots along with fulfillment depending on the action scope chosen.
Configure slot filling
Slot filling involves adding the required input entities for the AI engine. In the Slot filling section of the Actions page, add the input entities:
-
You can add the entities one by one in table format. For more information, see Add input entities in table format.
-
You can also use the JSON file and define the entities. See A Tour of JSON Schema for details.
Add input entities in table format
1 |
To add an input entity, click +New input entity. |
2 |
On the Add a new input entity page, specify the following details: |
3 |
Click Add to add the input entity. You can add as many input entities as you need. |
4 |
Click Add to add the action to the AI agent. |
5 |
Click Publish to make the AI Agent live. For more information, see the Publish your autonomous AI agent. After adding the action, use the Controls option to edit or delete the entities. |
Add entities using a JSON editor
- To add an input entity in JSON, click Use JSON instead.
- Enter the input parameter schema in JSON format.
- Click Add.
For more information, see A Tour of JSON Schema.
Input parameter structure
The input parameters must adhere to the following structure:
-
type—Data type of the parameters object. This is always 'object' to denote that the parameters are structured as an object.
properties—An object where each key represents a parameter and its associated metadata.
required—An array of strings listing the names of parameters that are mandatory.
properties Object
Each key in the properties object represents an input entity/parameter and contains another object with metadata about that parameter. The metadata should always include the following keywords:
-
type—Data type of the parameter. The allowed types are:
-
string—Textual data.
-
integer—Numeric data without decimals.
-
number—Numeric data that can include decimals.
-
boolean—True/false values.
-
array—A list of items, all of which are typically of the same type.
-
object—A complex data structure with nested properties.
-
-
description—A brief explanation of what the entity represents. This helps the AI engine understand the purpose and usage of the parameter. A description that's concise and consistent with the agent's instructions and action description is good for better accuracy.
-
Validation is enforced by the platform for ‘type’ only. ‘Description’ isn't enforced for all entities but it's highly recommended that it’s added. Other useful keywords for entity metadata are:
-
enum—The enum field lists the possible values for a parameter. This is useful for parameters that should only accept a limited set of values. Developers can define custom lists of values that a parameter should accept to use this.
- pattern—Use the pattern field with string types to specify a regular expression that the string must match. This is useful for validating specific formats, such as phone numbers, postal codes, or custom identifiers.
-
examples—The examples field provides one or more examples of valid values for the parameter. This helps the AI engine understand the kind of data it needs and can be especially useful for interpretation and validation purposes.
-
There are other keywords that can make the entity definition more accurate and robust. For more information, see A Tour of JSON Schema.
Example
The following example includes various types of entities and keywords:
{
"type": "object",
"properties": {
"username": {
"type": "string",
"description": "The unique username for the account.",
"minLength": 3,
"maxLength": 20
},
"password": {
"type": "string",
"description": "The password for the account.",
"minLength": 8,
"format": "password"
},
"email": {
"type": "string",
"description": "The email address for the account.",
"pattern": "\w+([-+.']\w+)*@\w+([-.]\w+)*\.\w+([-.]\w+)*"
},
"birthdate": {
"type": "string",
"description": "The birthdate of the user.",
"examples": ["mm/dd/YYYY"]
},
"preferences": {
"type": "object",
"description": "User preferences settings.",
"properties": {
"newsletter": {
"type": "boolean",
"description": "Whether the user wants to receive newsletters.",
"default": true
},
"notifications": {
"type": "string",
"description": "Preferred notification method.",
"enum": ["email", "sms", "push"]
}
}
},
"roles": {
"type": "array",
"description": "List of roles assigned to the user.",
"items": {
"type": "string",
"enum": ["user", "admin", "moderator"]
}
}
},
"required": ["username", "password", "email"]
}
This example includes the following entities:
- username—A string type with minimum and maximum length constraint.
- password—A string type with a minimum length and a specific format (password indicates that secure handling is required).
- email—A string type with a regex pattern to ensure it’s a valid email address.
- birthdate—A string type with examples to prescribe the format of the date.
- preferences—An object type with nested properties (newsletter and notifications), including a boolean with a default value and a string with specific allowed values (enum).
- roles—An array type where each item is a string limited to specific values
(enum).
The username, password, and email are mandatory as defined by the ‘required’ array.
In this example, the entities have descriptive names, clear descriptions, and follow a consistent structure and naming convention. Follow these best practices to create well-defined entities that are easy for the AI engine to interpret and enforce.
Configure fulfillment
You can configure fulfillment flows for AI agent actions on Webex Connect Flow builder. For more information, see Configure Fulfillment Flows for AI Agent Actions.
