Human in the Loop

Human-in-the-loop is a first-of-its kind capability that allows a human agent to guide GenerativeAgent in assisting a customer. GenerativeAgent may request human help in situations where it lacks the necessary API access, knowledge sources, or requires human approval to complete an action.

GenerativeAgent raises a ticket requesting specific help through your existing digital support tool. Available agents within your organization are part of dedicated human-in-the-loop queues where they receive and respond to tickets from GenerativeAgent, thereby resolving the customer issue.

The Human-in-the-loop Agents (HILA) within your organization wait in a queue and are directed to specific scenarios where they take on the action of helping in a customer’s issue and give back the conversation to the GenerativeAgent. HILAs can also transfer the conversation to a Live Agent so they take over the task from GenerativeAgent.

​

When should GenerativeAgent consult a human

The Human-in-the-loop capability can be invoked in the task instructions for the GenerativeAgent. You can specify scenarios or actions that the GenerativeAgent cannot perform automatically and require human intervention for information or confirmation. This is similar to actions a human agent cannot complete without supervisor approval.

Recommended scenarios for human assistance include:

• Insufficient permissions: When the GenerativeAgent should not act out in its own without HILA approval.
• No API to call: Whenever there is no API to call to retrieve specific customer information.
• No Knowledge Base information: Whenever the question or issue provided by the costumer has no content in the Knowledge Base source that GenerativeAgent can use.

​

HILAs

The primary function of the human-in-the-loop is to support and unblock the GenerativeAgent. These supervisors handle tasks requiring approvals or a deep understanding of the issues. Tier 1 agents can address simpler queries. When accepting a ticket from the GenerativeAgent in your digital support tool, agents access an embedded Human-in-the-loop UI.

The actions HILAs are set to do include:

• Ticket assignments
• Response edits/decisions
• Unlock GenerativeAgent
• Escalation to live agents

Human-in-the-loop agents assist the GenerativeAgent without directly interacting with customers, differing from live agents.

The key benefit is that a single agent can manage multiple customer conversations simultaneously without engaging in calls or chats.

If the Human-in-the-loop agent determines that a live agent would better serve the customer, they can use the Transfer option in the UI to hand off the conversation from the GenerativeAgent to a live agent.

When does GenerativeAgent transfer to a live agent?

The following scenarios are recommended for transferring a conversation to a live agent:

• A human-in-the-loop instructs GenerativeAgent to do so
• There are no Human-in-the-loop agents available. This is managed automatically and does not require explicit instructions.
• The customer explicitly requests it (if configured).

​

Agent UI

Enabling human-in-the-loop capability

Human-in-the-loop agents operate from the existing Agent Desk. To enable the Human-in-the-loop UI and task functions in GenerativeAgent, please contact your Implementation Manager.

The Human-in-the-loop agent UI is the primary application where agents can interact with the GenerativeAgent. Through this interface, they can:

• Respond to the GenerativeAgent
• Transfer conversations to live agents
• View the interaction thread history
• Access relevant customer information and summarized conversation context

This section provides an overview of important features available:

• Transfer: Allows the agent to transfer the conversation from the GenerativeAgent to a live agent.
• Ticket Assignment Timer: Tracks the time elapsed since the ticket was assigned to the agent.
• Prompt: Indicates the specific assistance the GenerativeAgent needs to unblock the customer. This is generated by the GenerativeAgent itself.
• Response: The Human-in-the-loop agent can respond to the GenerativeAgent through an open text field or structured options, depending on the configuration.
• Send: After selecting a response, the agent can click ‘Send’ to submit the response and close the ticket simultaneously.
• Context: Provides a summarized context of the conversation between the GenerativeAgent and the customer.
• Transcript: Displays the complete customer interaction thread prior to the ticket being raised.
• Customer: Shows the customer’s details, including company and specific account information for authenticated customers.

​

Next Steps

After seting-up Human-in-the-Lopp, you are ready to speed up customer replies and solve their inquiries.

You may find one of the following sections helpful in advancing your integration: