Implementing Automated Post-Call Summarization with Genesys Cloud Copilot

How-To-Articles
1

Implementing Automated Post-Call Summarization with Genesys Cloud Copilot

What This Guide Covers

  • You will architect a Generative AI-driven workflow that automatically summarizes customer conversations and extracts key action items using Genesys Cloud Copilot.
  • You will implement the “Auto-Summarization” feature to replace manual After Call Work (ACW), ensuring consistent, high-quality documentation for every interaction.
  • When complete, your agents will save 45-90 seconds per interaction on manual typing, significantly reducing Average Handle Time (AHT) while improving the reliability of your CRM data.

Prerequisites, Roles & Licensing

  • Licensing: Genesys Cloud CX 1, 2, or 3 with the AI Experience add-on (or included in “AI-Ready” bundles).
  • Permissions:
    • Copilot > Summary > View/Edit
    • Conversation > Communication > View (to access transcripts)
    • Routing > Queue > Edit (to enable on specific queues)
  • OAuth Scopes: copilot, conversations.
  • Foundational Requirement: Genesys Cloud Transcription must be enabled for the queues where Copilot will operate.

The Implementation Deep-Dive

1. The Evolution of ACW: From Subjective to Objective

Manual agent notes are notoriously inconsistent-one agent writes “Customer called about bill,” while another writes “Billing dispute resolved, credit applied.” Genesys Cloud Copilot uses Large Language Models (LLMs) to provide an objective, structured summary based on the actual transcript.

The Strategy:

  • Capture: The system transcribes the dual-channel audio (Agent + Customer).
  • Synthesize: The LLM identifies the Reason for Call, the Resolution, and the Sentiment.
  • Inject: The summary is presented to the agent during wrap-up for quick review and then saved to the interaction details.

2. Configuring Copilot and LLM Settings

You must configure the “Style” of the summary to match your business requirements.

The Step:

  1. Navigate to Admin > Copilot > Settings.
  2. Enable Auto-Summarization.
  3. The Critical Part: Select the Summary Length (Short, Medium, Long) and Format (Bullet points vs. Paragraph).
  4. Define Custom Action Item detection. You can instruct Copilot to specifically look for phrases like “I will call you back” or “Send me an email.”
  5. Architectural Reasoning: Bulleted summaries are generally superior for technical support, as they allow subsequent agents to “scan” previous interaction histories much faster than reading a paragraph.

3. Enabling Summarization on the Queue

Copilot is not “on by default” for the entire organization; it is enabled at the queue level.

The Step:

  1. Go to Admin > Routing > Queues.
  2. Select your target queue and click the Copilot tab.
  3. Toggle Enable Summarization to “On.”
  4. The Trap: Ensure that Dual-Channel Recording is enabled for the queue. If you are only recording a mono-mixed stream, the AI cannot distinguish between the agent’s voice and the customer’s voice, leading to “muddled” summaries where the AI attributes the agent’s words to the customer.

4. Customizing the Agent Experience

The agent should not be a “passive observer.” They must have the ability to edit the AI-generated summary before it becomes the permanent record.

The Step:

  1. In the Wrap-Up screen configuration, ensure the “Notes” field is populated by Copilot.
  2. Training: Teach agents to “Audit” the summary. While Copilot is ~95% accurate, it may miss specific industry jargon or nuanced sarcasm.
  3. Architectural Reasoning: This “Human-in-the-loop” model ensures data integrity while still providing 90% of the labor savings of full automation.

[THE TRAP]
The most common failure in Copilot deployments is “Silent Failures” due to poor audio quality. If your agents are using low-quality analog headsets or calling over high-jitter VoIP lines, the transcript quality will be poor. Generative AI is “Garbage In, Garbage Out”-if the transcription is only 60% accurate, the summary will be hallucinated or nonsensical. Always verify your Transcription Accuracy before blaming the Copilot model.

Validation, Edge Cases & Troubleshooting

Edge Case 1: Multi-Topic Conversations

  • The Failure: A customer calls about a technical issue, then switches to a billing complaint, then asks about a new product. The summary only mentions the first topic.
  • The Root Cause: The LLM “Prompt” is too restrictive or the conversation was too long for the context window.
  • The Solution: Use the “Long” summary format in Copilot settings. This instructs the model to capture multiple sub-intents and transitions in the conversation.

Edge Case 2: PII Redaction in Summaries

  • The Failure: The customer reads their credit card number, the transcription captures it, and the AI summary includes it in the “Resolution” notes.
  • The Root Cause: Transcription redaction was not configured or failed.
  • The Solution: Enable Native PII Redaction in the Genesys Cloud Telephony settings. This ensures the sensitive data is “scrubbed” from the transcript before it is ever sent to the Copilot LLM for summarization.

Edge Case 3: The “No-Talk” Interaction

  • The Failure: A call connects, there is 2 minutes of silence or music-on-hold, and then it disconnects. Copilot generates a weird summary like “The customer listened to music.”
  • The Root Cause: The AI is trying to find meaning where there is none.
  • The Solution: Implement a Minimum Duration rule in your reporting. If the conversation has fewer than 20 transcribed words, ignore the Copilot summary and mark the interaction as Empty_Interaction.

Official References