From Bright Pattern Documentation
Jump to: navigation, search
• 5.19


Using a Bot to Assist Agents in Chats

You can set up chat services to assist agents with an Amazon Lex Bot or IBM Watson Assistant. When configured, the chat transcript is analyzed by the bot, and bot responses are presented to the agent as reply options that can be selected with a click. The agent is still free to compose their own reply. Optionally, customer sentiment can be analyzed using Natural Language Understanding.


The interface displays (1) multiple bot suggestions and (2) NLU-driven customer sentiment analysis


Prerequisites


To assist agents in voice services, refer to the related guide Using Real-Time Voice Transcripts for Bot/Chat Suggestions.

See Integration with Bots to learn more about chatbot integrations in general.

Configuration Steps

  1. In the Properties tab of the service, choose the default bot integration account for suggestions with the Bot/Chat suggestions engine property. This can be overridden in the scenario if needed, see below.

  2. (Optional) In the Properties tab, indicate the desired NLU account for sentiment analysis with the Natural Language Understanding property. See Natural Language Understanding to learn more.

Configuration Details

Selecting the Bot Account

The Bot/Chat suggestions engine integration account that provides chat suggestions can be selected in the service properties, or set in the scenario.

For more information, see the Selecting the Bot / Chat Suggestions Engine Integration Account section in the Using Real-Time Voice Transcripts for Bot/Chat Suggestions tutorial.


Analyze Agent Responses when Making Suggestions

By default, only customer messages are analyzed by the suggestions bot. However, bot suggestion relevance can sometimes be improved by including agent messages for analysis as well. To include both customer and agent messages for bot analysis, follow the instructions Configure Suggestions Bot to Analyze Agent Responses in the Using Real-Time Voice Transcripts for Bot/Chat Suggestions tutorial.


Formatting Bot Suggestions

Suggestions for agents are derived from the response string returned by the bot account. Suggestions may contain formatting and URL links that provide additional information to the agent. Response configuration depends on the bot account provider:

  • For Amazon Lex, the content of the intent property "Response" is returned as a suggestion.
  • For IBM Watson, the content of the dialog property "Respond With" is returned as a suggestion


Info.40x40.png Configuring multiple responses for a single intent (Lex) or dialog node (Watson) allows the bot to offer agents several distinct reply suggestions when triggered.


For example, you could configure multiple responses to suggest different troubleshooting options when the bot recognizes a technical problem in the conversation.



There are two ways to format the response in the bot account: as a text string, or as a string that can be parsed into a JSON object.


Response as Text String

To configure a suggestion that does not require text formatting (bold, italics, clickable URLs, etc.) configure the bot response with the exact text to display as a suggestion.


Response as a JSON Object

To configure a suggestion with formatted text, you will need to set the bot response as a string that can be parsed into a JSON object.

For a detailed explanation, see Formatting Bot Suggestions in the Using Real-Time Voice Transcripts for Bot/Chat Suggestions tutorial.

< Previous | Next >