1 |
Go to the Actions tab on the AI agent configuration page. Choose the action for which you want to configure the fulfillment flow. |
2 |
In the Webex Connect Flow Builder Fulfillment section, configure the following settings: |
3 |
Click Save changes to complete the cofiguration. |
What to do next
Configure the knowledge base.
Configure knowledge base
Before you begin
Create an autonomous AI agent.
1 |
On the Dashboard page, click the AI agent that you've created. |
2 |
Navigate to the tab. |
3 |
Choose the required knowledge base from the drop-down list. You can't associate a knowledge base if you have already mapped it to an AI agent. |
4 |
Click Save changes. |
5 |
Click Publish to make the AI agent live. For more information, see Publish your autonomous AI agent. |
What to do next
Click language and voice for autonomous AI agent. For more information, see Configure language and voice.
Configure language and voice
1 |
On the AI agent configuration page, navigate to the Configuration > Language tab. The default language and voice are set to English. For other supported languages and voices, see the Supported languages and voices article. |
2 |
Choose the desired language and locale from the drop-down list. |
3 |
Choose the appropriate voice from the drop-down list. You can use the default voice or choose a different voice from the available list. The list of available voices appears automatically based on the chosen language. |
4 |
Click Publish to make the AI agent live. For more information, see the Publish your autonomous AI agent section. |
5 |
Choose the AI Agent supported languages from the drop-down list.
|
What to do next
Click Preview to preview the AI agent. For more information, see Preview your Autonomous AI Agent. Click Publish to make the AI agent live. For more information, see the Publish your autonomous AI agent section.
Preview your autonomous AI agent
You can preview the autonomous AI agents at the time of creating the AI agent, while editing, and after deploying the agent. You can open the preview from:
- AI agent dashboard—On hovering over an AI agent card, the Preview option for that AI agent become visible. Click to open the preview of the AI agent.
- AI agent header— Click on the AI agent card to open the AI agent. The Preview option is always visible in the header section.
- Minimized widget—After you launch the preview and minimize it, a chat head widget appears at the bottom right of the page. You can use this option to easily reopen the preview mode.
Webex AI Agent Studio also provides a shareable preview option. Click the menu on the top-right corner and select the Copy Preview Link option. You can share the preview link with other users such as testers or consumers of the AI agent.
Platform preview widget
The preview widget appears on the bottom-right section of the screen. You can provide utterances (or a sequence of utterances) to check the AI agent's responses and ensure it’s functioning correctly.
Also, you can minimize the preview widget, provide consumer information, and initiate multiple rooms to test the AI agent.
Publish your autonomous AI agent
After configuring and previewing your AI agent, you can publish the agent to make it live.
Before you begin
Create the autonomous AI agent.
1 |
On the AI agent configuration page, click Publish. |
2 |
On the Publish changes screen, enter the Version name and click Publish. You can view the version details on the History page. For more information, see the History section. |
View autonomous AI agent sessions and history
You can view the details of sessions established with the customers and the history of the configuration changes performed on the AI agent.
Sessions
The Sessions page provides a comprehensive record of all interactions between AI agents and users. To access Sessions:
-
On the Dashboard, click the autonomous AI agent for which you want to view the session details.
-
From the left navigation pane, click Sessions.
The Sessions page appears. Each session is displayed as a record that contains all the messages of the session. This information is useful to audit, analyze, and improve the AI agent.
The sessions table shows a list of all the sessions/rooms created for that AI agent. The table gets paginated if there are more rows than can be accommodated in one screen. Any of the fields in the table can be sorted or filtered using the Refine Results section on the left-hand side. The fields which are present represent the following information about any particular session:
-
Session ID—The unique room id or session id for a conversation.
-
Consumer Id—The id of the consumer who interacted with the AI agent.
-
Channels—Channel where the interaction took place.
-
Updated At—Time of the room closure.
-
Room Metadata—Contains additional information about the room.
-
Check the required check boxes:
-
Hide test sessions—To hide the test sessions and display only the list of live sessions.
-
Agent handover happened—To filter the sessions that are handed over to an agent. If agent handover happens, it displays the Headphone icon indicating the handover of the chat to a human agent.
-
Error occurred—To filter the sessions in which the error occurred.
-
Downvoted—To filter the downvoted sessions.
-
View session details
To view the session details:
-
Click on an individual row in the sessions table for a detailed view of that session. If the session is locked, you need to have permission to decrypt the session.
-
Click the Decrypt content to view the session data.
This functionality is applicable only when the Advanced data protection is set to true or enabled for the tenant.
-
The system displays the following sessions details:
-
The left panel displays details about the transactions.
-
The right panel displays details about slot filling and fulfillment related to all actions. Use the Expand all button to expand the transaction. The right panel displays details about the knowledge utilization with details about document names and files uploaded.
