Governance Types

SCAI offers multiple AI governance models to align with different business needs and complexity levels. Choose between NLU for structured, intent-based interactions, Agentics for dynamic problem-solving capabilities, or Composite approaches that blend both methodologies for flexibility and control. Learn more about their differences.

Read the explanations below to understand your options.

Intent-based Model

The Intent-based model is designed for structured, intent-driven interactions with predefined conversational paths. It excels in scenarios where user interactions are relatively predictable and can be mapped to specific intentions. Ideal for customer service applications with clear, repetitive workflows like order tracking, basic support inquiries, or information retrieval.

In this model, the system focuses on accurately identifying user intents and matching them to pre-configured responses. It works best when you have a well-defined set of user goals and want a reliable, consistent interaction model with minimal variability. Think of it as a sophisticated flow chart that efficiently routes user requests to the most appropriate predetermined response.

Agentic Model

The Agentic model (LLM-based) represents a more dynamic and adaptive approach to interactions. This approach enables complex problem-solving, creative reasoning, and context-aware responses. It's particularly suitable for scenarios requiring nuanced understanding, open-ended problem-solving, or interactions that cannot be easily predefined. In this model, AI agents can break down complex tasks and adapt their approach based on contextual cues and handle unexpected queries, providing provide more human-like conversational experiences.

Composite (Hybrid) Model

The Composite models combines structured intent recognition and dynamic problem-solving by combining NLU and Agentic approaches. This hybrid model offers two primary configurations:

Intent-first

In this configuration, the system prioritizes intent recognition while maintaining the ability to leverage AI agents for more complex tasks. A Supervisor agent first attempts to match user inputs to predefined intents, ensuring efficient handling of common, predictable interactions. When an intent cannot be directly matched or requires more nuanced processing, specialized AI agents are activated to provide more flexible, context-aware responses. This approach is perfect for organizations wanting to maintain the reliability of intent-based systems while introducing adaptability for edge cases.

AI Agent in Intent-first model

Combining intent classification with an AI agent's dynamic capabilities can be achieved using a Transfer cell. This component follows the intent classification stage, allowing seamless interaction between identifying user intent and leveraging AI functionality to respond effectively, as shown in the example below.

This is a good option if you want to migrate an Intent-based agent to agentics models.

Agentics-first

Conversely, this configuration prioritizes AI agent capabilities while retaining some intent-based routing mechanisms. The system first leverages the dynamic problem-solving capabilities of AI agents, using intents as a secondary mechanism, conserving specific and/or deterministic use cases. This model is ideal for complex environments where user interactions are diverse and unpredictable, but some level of structured routing can still enhance efficiency. It allows for more creative and adaptive responses while maintaining a soft structure through intent-based insights.

Models Compatibility

After creating your project, some restrictions for changing the models apply:

Model compatibility matrix