DESIGN BASELINE • VERSION 1.0 • 22 SEPTEMBER 2026
MICROSOFT & AZURE CENTRED GRC COMMERCIAL MODEL
Ω
aioperatingmodel.work
AI Governance Board Framework
Abstract AI Governance Sculpture
A MICROSOFT-CENTRED GOVERNANCE, RISK & COMPLIANCE OFFERING

Make accountable decisions about AI by turning platform evidence into three connected scorecards.

We help organisations establish a prioritised action backlog and a recurring AI Governance Board across Microsoft 365 Copilot, Copilot Studio agents, Azure Foundry, and external AI services.

Explore Three Connected Scorecards View 30-Control Live Matrix 6 Layers Taxonomy
TAXONOMY
6 Layers
Adopt • Defend • Govern
CONTROLS
30 Boxes
Every box navigable
ROLLUP RULE
0% Average
Strict deterministic worst-case
LINEAGE
100% Trace
EV → AS → F → T → D
ESTATE SCOPE
M365 & Azure
Copilot, Agents & Custom AI
01 / EXECUTIVE RECOMMENDATION

Establish an AI Governance Board with delegated management authority.

The forum decides where AI may operate, what conditions apply, who owns residual risk, and what must change. The service provider supplies evidence, analysis, and coordination; the client retains institutional accountability.

Buyer Problem Service Response Management Outcome
AI use is fragmented across products and teams Inventory systems, owners, models, tools, and data dependencies An agreed scope and a visible list of unknowns
Reports show activity but do not create decisions Use common RAG rules with an explicit decision required field Prioritised approvals, restrictions, and remediation
Business ownership is weaker than technical ownership Assign a business accountable owner for each use case A named person accepts benefits, impacts, and risk
Audit evidence is assembled after the event Capture evidence and decision lineage continuously A reproducible record of what was known and approved
New agents introduce actions and external connections Gate permissions, MCP tools, and release changes Controlled expansion of agent execution authority

5 Decisions Required to Launch

  • 1. Appoint Chair: Designate the executive sponsor and board chair (typically CIO, COO, or delegated senior executive).
  • 2. Approve Taxonomy: Ratify the six-layer control taxonomy and explicit risk appetite thresholds.
  • 3. Select Scope: Agree the initial business unit and 10 to 25 production AI systems.
  • 4. Authorise Evidence: Grant required telemetry access and formally name control owners.
  • 5. Fund Remediation: Fund the baseline and reserve technical capacity to remediate findings.

The Minimum Useful Product

A functioning board, a reconciled inventory, one operational scorecard across all 30 controls, one business scorecard, and one executive rollup. Every red or amber finding must have an owner, an action, and a due date. A monthly report without a closed action loop is rejected.

Demonstrating Success

Measure coverage of the agreed estate, unresolved critical issues, evidence freshness, overdue actions, and decision turnaround. Cost and adoption provide business context, but efficiency gains never offset security or compliance gaps.

02 / TAXONOMY & CONTROL STRUCTURE

6 Layers for AI Governance

Derived from the foundational architecture diagram: 30 operational controls organised into Adopt, Defend, and Govern groups. The six layers operate continuously throughout the lifecycle; they are not sequential project milestones.

Abstract 3D Architectural Tiers
3D Parametric Tier Alignment • 30 Operational Control Points Microsoft 365 Copilot & Azure Foundry Native
GROUP: ADOPT Build responsibly
10 Controls
LAYER 01 Owner: AI Estate Lead

AI Inventory

"Do we know what AI exists and who owns it?"

01A System Inventory 01B Risk Classification 01C Ownership & Roles 01D Model & Tool Registry 01E Usage Visibility
LAYER 02 Owner: AI Delivery Lead

Responsible Deployment

"Are intended uses and changes approved and safe to release?"

02A Use Case Selection 02B Architecture & Model 02C Deployment Practices 02D Change Control 02E DevSecOps Integration
GROUP: DEFEND Test and protect
10 Controls
LAYER 03 Owner: CISO Delegated Lead

AI Security & Access

"Can AI access or change only what it is authorised to?"

03A Identity & Access 03B Context Protection 03C Tool & MCP Controls 03D Access Controls 03E Data Integrity
LAYER 04 Owner: AI Quality & Ops Lead

Testing & Monitoring

"Can we detect harmful behaviour and respond?"

04A Pre-Production Eval 04B Red Teaming & Threats 04C Runtime Monitoring 04D Drift Detection 04E Incident Response
GROUP: GOVERN Authorize and oversee
10 Controls
LAYER 05 Owner: Business AI Owner

Human Oversight

"Can accountable people challenge, intervene and stop AI?"

05A Decision Review 05B Escalation Paths 05C Override Authority 05D Output Validation 05E Accountability Mapping
LAYER 06 Owner: Risk & Compliance Lead

Compliance & Audit

"Can we explain obligations, decisions and supporting evidence?"

06A Policy & Decision Rights 06B Regulatory Alignment 06C Audit Evidence 06D Incident Reporting 06E Audit Trails & Logs
03 / SCORECARD ARCHITECTURE

Three Connected Scorecards

One underlying evidence model so executives, business leaders, and operators inspect the same risks at different levels of detail. Synthetic frozen reporting period: September 2026.