-
History
The History page allows you to view the details of the configuration changes performed on the AI agent. To view the history of a specific agent:
-
On the Dashboard, click the autonomous AI agent for which you want to view the history.
-
From the left navigation pane, click History.
The History page appears with the following tabs:
-
Version history—Click the Version History tab to view the various versions of the autonomous AI agent.
-
Change logs—Click the Change Logs tab to view the changes made to the AI agents.
Version history
Whenever you publish the autonomous AI agent, a version of the autonomous AI agent is saved and is available in the Version history tab. You can view the various versions of the AI agent from the Version history tab.
-
Version description—A brief description about the version of the AI agent.
-
AI engine—The AI engine used for that version of the AI agent.
-
Updated at—Date and time when the version was created.
-
Actions—Allows you to perform the following actions on the AI agent:
-
Load as draft—All changes on the AI agent is lost. You must perform the configuration again.
-
Export—Use to export the AI agent.
-
Change logs
The Change logs tab tracks the changes made to the autonomous AI agent. You can view the details of the changes for the past 35 days. The Change logs tab displays the following details:
Users with Admin or AI agent developer roles can only access the Change logs tab. Users with custom roles that have the ‘Get Audit log’ permission can also view the audit logs.
-
Updated at—The data and time of the change.
-
Updated by—The name of the user who incorporated the change.
-
Change Location—The specific section of the AI agent where the change was made.
-
Description—Additional information about the change.
You can search for a specific audit log using the Updated by, Change Location, and Description search options. You can sort the logs based on the Updated at and Updated by fields.
View Autonomous AI agent performance using analytics
The AI agent analytics section provides a graphical representation of the key metrics to evaluate the AI agent performance and effectiveness. To generate the analytics of the Autonomous AI agent:
- Choose the AI agent from the Dashboard.
- On the left navigation pane, click Analytics. An overview of the AI agent performance appears in both tabular format and graphical representation.
The first section displays the following statistics about sessions and messages for the AI agent.
- Total sessions and sessions handled by the AI agent without human intervention.
- Total agent handovers, which is a count of number of sessions handed over to human agents.
- Daily average sessions.
- Total messages (human and AI agent messages) and how many of those messages came from users.
- Daily average messages.
The second section displays the statistics about the users. It provides a count of total users and information about average sessions per user and daily average users.
The third section displays the AI agent responses and agent handovers.
Set up scripted AI agent
Scripted AI agents enhance the no-code agent-building capabilities of the Webex AI Agent Studio platform. They enable multiturn conversations, gathering relevant data from customers to perform specific tasks. This includes:
-
Running simple commands—Follow instructions to complete predefined actions.
-
Processing data—Manipulate and transform data according to specified rules.
-
Interacting with other systems—Communicate with and control other solutions.
Scripted AI agents are knowledge-driven agents whose knowledge base consists of a corpus of questions and answers. Scripted AI agent can provide answers based on a user-created training corpus, which is a collection of examples and answers. This capability is useful in scenarios where:
-
Specific knowledge is required—The agent needs to answer questions within a predefined domain.
-
Consistency is important—The agent must provide consistent responses to similar queries.
-
Limited flexibility is needed—The agent's responses are constrained by the information in the training corpus.
Create a scripted AI agent
1 |
Log in to the Webex AI Agent Studio platform. |
2 |
On the dashboard, click + Create agent. |
3 |
On the Create an AI Agent screen, choose Start from scratch and click Next. You can also choose a predefined template to create your AI agent quickly. You can
filter the AI agent type to |
4 |
Choose Scripted agent type. |
5 |
Specify the following details: |
6 |
Click Create. You have now successfully created the scripted AI agent which is now available on the Dashboard. On the AI Agent header, you can perform the following tasks:
Also, you can import the AI agents. For more information, see Import AI agent. |
What to do next
Configure scripted AI agent
The following sections guide you on how to configure scripted AI agent for your specific needs:
Update scripted AI agent profile
Before you begin
Create a scripted AI agent.
1 |
On the Dashboard, click the AI agent that you've created. |
2 |
Navigate to the tab and configure the following details: |
3 |
Click Publish to make the AI agent live. |
Update AI engine settings
Scripted AI agents use AI engines (powered by machine learning) to understand and respond to customer queries. Here's a quick overview of the AI engines used:
-
Webex AI Pro 1.0 (with Swiftmatch)—A fast and lightweight training engine that supports multiple languages.
-
Webex AI Pro 1.0 (with RASA)—A top open-source framework for building conversational AI.
-
Webex AI Pro 1.0 (with Mindmeld)—An advanced framework for creating high-quality conversational flows with various natural language understanding (NLU) features.
