Configuring
AI Workflows

AI Governance — Module 3 of 4

Click to start

Learning objectives

“As an administrator, I want to configure workflows that support AI governance — including intake, approvals, human oversight, and escalation paths.”


This third module is part of a four-part series:

AI Governance

Module 3 — Contents

  1. Intake workflows
  2. Approval workflows
  3. Human oversight controls
  4. Escalations

Training is interactive — you explore ​ValidMind live. Try it!

→ , ↓ , SPACE , N — next slide     ← , ↑ , P , H — previous slide     ? — all keyboard shortcuts

Intake workflows

What is an intake workflow?

Intake workflows control how new AI systems enter your governance program. They ensure:

  All AI systems are registered

  Required information is captured

  Initial classification is performed

  Ownership is assigned

Configuring intake

In ​ValidMind:

  1. Define required fields for registration
  2. Set up validation rules
  3. Configure automatic notifications
  4. Route new registrations for review

Example: Risk classification workflow

A screenshot showing a Risk Classification Workflow with model stage changes, user actions, approvals, and condition branches for different risk levels

 Click to enlarge

Risk classification workflow steps

This example intake workflow demonstrates key AI governance patterns:

  • Model Stage Change — Sets model stage to In Risk Classification
  • User Action — Collects risk classification materials with deadline
  • Broadcast — Notifies stakeholders of pending risk rating
  • Approval — Routes to Model Validation Governance for review
  • Condition Branch — Different paths for High Risk, Medium Risk, or Default

Approval workflows

Purpose of approval workflows

Approval workflows ensure AI systems receive appropriate review before deployment. They implement:

  Risk-based approval gates

  Multiple approver requirements

  Segregation of duties

  Documented decision-making

Designing approval flows

Consider these design principles:

Risk level Approval requirements
High-risk Multiple approvers, governance committee
Medium-risk Manager + compliance review
Low-risk Single approver

Configuring approvals

​ValidMind supports:

  • Workflow states and transitions
  • Role-based approval authorities
  • Sequential and parallel approvals
  • Approval deadline tracking

A screenshot showing approval step configuration with role selection in the workflow builder

Approval step configuration

Human oversight controls

Regulatory context

The EU AI Act (Article 14) requires human oversight** for high-risk AI systems. Human oversight includes:

  Ability to understand AI outputs

  Capability to intervene

  Authority to override decisions

  Mechanisms for feedback

Implementing oversight

Use ​ValidMind to document and enforce oversight:

  • Record oversight requirements in documentation
  • Configure review workflows for AI outputs
  • Track intervention decisions
  • Maintain evidence of human involvement

Review workflows

Set up periodic review workflows:

  1. Define review frequency by risk tier
  2. Configure reviewer assignments
  3. Track review completion
  4. Escalate overdue reviews

Escalations

Escalation paths

Configure escalations for:

  Overdue approvals

  High-risk validation issues

  Policy violations

  Compliance incidents

Escalation ensures issues reach appropriate decision-makers.

Next steps

Continue to Module 4 to learn about compliance and oversight.

Module 4: Compliance and oversight