Designing CXone Auto-Summary and Disposition Mapping with Enlighten AI

How-To-Articles
1

Designing CXone Auto-Summary and Disposition Mapping with Enlighten AI

What This Guide Covers

This guide details the implementation of NICE Enlighten AI to automate the post-interaction lifecycle. You will learn how to configure the AI engine to generate objective interaction summaries and automatically map detected customer intents to ACD disposition codes, significantly reducing After-Call Work (ACW) and eliminating agent subjectivity in CRM reporting.

Prerequisites, Roles & Licensing

  • Licensing: NICE CXone Enlighten AI (Auto-Summary / Copilot) license tier.
  • Permissions:
    • Security > Roles > NICE CXone Central > Enlighten > View/Edit
    • Security > Roles > NICE CXone Central > ACD > Dispositions > View/Edit
  • Software: Agents must be using the latest version of the MAX (Multi-Channel Agent Experience) desktop to view and validate AI-generated summaries.

The Implementation Deep-Dive

1. The Enlighten AI Engine: Intent Detection vs. Transcription

It is a common misconception that auto-summarization is just “summarized transcription.” Architecturally, Enlighten AI uses two distinct layers.

The Implementation:

  1. Transcription Layer: Converts the stereo audio (Agent/Caller) into text.
  2. Intent Layer: Maps the text to pre-trained models (e.g., “Billing Dispute,” “Address Change,” “Retention Request”).

The Trap:
Attempting to build “Universal Summaries.” If you do not define your business-specific intents in the Enlighten portal, the AI will provide generic summaries like “The customer called about a problem and the agent helped them.” This provides zero analytical value.
The Solution: Use Custom Topic Training. Identify your top 10 call drivers and create Enlighten “Topics” with specific keyword clusters. The AI will then prioritize these topics in the summary generation.

2. Mapping AI Intents to ACD Disposition Codes

The most powerful feature of Enlighten is its ability to pre-select the disposition code for the agent.

The Implementation:
In the Enlighten configuration portal, create a mapping between an Enlighten “Topic” and a CXone “Disposition Code.” When the call ends, the AI evaluates the dominant topic and automatically populates the {DispositionID} in the agent’s ACW window.

The Trap:
Mapping multiple disparate intents to a single “General” disposition. While it simplifies the agent’s life, it destroys the granularity of your reporting.
The Architectural Reasoning: Implement a “Confirmation Loop.” Instead of forcing the AI’s choice, configure the MAX desktop to “Suggest” the disposition. The agent sees a highlighted code based on the AI’s confidence score (e.g., >85%). This maintains data integrity while still providing the speed benefit of automation.

3. Objective Summarization vs. Agent Subjectivity

Agents often summarize calls through the lens of their own experience (e.g., “Customer was rude”), whereas Enlighten AI summarizes based on factual milestones (e.g., “Customer requested a refund for invoice #123; agent denied based on policy”).

The Implementation:
Configure the Summary Delivery Target. You can choose to push the summary to:

  1. The MAX ACW Window (for agent validation).
  2. The CXone Interaction Metadata (for immediate searchability).
  3. An external CRM via a Webhook (for a permanent record in Salesforce/Dynamics).

The Trap:
Pushing the summary to the CRM before the agent has a chance to edit it. If the AI misinterprets a complex multi-topic call, an incorrect summary is committed to the customer’s permanent record.
The Solution: Implement a 30-second “Review Buffer” in the MAX desktop. If the agent does not edit the summary within 30 seconds of the call ending, the AI summary is automatically committed.

Validation, Edge Cases & Troubleshooting

Edge Case 1: The “Multi-Topic” Conflict

The Condition: A customer calls to pay a bill but ends up complaining about a previous service failure. The AI summarizes only the bill payment.
The Root Cause: “Dominant Intent” bias. Most AI models prioritize the first detected intent or the intent that occupied the most talk time.
The Solution: Enable “Secondary Intent Detection” in the Enlighten settings. This allows the AI to output a primary and secondary topic, ensuring that the “Service Failure” complaint is not lost in the reporting.

Edge Case 2: Low-Confidence Hallucinations

The Condition: The AI summary contains information that was never discussed (e.g., a specific date or dollar amount that is incorrect).
The Root Cause: Low confidence in the transcription due to background noise or poor audio quality (VOIP jitter).
The Solution: Set a Confidence Threshold. In your integration middleware, only accept summaries with a confidence score above 75%. For scores below this threshold, revert the agent to manual summarization to prevent “Data Poisoning” in your CRM.

Official References