To change the AI engine for the AI agent:
1 |
On the Dashboard, click the AI agent that you've created. |
2 |
Navigate to tab. |
3 |
Click the icon next to the AI engine. |
4 |
In the Manage AI Engine page, configure the following fields:
|
5 |
Click Update to change the AI engine settings for the AI agent. |
6 |
Click Save changes to update the AI engine settings. |
Configure scripts
Scripts are the building blocks that power your AI agent's understanding and responses. This section describes three key components:
-
Intents capture the various goals or actions users want to accomplish when interacting with your AI agent. Mapping user intents enables the AI agent to recognize and respond appropriately to user requests. To create an intent, see Create an intent.
-
Entities are the specific pieces of information your AI agent needs to extract from user inputs. These include dates, product names, or custom values unique to the customer use case. Entities are important variables that your AI agent must understand to fulfill the user requests effectively. To create an entity, see Create an entity.
-
Responses are your AI agent's carefully crafted replies to user requests. They dictate how your AI agent communicates with the users after understanding their intent and gathering the necessary entities. To create a response, see Create a response.
Together, these components work together to create fluid and purposeful conversations between your agent and its users. For more information, see Understand intents, entities, and responses in AI Agent Studio.
Create an intent
Before you begin
Before creating an intent, ensure you create entities to link to the intent. For more information, see Create an entity.
1 |
On the Dashboard, click the AI Agent that you've created. |
2 |
Navigate to Configurations > Scripts > Intents. |
3 |
Click +Create intent. |
4 |
On the Add a new intent screen, specify the following details: |
5 |
Click the Add to create an Intent. |
6 |
Click Publish to make the AI Agent live. |
Create an entity
1 |
On the Dashboard, click the AI Agent that you've created. |
2 |
Navigate to Configurations > Scripts > Entities. |
3 |
Click+Create entity. |
4 |
On the Create entity window, specify the following fields: |
5 |
Click Save changes. You can use Edit and Delete options in the Actions column to perform related actions. You can edit only the entity name and not the entity type. |
Create a response
1 |
On the Dashboard, click the AI Agent that you've created. |
2 |
Navigate to Configurations > Scripts > Responses. By default, you can use the predefined responses. You can use the Edit icon in each response to modify the response settings. |
3 |
To create a custom response, click +Create response.
Add a new response screen appears.
|
4 |
On the Add a new response screen, enter the new Response name. |
5 |
For the configured language, add a conditional response by clicking Add condition.
|
6 |
Click Create to create the response for the scripted agent. You can also use the code interpreter to configure the responses for various channels. |
Configure agent handover
Before you begin
Create the scripted AI agent.
1 |
On the Dashboard, click the scripted AI agent that you've created. |
2 |
Navigate to and toggle on or off the required settings: |
3 |
Click Save changes to save the handover settings. |
What to do next
Configure language and voice
You can configure multiple languages and language-specific voices for the scripted AI agent to handle customer interactions.
Before you begin
Create the Scripted AI Agent.
1 |
On the AI agent configuration page, navigate to the tab.The default language and voice are set to English. For other supported languages and voices, see the Supported languages and voices article. |
2 |
Click +Add language to add new languages to the AI agent. |
3 |
Choose the desired language and locale from the drop-down list and click Add. |
4 |
Click Add languages. |
5 |
Choose the appropriate voice from the drop-down list. The available voices automatically appear based on the chosen language. |
6 |
Turn on the toggle in the Enabled column to enable the language and voice. |
7 |
Click Make default in Actions to set the language and voice as defaults for the AI agent. You can change the configured language and voice but can’t delete them. However, changing the language and voice may impact the AI agent’s functionality. |
Preview your scripted AI agent
Webex AI Agent Studio allows you to preview your AI agents during and after development. This way, you can test the AI agent's functioning and determine if it generates the desired responses for the respective input queries. You can preview your scripted AI agent in the following ways:
- AI Agent dashboard—Hover over an AI Agent card to view the Preview option for that AI agent. Click Preview to open the AI Agent preview widget.
- AI Agent header—After entering the Edit mode for any AI Agent by clicking on the AI Agent card or the Edit button on the AI Agent card, the Preview option is always visible on the header section.
- Minimized widget—After launching and minimizing a preview, a chat head widget appears at the bottom right of the page. This lets you easily reopen the preview mode.
In addition to this, you can copy the shareable preview link from within an AI agent. On the AI Agent card, click the Ellipses icon in the top right, and click Copy Preview Link. You can share this link with the other users of the AI agent.
Platform preview widget
The preview widget appears at the bottom right of the screen. You can provide utterances (or a sequence of utterances) to see how the AI agent responds, ensuring it performs as expected. The AI agent preview supports multiple languages and can autodetect the language of utterances to respond accordingly. You can also manually select the language in the preview by clicking the language selector and choosing from the list of available options.
