This page shows you how to set up Agent Assist for Contact Center AI Platform (CCAI Platform). Agent Assist provides suggestions to your human agents when they're in a conversation with an end-user. Suggestions are based on your own uploaded data and therefore can be tailored to your specific business needs. Agent Assist is not available in the email channel.
Service account requirements
Before you can set up Agent Assist, you need to create a Google Cloud Service Account and key in the project where you created your Agent Assist conversation profile. In addition, you must grant the following IAM roles to your service account:
Add an Agent Assist platform
To add an Agent Assist platform, follow these steps:
In the CCAI Platform portal, click Settings > Developer Settings. If you don't see the Settings menu, click Menu.
Go to the Agent Assist Platform pane, and then click Add platform. The Add an Agent Assist platform dialog appears.
In the Name field, enter a name for your platform.
In the Service field, select CCAI Conversation Profile, and then click Create.
Click Upload key to upload your JSON service account key. For more information, see Service account requirements.
Click Close to close the dialog.
Configure a queue for Agent Assist
After you add your Agent Assist platform, you can assign the platform to a queue where you want to use Agent Assist capabilities. Agent Assist can only be assigned to leaf queues.
To configure a queue for Agent Assist, follow these steps:
In the CCAI Platform portal, click Settings > Queue. If you don't see the Settings menu, click Menu.
Depending on the channel you want, click Edit / View in one of the following panes: IVR (Interactive Voice Response), Mobile, or Web.
Click the queue that you want to edit.
In the Settings pane, under Agent Assist, click Configure. The Agent Assist Profile Configuration pane appears.
In the Platform Name field, select a platform that you added previously. The Profile Name field appears.
In the Profile Name field, select a profile.
Click Save. The Settings pane appears.
Under Agent Assist, click the Enable Agent Assist toggle to the on position.
Agent Assist for calls
The following call capabilities are available with Agent Assist. To enable these capabilities, select them when you turn on and configure Agent Assist for calls.
Knowledge assist: provides knowledge base suggestions to human agents during a call. An agent can examine suggestions and decide which to read or share with the end-user.
The knowledge base for knowledge assist is set up in the Google Cloud Agent Assist console. The following tutorial walks you through the steps to create a knowledge base: Agent Assist - Knowledge Base.
Generative knowledge assist: provides generative AI knowledge-base suggestions to human agents during a call.
Sentiment analysis: analyzes messages in real time from both the human agent and the end-user to determine emotional intent. The score is visible in the agent adapter.
Real-time call transcription: transcribes calls in real time for agents to reference during the call or for analysis after the call
Generative AI session summarization: displays an auto-generated summary of the conversation when the call enters wrap-up. This summary provides a brief overview of the call, including key discussion points and resolutions. The agent can optionally add their own notes to the summary. To keep the summary, the agent clicks Insert summary before saving the wrap-up screen. A saved summary is recorded in your CRM app or in external storage, depending on your configuration.
Turn on and configure Agent Assist for calls
This procedure shows you how to turn on Agent Assist for calls globally, and to select the Agent Assist call capabilities that you want. In addition, you need to set up Agent Assist, which includes adding an Agent Assist platform, and configuring the queues where you intend to use Agent Assist.
To turn on Agent Assist for calls, follow these steps:
In the CCAI Platform portal, click Settings > Call. If you don't see the Settings menu, click Menu.
Go to the Agent Assist pane.
Select the Agent Assist capabilities that you want to turn on, and then click Save Agent Assist. For information about Agent Assist call capabilities, see Agent Assist for calls.
Agent experience
The following sections explain the agent experience for each Agent Assist capability.
Knowledge assist
This feature follows a conversation between a live agent and a end-user and provides the live agent with relevant document suggestions. Agents are alerted to knowledge assist document suggestions when the Agent Assist icon in the agent adapter has a green dot next to it.
Click the Agent Assist icon to view the suggestions in the Knowledge tab in the agent adapter.
Click the External Link icon next to a suggested article to share a document with the end-user.
Real-Time sentiment analysis
Sentiment Analysis evaluates the conversation between a human agent and end-user to determine emotional intent. The analysis is generated in real-time to display the sentiment of the present session, or as a one-time score upon Virtual Agent escalation. Sentiment Analysis is available for both Voice and Chat.
For Voice: The Sentiment feature panel displays on the Agent Adapter during a live session.
Sentiment Analysis breakdown
The Real-Time Sentiment Analysis displays 3 scores: last, session, and previous leg.
A (--) displays when no score is available.
| last | sentiment score for the last consumer response (updates in real-time) |
| session | current average sentiment score (updates in real-time) |
| prev. leg | average sentiment score of the previous segment |
There are three sentiments types (neutral, happy, and negative) in the Real-Time Sentiment Analysis and scores displayed fall into the following ranges:
40-75 = Neutral
Over 75 = Happy
Below 30 = Negative
During the session, the Real-Time Sentiment Analysis scores are sent the to the Agent Adapter. After the session, the session's average score will be posted in the CRM ticket.
