Skills in Luma Virtual Agent

Any request to be fulfilled by the Virtual Assistant/Bot must be defined in a logical entity called Skill, where you define the intention of the user request, the required information to process the user request, and fulfillment details that resolve the user’s request. A conversation is started and completed when the user request/intent understood and fulfilled by the Virtual Agent. These components together provide the back-and-forth conversation between the bot and the user and can enhance the end user's service management experience. This article details building skills and taking advantage of the many configurable options.

Luma Virtual Agent supports two distinct paradigms for skill creation and classification. These models define how the system identifies and executes the appropriate skill in response to a user’s request.

Intent-Based Skills (Semantic Skills)

Intent-Based Skills, also referred to as Semantic Skills, represent the latest advancement in skill creation within the Luma Virtual Agent platform. Semantic Skills are a new paradigm for defining and executing virtual agent skills based on intent and classification. These skills are identified based on the user’s intent, rather than relying on a set of manually trained user phrases. This eliminates the inefficiencies of traditional skill training by enabling purpose-driven, AI-powered, and scalable skill identification.

Once Semantic Skills are enabled, the skill developer is guided by the platform itself. Luma prompts the user to define critical aspects such as the skill’s purpose, trigger criteria (phrases or keywords that should activate the skill), and exclusion criteria (phrases that should not activate the skill). Based on these inputs, Luma automatically generates the necessary classification logic using AI-powered models.

Key Advantages:

• No Manual Phrase Training Required
  The system eliminates the need to manually add and tag user phrases. Instead, Luma uses AI to interpret the purpose and intent of the skill based on semantic understanding.

• Natural Language Flexibility
  Because classification is intent-based, Luma can to handle a wide range of user phrasings and linguistic variations without needing explicit examples.

• AI-Based Entity Recognition (NER)
  Entities are automatically extracted from the user's message using Luma’s AI-driven NER capabilities. This removes the need for manual tagging of attributes in training phrases, making the process more scalable and less error-prone.

• Minimal Configuration and Faster Time-to-Deploy
  Skill developers can rapidly define new skills without the overhead of manual training.

• Reduced Maintenance Effort
  There is no need for training with every possible version of the user phrase. Additionally adding evolving business language and usage patterns is easy, making the semantic skills more resilient and easier resilient.

• Prevents Redundancy
  Luma proactively identifies and prevents the creation of duplicate or overlapping skills by validating skill definitions during setup. This ensures a streamlined and non-redundant skill library, resulting in more consistent and reliable skill execution.

2. Phrase-Based Skills (Traditional Skills)

Phrase-Based Skills are the traditional method of skill creation used in Luma (tenant-created up untill Release 3.7). In this model, skill recognition is dependent on a comprehensive set of manually authored training phrases. Skill Developer provides multiple utterances to train the system on how to recognize the most relevant skill.

Additionally, developers can tag entities (such as dates, names) within each phrase so that Luma can extract data during skill execution. The effectiveness of these skills is directly tied to the quantity and quality of training data provided.

Key Advantages

• Fine-Grained Control Over Skill Recognition
  Developers have full control over the exact phrases and patterns that trigger a skill, allowing precise customization of the recognition logic.

• Predictable Behavior
  Because skills respond only to explicitly defined phrases, their behavior is highly predictable and deterministic—useful in environments where strict control is required.

• Effective in Narrow, Well-Defined Use Cases
  Phrase-based skills are suitable for scenarios where user inputs are limited in variation, such as structured requests or guided workflows.

• Supports Attribute-Level Tagging
  Developers can manually tag and map specific words or phrases to attributes, providing precision in entity extraction where AI-based models may be overkill.

• Compatibility with Legacy Deployments
  These skills are fully supported in Luma environments up to version 3.7, making them a reliable choice for customers not yet ready to transition to AI-driven models.

Intent-Based Skills vs Phrase-Based Skills

The evolution from traditional Phrase-Based Skills to AI-powered Intent-Based (Semantic) Skills marks a significant leap in how Luma interprets and responds to the end user requests. While Phrase-Based Skills served as the foundation in earlier versions, their limitations become increasingly apparent as organizations scale.

Below are key reasons why Intent-Based Skills provide a more robust, scalable, and future-ready alternative:

✅ Reduced Manual Effort

Unlike Phrase-Based Skills, which require developers to manually create and tag 10–15 or more training phrases per intent-Intent-Based Skills leverage AI to automatically classify user inputs. Luma guides the user through a simple definition process and generates the necessary classification details, eliminating the need for intensive training and tagging.

✅ Elimination of Redundancy and Conflicts

Intent-Based Skills include built-in mechanisms to prevent duplication. Luma intelligently checks for overlap and rejects redundant skill definitions, ensuring a cleaner skill library and reducing classification errors common in large Phrase-Based environments.

✅ Better Adaptability to Natural Language

Powered by semantic understanding and AI-based natural language processing (NLP), Intent-Based Skills can interpret a broader range of user expressions without needing exact matches. This improves the user experience and supports more conversational interactions.

✅ AI-Powered Entity Extraction

With Named Entity Recognition (NER), Luma can extract entities directly from user input-without requiring developers to manually tag attributes. This reduces setup effort while increasing accuracy in capturing relevant information.

✅ Minimal Maintenance Overhead

Because skills are driven by intent rather than rigid phrase patterns, changes to business logic or user behavior typically do not require retraining. The model is inherently more adaptable, lowering the maintenance burden over time.