You can maximize the preview widget for a better view. Also, you can provide consumer information and initiate multiple rooms to thoroughly test the AI agent.
Publish your scripted AI agent
After configuring and previewing your AI agent, you can publish the agent to make it live.
Before you begin
Create the scripted AI agent.
1 |
On the AI agent configuration page, click Publish. |
2 |
Enter the version name and click Publish. You can view the version details on the History page. For more information, see the History section. |
Common management sections for Scripted AI Agent
The following sections appear on the left panel of the AI Agent configuration page:
Test scripted agent
As AI agents evolve and become more complex, changes to their logic or Natural Language Understanding (NLU) can sometimes have unintended consequences. To ensure optimal performance and identify potential issues, the AI agent platform offers a convenient one-click AI agent testing framework. You can:
- Easily create and run a comprehensive set of test cases.
- Define test messages and expected responses for various scenarios.
- Simulate complex interactions by creating test cases with multiple messages.
Define tests
You can define tests using the following steps:
- Log in to the Webex AI Agent Studio platform.
- On the Dashboard, click the scripted AI agent that you've created.
- Click Testing in the left pane. By default, the Testcases tab appears.
- Select a test case and click Execute selected tests.
Each row in the table represents a test case having the following parameters:
Parameter | Description |
---|---|
Message | A sample message that represents the types of queries and statements you can expect users to send to your AI agent. |
Expected language | The language in which the you interact with the AI agent. |
Expected intent | Specify the intent to be displayed in response to a particular user message. To assist you in finding the most relevant intent, this column features a Smart auto-complete function. As you enter, the system suggests matching intents based on the text entered so far. |
Reset previous context | Click the check box to isolate test cases and run them independently of any existing AI agent context. When enabled, each test case is simulated in a new session, preventing interference from previous interactions, or stored data. |
Include partial matches | Enable this toggle to consider test cases successful even if the expected intents only partially match the actual response. |
Import from CSV | Import test cases from a comma-separated file (CSV) file. In this case, all existing test cases are overwritten. |
Export to CSV | Export test cases to a comma-separated file (CSV) file. |
Test callbacks | Enable this toggle to simulate incoming callbacks and test the flow behavior without requiring actual incoming calls. |
Callback in flow | Click the check box in this column to indicate that an intent must trigger a callback. |
Expected callback template | Specify the template key to activate when the callback occurs. |
Callback timeout (s) | The maximum amount of time (in seconds) the AI agent waits for a callback response before considering the callback as timed out. The system allows a maximum of 20-second timeout. |
Execute tests
On the Execution tab, click Execute selected tests to initiate a sequential execution of all selected test cases.
You can also execute test cases from the Test cases tab.
.To view test cases with specific outcomes, click the desired result (for example,
Passed
, Passed with partial match
,
Failed
, Pending
) in the summary ribbon. This filters the
test case list to display only those matching the selected result.
The session ID
associated with each test case is displayed in the results. This allows you to quickly cross-reference test cases and view transaction details. To perform this, choose the Transaction Details
option in the Actions column.
Execution history
On the History tab, access all executed test cases.
- Click the Download icon from the Actions column to export the executed test data as a CSV file for offline analysis or reporting.
- Review the specific engine and algorithm settings used for each test case execution. This information helps developers optimize the AI agent's performance.
- To view the advanced algorithm configuration settings used for a particular training engine, click the Info icon next to the training engine name. This provides insights into the parameters and settings that influenced the AI agent's behavior during testing.
View agent sessions
The Sessions section provides a comprehensive record of all interactions between AI agents and customers. Each session includes a detailed history of messages exchanged. You can export session data as a CSV file for offline analysis and auditing. Use this data to analyze user interactions, identify areas for improvement, and refine AI agent responses.
It can handle large data sets by displaying results in pages. You can use the Refine Results section to filter and sort sessions based on various criteria. Each row in the table displays essential session details, including:
- Channels—The channel where the interaction occurred (for example, chat, voice).
- Session ID—A unique identifier for the session.
- Consumer ID—The unique identifier of the user.
- Messages—The number of messages exchanged during the session.
- Updated at—The last updated system time.
- Metadata—Additional information about the session.
- Hide test sessions—Select this check box to hide the test sessions and display only the list of live sessions.
- Agent handover happened—Select this check box to filter sessions that we hand over to an agent. If agent handover happens, it displays the headphone icon indicating the handover of the chat to a human agent.
- Error occurred—Select this check box to filter the sessions in which an error occurred.
- Downvoted—Select this check box to filter the downvoted sessions.
Click on a row to access the detailed view of a specific session. Use check boxes to filter sessions based on agent handover, errors, and downvotes. Decrypting sessions requires user-level permission and advanced data protection settings. Click Decrypt content to view the session details.