Session summarization
Session summarization automatically creates a transcript of the agent and end user conversation during a call, which is displayed in real-time under the Transcript tab in the agent adapter. When the call goes into wrap-up, the session transcript becomes available as an attachment in the notes area of the agent adapter.
Agents can also add notes related to the call in the disposition notes field. Agents can further improve their notes by using the automatically generated session summary feature, which provides a brief overview of the call, including key points discussed and any resolutions or solutions agreed upon.
After the summary is added, the agent can save the disposition summary, which will be shared with the CRM based on the preferred company configuration. Overall, this feature provides agents with a comprehensive and efficient way to manage their conversations with customers, ensuring that all important details are recorded and accessible.
When a call is initiated, the Transcript tab will display the transcript of the conversation in real-time.
When the chat goes into wrap-up, the session transcript will be available as an attachment in the notes area of the agent adapter.
In the disposition notes field, you can add any additional notes related to the call, such as the resolution or solution.
If you'd like to add the Automatically Generated Summary to the disposition summary, click the Insert Summary button.
Review the summary and make any necessary edits.
Save the disposition summary. Based on your company's CRM and external storage configuration, the disposition summary will be shared with the CRM as configured.
Agent Assist for chat
The following chat capabilities are available with Agent Assist. To enable these capabilities, select them when you turn on and configure Agent Assist for chat.
Smart compose: monitors a conversation in real time and provides suggestions to human agents that auto-complete their responses to end-users
Smart reply: provides chat responses to your human agents in real time during a conversation
Generative knowledge assist: provides generative AI knowledge-base suggestions to human agents during a chat
Sentiment analysis: analyzes messages in real time from both the human agent and the end-user to determine emotional intent. The score is visible in the agent adapter.
Generative AI session summarization: displays an auto-generated summary of the conversation when the chat enters wrap-up. This summary provides a brief overview of the chat, including key discussion points and resolutions. The agent can optionally add their own notes to the summary. To keep the summary, the agent clicks Insert summary before saving the wrap-up screen. A saved summary is recorded in your CRM app or in external storage, depending on your configuration.
Spelling and grammar correction: provides spelling and grammar suggestions for messages that agents type into the message field of the chat adapter.
Turn on and configure Agent Assist for chat
This procedure shows you how to turn on Agent Assist for chat globally, and to select the Agent Assist chat capabilities that you want. In addition, you need to set up Agent Assist, which includes adding an Agent Assist platform, and configuring the queues where you intend to use Agent Assist.
To turn on Agent Assist for chat, follow these steps:
In the CCAI Platform portal, click Settings > Chat. If you don't see the Settings menu, click Menu.
Go to the Agent Assist pane.
Select the Agent Assist capabilities that you want to turn on, and then click Save Agent Assist. For information about Agent Assist chat capabilities, see Agent Assist for chat.
Agent experience
The following sections explain the agent experience for each Agent Assist capability.
Smart Compose
Smart Compose auto-completes an agent's chat responses using context from the ongoing conversation. This feature provides suggestions at the end of the words being typed to help finish the sentence.
How it works: Suggestions to complete the words or sentences appear after the typed message in lighter font color.
A tab key icon is displayed at the end of the autocomplete suggestion. Click the tab key on the physical keyboard to place the suggested autocomplete text content in the agent chat entry field.
Smart Reply
Smart Reply suggests chat responses to the agent using context from the ongoing conversation. While Smart Compose suggests characters to complete a partially-typed response, Smart Reply suggests entire responses in bubbles above the text entry box.
How it works: Suggestions appear above the text entry field. Select the suggestion that you want and paste it into the field and send the reply to the end-user.
Knowledge Assist
This feature follows a conversation between a human agent and an end-user and provides the human agent with relevant document suggestions.
How it works: The Knowledge Assist button is located in the top right hand corner of the Agent Adapter. A pink circle indicates that articles are available. Click the icon to show the articles. The icon expands to show a list of articles.
Click the Share button next to the article to send it to the end-user.
Real-Time Sentiment Analysis
Sentiment Analysis evaluates the conversation between a human agent an end-user to determine emotional intent. The analysis is generated in real-time to display the sentiment of the present session, or as a one-time score upon Virtual Agent escalation. Sentiment Analysis is available for both Voice and Chat.
Sentiment Feature Panel
For Chat: The Sentiment feature panel will be displayed in an open state by default for all new chats. Click the X or the Sentiment Stars Icon to open or hide the panel.
Sentiment Analysis breakdown
The Real-Time Sentiment Analysis displays 3 scores: last, session, and previous leg.
A (--) displays when no score is available.
| Score | Description |
|---|---|
| last | Sentiment score for the last consumer response. Updates in real time. |
| session | Current average sentiment score. Updates in real time. |
| prev. leg | Average sentiment score of the previous segment. |
There are three sentiments types (neutral, happy, and negative) in the Real-Time Sentiment Analysis and scores displayed fall into the following ranges:
- 40-75 = Neutral
- Over 75 = Happy
- Below 30 = Negative
During the session, the Real-Time Sentiment Analysis scores are sent the to the Agent Adapter. After the session, the session's average score will be posted in the CRM ticket.