Phase

Characteristics

Recommendations

1

• Consume generic models
• Unpredictable and reactive approach
• Localized efforts
• Limited understanding

• Develop basic awareness
• Initiate pilot projects

2

• Fit-for-purpose models on primary Gen AI env.
• Inconsistent processes
• Initial documentation
• Recognizing data quality needs

• Establish centralized strategy
• Basic training
• Evaluate data standards

3

• Leverage enterprise-wide data on Gen AI env.
• Organization-wide standards
• Established governance
• Focus on ethics

• Enhance collaboration
• Address GenAI challenges
• Continuous feedback mechanisms

4

• Run and infer Gen AI models to scale compute/costs
• Active metrics tracking
• Quantitative evaluation
• Data-driven decision-making

• Advanced analytics
• Link to business objectives
• Robust risk management

5

• Build & use models cross env. securely at optimal costs
• Continuous refinement
• Established feedback loops
• Proactive approach

• Foster innovation
• Engage with experts
• Review governance framework

Phase

GenAI Capabilities

Governance Maturity

1

• GenAI Resources: Hardware and platform basics.
• Basic Data Management: Initial storage and management of data.

No AI lifecycle governance

2

• Model Hub: Basic model importing and data importing capabilities.
• Supporting Capabilities: Basic IT operations for GenAI.
• Initial steps in GenAI Application Development: Beginning to tune foundational models.

Some AI policies available to guide AI lifecycle

3

• Model Hosting: Deployment of models as API services, and Model Access Policy Management.
• Model Customization: Introduction to tuning and training models for specific needs.
• GenAI Tuning: Basic customization using Prompt Engineering and Model Fine-tuning.

Common set of metrics to govern AI lifecycle

4

• Model Governance: Addressing risks like bias introduction, regulatory and compliance adherence.
• Model Monitoring: Real-time monitoring capabilities including Bias Detection and HAP Detection.
• GenAI Application Capabilities: Incorporating advanced features like Orchestration and Intent Detection.

Automated validation and monitoring

5

• Advanced Model Customization: Employing cloud platforms for dynamic needs.
• Prompt Monitoring and Security: Ensuring models are protected from advanced threats.
• Advanced GenAI Tuning: Thoroughly customizing models to enterprise-specific jargon and processes.
• Advanced GenAI Application Development: Full-feature generative AI application development and potential model creation from scratch.

Fully automated AI lifecycle governance