Designing CXone Digital (DFO) Intent-Based Routing using Enlighten AI

How-To-Articles
1

Designing CXone Digital (DFO) Intent-Based Routing using Enlighten AI

What This Guide Covers

  • Architecting an advanced intent-based routing strategy for digital channels (Email, SMS, Web Chat) within NICE CXone Digital First Omnichannel (DFO).
  • Utilizing Enlighten AI to automatically classify customer intent based on the initial message payload and dynamically routing interactions to specialized skill groups, skipping traditional rigid IVR/menu selections.
  • The end result is a frictionless customer experience where the underlying AI understands why the customer is reaching out before an agent is even assigned, reducing handle time and eliminating unnecessary transfers.

Prerequisites, Roles & Licensing

  • Licensing: NICE CXone with Digital First Omnichannel (DFO) and Enlighten AI Routing/Classification enabled.
  • Permissions: ACD > DFO > Administration, ACD > Routing > Skills, Enlighten AI > Model Management.
  • Infrastructure: Configured Digital Channels (e.g., an active inbound Email or Web Chat point of contact).

The Implementation Deep-Dive

1. The Limitation of Traditional Digital Routing

In traditional digital routing, an email arrives at support@company.com, and it is blindly routed to the “Support” skill queue. The first available agent opens it, reads it, realizes it’s a billing dispute, and manually transfers it to the “Billing” queue.

The Trap:
This manual triage creates massive inefficiencies. The customer waits in the Support queue just to be told they are in the wrong place, and then waits again in the Billing queue. Furthermore, manual transfers skew your SLA metrics and increase the Average Handle Time (AHT) artificially across departments.

2. Enabling Enlighten AI Intent Classification

To eliminate manual triage, you must intercept the digital interaction before it hits a queue and run it through Enlighten AI.

Implementation Steps:

  1. Model Selection: In the CXone Admin portal, navigate to the Enlighten AI section. Select the pre-trained intent model that matches your vertical (e.g., Financial Services or Retail), or deploy a custom model trained on your historical chat transcripts.
  2. DFO Channel Configuration: Navigate to ACD > DFO > Channels. Select your target channel (e.g., Inbound Email).
  3. Routing Configuration: Under the Routing settings for the channel, locate the Automation & AI or Pre-Routing Logic section. Enable Enlighten AI Classification.
  4. Configure the system to extract the Body of the email or the first Message of the Web Chat and pass it to the Enlighten model as the evaluation text.

3. Architecting the DFO Routing Rules

Once Enlighten evaluates the text, it assigns an Intent Label (e.g., Billing_Dispute, Password_Reset, Where_Is_My_Order) and a Confidence Score (e.g., 0.92). You must build routing rules to act on this data.

Architectural Reasoning:
Do not blindly trust the AI. If the Confidence Score is too low, the AI is guessing. You must build a fallback mechanism.

Implementation Steps:

  1. Navigate to ACD > DFO > Routing Rules.
  2. Create a new Rule named Enlighten_Billing_Escalation.
  3. The Condition: Set the condition to evaluate the Custom Data field populated by Enlighten:
    • Intent Equals Billing_Dispute AND ConfidenceScore > 0.85.
  4. The Action: Set the action to Assign to Skill: Digital_Billing_Tier2 and increase the priority of the interaction by +10.
  5. The Fallback Rule: Create a second rule named Enlighten_Low_Confidence_Fallback.
    • Condition: ConfidenceScore <= 0.85.
    • Action: Assign to Skill: Digital_Triage_General.
  6. Ensure your rules are ordered correctly. DFO evaluates rules top-down. Put high-confidence, specific intent rules at the top, and the general fallback rule at the bottom.

4. Passing Intent Context to the MAX Agent

Routing the interaction correctly is only half the battle. The agent must know why it was routed to them.

Implementation Steps:

  1. When the interaction lands in the MAX agent workspace, you want the agent to see the AI’s conclusion immediately.
  2. In your DFO Configuration, map the Enlighten Intent and ConfidenceScore variables to Custom Fields within the Customer Card or Interaction Details panel.
  3. When the agent accepts the chat or email, they see: “AI Classified Intent: Billing Dispute (92% Confidence)”.
  4. This allows the agent to immediately open the billing system and greet the customer proactively: “Hi, I see you have a question about your recent invoice. How can I help?”

Validation, Edge Cases & Troubleshooting

Edge Case 1: Sarcasm and Mixed Intents

  • The Failure Condition: A customer emails: “Your website is amazing. I love how I can’t reset my password and my credit card was double charged. Great job.” Enlighten flags the intent as Praise due to the positive words, routing it to the Customer Success queue instead of Billing/Support.
  • The Root Cause: Sarcasm is notoriously difficult for standard NLU models to parse, and the email contains two distinct intents (Password Reset AND Billing).
  • The Solution: Leverage Enlighten Sentiment Analysis in parallel with Intent Classification. If Sentiment == Highly Negative AND Intent == Praise, create a specific Routing Rule that flags this as an anomaly and routes it to a specialized escalation queue for human review. Furthermore, tune your model to identify multi-intent payloads and prioritize the most severe intent (Billing over Password Reset).

Edge Case 2: The “Short Utterance” Trap in Web Chat

  • The Failure Condition: In Web Chat, the user types “Hi” and hits enter. Enlighten evaluates “Hi”, returns a low confidence score, and routes the chat to the general triage queue. The user then types “I need to cancel my order”. It is too late; the chat is already in the wrong queue.
  • The Root Cause: Evaluating the very first message is dangerous in asynchronous channels.
  • The Solution: Implement a Virtual Agent / Bot Gateway in front of the human queue. Have the bot greet the customer and explicitly ask: “Briefly describe what you need help with today.” Only pass the response to that specific prompt to the Enlighten AI model for classification, ensuring you have enough context to make an accurate routing decision.

Official References