Workflows for models
- Model Lifecycle
- This workflow takes an AI model from the completion of the candidate process through to approval
for deployment. It includes multiple stages and sub-workflows that involve various stakeholders.
- Foundation Model Onboarding workflow
- The Foundation Model Onboarding workflow takes a foundation model through an approval process.
The approval process includes stages for legal approval, AI ethics approval, and finance
approval.
- Model Candidate workflow
- This workflow allows a user to add a Model object to the inventory as a candidate. The Model candidate is submitted for Approval as either a Model or a Non Model. The approver can override the candidate proposal. After a model candidate is confirmed as a Model, the Model Development and Documentation process can begin.
- Model Validation workflow
- This workflow is performed at the completion of the model development and documentation workflow. The Review Planning team that is identified on a Preference object is responsible for completing this review. This workflow is also used for conducting reviews after the model is in production.
- Model Deployment
- The Model Deployment workflow is used to govern the process of model deployment when development is complete. If the model is being deployed in production, there is an additional step to make the model ready for production before deployment is complete.
- Model Risk Assessment
- This workflow performs a model risk assessment on the model, the results of which are used to assign a tier to the model. The Model Risk Scorecard calculation then uses the values in Preference records to compute risk scores and tier. At the end of the workflow, the scores and tier are copied to the parent Model.
- Model Attestation workflow
- This workflow is typically started by an MRG administrator and records a model owner's response to a request for attestation.
- Model Decommission workflow
- This workflow is used to remove a Model from production and retire it.
- Model Change Request workflow
- This workflow provides governance for changes to Models. A workflow can be based on changes in the business or to the data and other inputs to a Model. Users can accept, approve, or reject the change and decide whether it is material or not.
Workflows for use cases
- Use Case Request
- The Use Case Request workflow takes a proposed use case through an approval process. All stakeholders reviews must be completed before the use case is approved.
- Use Case Stakeholder Review
- The Use Case Stakeholder Review workflow takes a use case through a series of reviews by stakeholders. Information about each review is captured in a Use Case Review object.
- Use Case Development and Validation
- The Use Case Development workflow takes a use case through a development and validation process.
After the validation process is complete, the Use Case Deployment Approval workflow starts.
- Use Case Deployment Approval
- The Use Case Deployment Approval workflow takes a use case through an approval process for
deployment of the model.
- Model Use Case Request
- This workflow is disabled by default. The Use Case Request workflow replaces it.
Workflows for metrics
- Metric Value workflow
- This workflow automates the Breach Status calculation and facilitates performance monitoring of
deployed models. This process is critical to the ability to proactively decide to change the model
or its usages or to remove a model from production.
Typically, metrics and their values are pulled in from watsonx.governance. But metrics can also be created and managed in Governance console. In this case, an MRG administrator creates Metric and Metric Value objects, a Metric Capturer provides the latest data for the metric, and the Metric Owner reviews and approves it. The workflow calculates breach status for the Metric and copies the most recent Metric Value information to the Metric.
- Metric Value Creation workflow
- When the next collection date comes due for an active metric, this workflow automatically
creates a new metric value, and populates it with owner, capturer, and threshold information from
the parent metric. The collector can then populate the metric value information and submit it for
review. Note: This workflow is not applicable to Governance console metrics.
Workflows for questionnaires
- AI Assessment Workflow
- The AI Assessment Workflow manages a series of assessments for use cases.
The Applicability Assessment stage uses the predefined EU AI Act Applicability Assessment questionnaire assessment. After a user completes the assessment, the workflow triggers the questionnaire response actions, which set the value of the EU AI Risk Category field of the use case.
The Risk Identification Assessment stage uses the predefined AI Risk Identification Questionnaire questionnaire assessment. After a user completes the assessment, the workflow triggers the questionnaire response actions, which associate risks to the use case.
To use the other assessment stages in this workflow, create questionnaire templates and assessments for each stage.
-
- Questionnaire Assessment Workflow
- The Questionnaire Assessment workflow moves a questionnaire assessment through the information gathering, review, and approval stages.
Other workflows
- Challenges workflow
- This workflow is started against a Model, one of its Model Deployments, or a Review. The result can be no action or changes to a Model or Model Deployment.
- Model Risk Assessment Workflow
- This workflow manages risk assessments for a use case by using the Risk object type.
The owner of the risk is asked to assess the use case, including the inherent risk, inherent likelihood of the risk, and the mitigation strategy. When the risk assessment is complete, the workflow sets the status of the risk to Approved.