Session details of a particular session in the Scripted AI Agent for performing actions
The Transaction Info tab in the Scripted AI Agent for performing actions provides a detailed breakdown of a specific interaction, categorizing information into four sections:
Intents Identified section:
- Displays the intents identified for the customer's query.
- Indicates the confidence level associated with each identified intent.
- Lists the slots that are associated with the identified intent. Click the slot to view additional information about its value and how the system extracts it from the user's query.
Entities Identified section lists the entities that the system extracts from the customer's message and associates it with the active consumer intent. These entities represent the key pieces of information that the AI Agent identified within the user's query.
The Algorithm Results section provides insights into the underlying processes that led to the AI Agent's response. Here's a breakdown of the information displayed:
- List of Intents—Shows the identified intents and their corresponding similarity scores.
- Entity List—Displays the entities that were extracted from the user's message.
The Other Info displays:
- Processed Query—Indicates the preprocessed version of the customer's input after the AI Agent's natural language understanding (NLU) pipeline processes it.
- Language detection provider—The provider who offers technology that can automatically identify the language of a given text.
- Language detected—The language detected by technology.
- Agent Handover—Indicates whether an agent handover occurred during the session. Check the Agent Handover by Rules check box if an agent handover was triggered by specific rules.
- Template Key—Indicates the template key associated with the intent that triggered the AI agent's response.
- Response Type—Indicates the type of response generated by the AI agent, such as a code snippet or a conditional response.
- Response Condition—Indicates the specific condition or rule that triggered the AI agent's response.
- NLU AI Engine—Identifies the NLU AI engine used to process the customer's query (for example, RASA, Switchmatch, or Mindmeld).
- Vector model—A way of representing text as numerical vectors.
- Min threshold Scores—Displays the minimum threshold score.
- Partial match score difference—The partial match score difference configured in the Handover and Inference settings. The system determines whether a query is out of scope or requires agent intervention based on these values.
- Debug Logs—Provides a list of debug logs associated with the specific transaction ID. Advanced logs are typically retained for 180 days.
You can also download and view the transaction info in the JSON format using the download option.
The Metadata tab displays:
- NLP Metadata—Review the preprocessing steps applied to the customer's input in the NLP tab.
- Datastore and FinalDF—Access data related to the session in the Datastore and FinalDF tabs for AI Agents.
- Search Functionality—Use the built-in search bar to find specific utterances within a conversation.
View version history and change logs
Whenever you add or modify intents or entities, it's essential to retrain your scripted AI agent to ensure it's the latest version. After each training session, thoroughly test your AI Agent to verify its accuracy and effectiveness. Whenever you publish a Scripted AI agent, a version of the Scripted AI agent is saved and is available in the Version History tab. You can view the various versions of the Scripted AI agent from the Version History tab.
The History page allows you to access the following updates made to your agents:
- Track when you published the version history and the changes made in the form of a note left by developers when publishing.
- See what NLU engine was used for each published version along with NLU engine settings. You can also see the time elapsed to get each version ready to publish.
- Monitor changes to settings, intents, entities, responses, and curation in the Change logs tab.
- Publish, Preview, or Load an older version as draft if needed.
- View Training History—Track when you trained a corpus and the changes made.
- Compare Training Engines—Review the training engines used for different iterations and their corresponding training durations.
- Track changes—Monitor changes to settings, intents, responses, NLP, and curation.
- Revert to previous versions—Easily revert to an older training set if needed.
The History section provides convenient tools for managing your knowledge base:
- Activate intents—Make previously inactive intents Live to include them in the AI Agent's responses.
- Edit intents—Create a new version of an existing intent while preserving the original for reference.
- Preview Performance—Evaluate the AI Agent's performance with a specific knowledge base using the Preview feature.
- Download intents—Export your knowledge base intents as a CSV file for offline analysis or reference.
Change Logs
The Change Logs section provides a detailed record of modifications made to your Scripted AI Agent within the past 35 days. To access Change Logs:
- Navigate to the Dashboard and click the AI agent that you've created.
- Click the History tab to view the AI Agent's history.
- Click the Change Logs tab to see a detailed log of changes:
- Updated At—The date and time the system made the change.
- Updated By—The user who made the change.
- Field—The section of the bot where the modification occurred (for example, Settings, Intents, and Responses).
- Description—Additional details about the change.
-
Use the
Updated by
andField
search options to find specific change log entries. -
The Model History tab displays a maximum of 10 corpora for each AI Agent.
View curated agent details
The system adds messages to the Curation console based on the following criteria:
- Fallback Messages—When the AI Agent fails to understand your message and triggers the fallback intent.
- Default Fallback Intent—If you enable this toggle, we send messages that activate the default fallback intent to the Curation console.
- Downvoted Messages—Messages that users have downvoted during AI Agent previews.
- Agent Handover—Messages that result in a human agent handover due to configured rules.
- From Session—Messages flagged by users as not receiving the desired response from session or room data.
- Low Confidence—Messages with a confidence score falling within the specified low-confidence threshold.
- Partial Match—Messages where the AI Agent couldn't figure out the right intent or response.
Resolve issues
The Issues tab allows you to review and address messages flagged for curation. You can do the following:
- Choose to resolve or ignore issues based on their severity and relevance.
- Examine the original user utterance, the AI Agent's response, and any attached media.
If you enable the Advanced Data Protection in the backend, the system grants the decrypted access at the user level.
To resolve an issue, you can:
-
Link to an existing intent—To connect an issue to an existing intent, select the Link option and search for the desired intent.
-
Create new intent—Use the Add to a New Intent option to create a new intent directly from the Curation Console.
-
Ignore issues—Resolve or ignore issues to remove them from the Curation Console.
- You can't link to default intents (welcome message, fallback message, partial match).
- For a scripted AI agent for performing actions, select the appropriate intent from the drop-down list and tag any relevant entities.
- After making changes, retrain your AI Agent to ensure that it reflects the new knowledge in its responses.
- Resolve or ignore multiple issues simultaneously for efficient management.
The Resolved tab displays all issues addressed by the system. You can view a summary of each resolved issue, including whether we linked it to an existing intent, created a new intent, or ignored it. If you see responses that you don't like that the system didn't catch, you can manually add specific examples to the Curation Console.
To add issues from sessions:
- Identify the Utterance—Locate the utterance that triggered the incorrect response.
- Check Curation Status—If the issue isn't already in the Curation Console, the system displays the
Curation Status
toggle. - Toggle the Flag—Enable the
Curation Status
toggle to add the utterance to the Curation Console for review and resolution.
If the issue is already in the Curation Console, the toggle's appearance changes to show its status.
View your Scripted AI performance using Analytics
The Analytics section provides a graphical representation of key metrics to evaluate the AI agent performance and effectiveness. The key metrics are divided into four sections represented as tabs, namely Overview, Responses, Training, and Curation.
On visiting the analytics screen, developers can select the AI agent they want to see the analytics for. They can customize the analytics view by choosing the channel, date range, and data granularity. By default, the system displays analytics data for the last month for all channels, with each day as a data point.
Overview
The overview contains key metrics and graphs that provide a snapshot of overall AI agent usage and performance to the developers.
- From the Dashboard, choose the AI agent that you've created.
- On the left navigation pane, click Analytics. An overview of the AI agent performance appears in both tabular format and graphical representation.
Sessions and messages
The first section in the overview displays the following statistics about sessions and messages for the AI agent:
- The count of the total sessions and the sessions that the AI agent handles without human intervention.
- Total agent handovers, which is a count of the number of sessions handed over to human agents.
- Daily average sessions
- Total messages (human and AI agent messages) and how many of those messages came from users.
- Daily average messages
The system follows this with a graphical representation of sessions (stacked column representing sessions handled by the AI agent and sessions handed over) and the total responses sent out by the AI agent.
Users
The second section in the overview contains stats about users for the AI Agent. It provides a count of total users and information about average sessions per user and daily average users. This is followed by a graph displaying new and returning users for each unit depending on the selected granularity.
Performance
The third section provides statistics about the AI agent’s responses to users. Here one can see the total responses sent out by the AI agent and the split up between responses where the AI agent:
- Identified the user’s intent.
- Responded with a fallback message.
- Responded with a partial match message.
- Informed the user of an agent handover.
The same is aggregated in a pie chart and an area graph provides information based on selected granularity.
Training
The training section represents of the ‘health’ of an AI Agent corpus. It’s recommended that developers configure 20+ training utterances for each intent in their AI Agents. This section displays all intents as rectangles, with color and size indicating the amount of training data. The closer an intent is to white color, the more training data it needs for your AI Agent’s accuracy to improve.
Responses
This section gives the developers a detailed view of what the users are asking about and how often they are asking it. It graphically shows the most popular intents for AI Agents for answering questions and response templates for AI Agents for performing actions.
Curation
This section visually summarizes the number of curation issues that arise each day and the number that AI agents resolve.
Use AI agents for customer interactions
This section outlines how to integrate AI agents with both voice and digital channels to manage customer conversations.
Use AI agents for voice and digital interactions
After you've created and configured your autonomous or scripted AI agents in the Webex AI Agent Studio platform, the next step is to integrate them with the voice and digital channels. This integration allows the AI agents to handle both voice-based and digital conversations with your customers, providing a seamless and interactive user experience.
For more information, see Use AI agents for voice and digital interactions article.
Manage custom reports for AI agents
You can use global variables to generate custom reports and analyze calls routed to the AI agent.
Currently, out-of-the-box reports for the AI agent are not available in Analyzer.
Create a global variable
1 |
Sign in to Control Hub. |
2 |
From the Contact Center navigation pane, choose . |
3 |
Click Create a new global variable and provide the name and description for the variable. Create a variable with the name CustomAIAgentInteractionOutcome. Choose String as the variable type. |
4 |
Toggle Make Reportable on to display the variable in Analyzer for reporting purposes. |
5 |
Click Save. |
Add the global variable to the flow
The following instructions are also available within the linked sample flow import.
1 |
Sign in to your customer organization using Control Hub. |
2 |
Navigate to . The Flows page appears. |
3 |
Click the Go to Flow Designer icon beside the flow. The Flow Designer window appears. |
4 |
In the Global Flow Properties pane, scroll down to section. |
5 |
In the Global Variables section, click Add Global Variables. Add the global variable CustomAIAgentInteractionOutcome to your flow. |
6 |
Use the Set Variable activity to assign the value ABANDONED to the variable CustomAIAgentInteractionOutcome. |
7 |
Configure your Virtual Agent V2 activity in the flow. |
8 |
Connect the Handled outcome of the Virtual Agent V2 Activity and use the Set Variable activity to assign the value HANDLED to the variable CustomAIAgentInteractionOutcome. |
9 |
Connect the Escalated outcome of the Virtual Agent V2 activity and use the Set Variable activity to assign the value ESCALATED to the variable CustomAIAgentInteractionOutcome. |
10 |
Connect the errored path of the Virtual Agent V2 activity and use the Set Variable activity to assign the value ERRORED to the variable CustomAIAgentInteractionOutcome. |
11 |
Complete the rest of the flows based on your business logic and publish them. Any calls going through this flow will have the value of the variable CustomAIAgentInteractionOutcome set to Abandoned, Handled, Escalated or Errored, depending on the path the call takes. |
Create custom visualizations
You can create custom reports for AI agent call records and AI agent outcome distribution in Analyzer.
Create AI Agent Call Records visualization
1 |
Download the AI_Agent_ Call_Records file from https://github.com/WebexSamples/webex-contact-center-api-samples/pull/275. |
2 |
Log in to Analyzer. |
3 |
On the Home page, click the Visualization icon in Analyzer. |
4 |
Click Import. |
5 |
Click Browse to select the file (.json format) to be imported. |
6 |
Click Import to import the AI_Agent_ Call_Records file. |
7 |
Click Edit to modify the imported Visualization. |
8 |
Click on Edit Filters for CustomAIAgentInteractionOutcome. |
9 |
Click the is in radio button and add the values ABANDONED, ESCALATED, ERRORED, HANDLED. |
10 |
Save and initiate a few test calls. |
11 |
Run the visualization to view the results. |
Create AI Agent Outcome Distribution visualization
1 |
Download the AI_Agent_ Outcome_Distribution.json file from https://github.com/WebexSamples/webex-contact-center-api-samples/pull/275. |
2 |
Log in to Analyzer. |
3 |
On the home page, click the Visualization icon in Analyzer. |
4 |
Click Import. |
5 |
Click Browse to choose the file (in JSON format) to be imported. |
6 |
Click Import to import the AI_Agent_ Outcome_Distribution.json file. |
7 |
Click Edit to modify the imported Visualization. |
8 |
Click on Edit Filters for CustomAIAgentInteractionOutcome. |
9 |
Click the is in radio button and add the values ABANDONED, ESCALATED, ERRORED, HANDLED. |
10 |
Save and initiate a few test calls. |
11 |
Run the visualization to view the results. |
Understand AI Compliance
This sections helps you understand AI development, data privacy, security, and safety.
AI development, data privacy, security, and safety
For every AI-powered feature, we undergo an AI Impact Assessment against our Responsible AI principles, and adhere to the Responsible AI Framework, in addition to existing Security, Privacy, and Human Rights by Design processes.
Privacy and SecurityWe don’t retain customer input data after the inference process, and the third-party model provider, Microsoft, doesn’t access, monitor, or store Cisco customer data. For more detail on feature-specific data retention policies, see Webex AI Agent AI transparency technical note in the Cisco Trust Portal.
Data Sources for Training and EvaluationOur third-party model provider, Microsoft doesn't use customer content to improve Azure OpenAI models and doesn’t store or retain Cisco customer data in Azure infrastructure.
Safety and Ethical ConsiderationsAll generative AI features are prone to errors, so we prioritize content safety for AI features by opting in to Content filtering, provided by Azure OpenAI.
Model Evaluation and PerformanceWe prioritize the performance and accuracy of AI Assistant by involving humans in the review, testing, and quality assurance of the underlying model.