CURRENT ROLLUP STATUS ENTERPRISE OVERALL: RED (CONTAINED)
Crystalline Decision Matrix
120 Applicable Scoped Assessments • 4 Active Production AI Systems Coverage: 116 / 120 (96.7%) • 0 Missing Inventory Records
EXECUTIVE ROLLUP VIEW

C-Suite Scorecard

STATUS: RED Unchanged since prior period
EVIDENCE COVERAGE
96.7%
116 / 120 (4 Stale)
RAG DISTRIBUTION
115 G / 4 A / 1 R
Zero optimistic averaging
CRITICAL FINDINGS
1 Open / 1 Contained
Containment is not closure
CORRECTIVE ACTIONS
5 Tasks (2 Overdue)
1 Task per non-Green item

Enterprise Layer Rollup Statuses

!
Executive Summary & Pending Decision Request

Overall status remains Red. The red tool access finding (CRM write overprivilege in Customer Operations) is contained by disabling the write action; the underlying permission defect remains open.

Board Approval Request: "Approve two engineering days to reduce permission scope and verify the fix. Do not approve wider autonomous deployment until the release authority receives the negative retest."
BUSINESS UNIT VIEW

Business Scorecard: Customer Operations

Accountable Owner Customer Operations Director
Business Review Mandate: Assisted customer service continues with the CRM write action disabled. Staff must record updates manually. The business owner accepts temporary operational impact and allocates an engineer and supervisor to test corrected permission scope. Re-enabling the tool requires release approval, not just ticket completion.
OPERATIONAL VIEW

Complete 30-Control Register (Live Matrix)

120 assessments across CO1, CO2, FIN, and Shared Platform. Filter by status to inspect exceptions.

Box Control Name CO1 (AGT 001) CO2 (AGT 002) FIN (Finance) SH (Shared) Action Task
EXECUTION CONTROL

Prioritised Action Backlog

5 Open Tasks
INTERACTIVE LINEAGE TRACER

Trace One Finding Through All Three Views

From raw Entra/API evidence to the Board Decision: zero colour averaging, immutable accountability.

DETERMINISTIC CHAIN
04 / COMPLETE SPECIFICATION

30 Operational Control Catalogue

Every control box from the architecture baseline with defined entities, measures, Green/Amber/Red thresholds, and corrective actions.

05 / TECHNOLOGY INTEGRATION

Microsoft-Centred Reference Architecture

Grounding governance in native Microsoft 365 Copilot, Copilot Studio, Azure Foundry, Purview, and Entra ID controls, supported by an immutable evidence store in Dataverse or Azure SQL.

Microsoft AI Reference Architecture Art
6 Architectural Planes • Purview, Entra, Foundry & Copilot Azure Graph & Audit Ingestion

Platform Coverage & Mixed Technology Estates

Delivery caveats and telemetry boundaries across hybrid enterprise environments.

Estate Segment Evidence and Controls to Validate Boundary or Delivery Caveat
M365 Copilot & SharePoint Admin inventory and usage; source permissions; Purview policies and audit log export [S1] Grounding safety depends on underlying SharePoint access. Verify exact workloads and retention.
Copilot Studio & Power Platform Environment policies, agent authentication, connectors, publishing lifecycle evidence [S2] Validate enforcement for each channel. Do not assume one policy covers every custom agent.
Custom AI on Microsoft Foundry Versioned evaluations, traces, monitored application signals, safety benchmarks [S3–S4] Application instrumentation and scenario-specific red teaming remain required.
External Models via Azure Gateway Provider identity, model version, region, contractual terms, application telemetry Hosting route changes obligations and available telemetry. Brand name alone is insufficient.
Direct External AI SaaS Supplier diligence, SSO enforcement, admin audit exports, user attestations No assumed parity with Microsoft telemetry; record missing evidence as Unknown explicitly.
MCP Servers & Enterprise Tools Tool manifest, schema authentication, delegated authority, negative permission tests Protocol compatibility does not prove trust. Review every tool allowlist and target API destination.

Insight to Action: The 8-Step Closure Pipeline

A closed loop ensures findings are never lost in reporting. Only independent verification closes a Red finding.

1
Collect
Data Engineer
2
Assess
Rules Engine
3
Explain
Control Owner
4
Recommend
AI Specialist
5
Authorise
Business Owner
6
Execute
Operator
7
Verify
Ind. Reviewer
8
Report
Service Mgr
06 / OPERATING CHARTER

Board Charter & Lifecycle Decision Rights

The Board operates with delegated authority from executive risk governance within approved risk appetite. It does not replace statutory boards, DPOs, CISOs, or incident command.

AI Governance Board Forum Membership (60-minute standing monthly agenda)

6 Strict Lifecycle Release Gates

Governing progression from proposal to safe decommissioning.

Internal Risk Tiers

Internal routing rules determining depth of control verification.

TIER 1
Low-Impact Assistive
Internal drafting, search, code assistance without external execution authority.
TIER 2
Sensitive or Customer-Facing
Internal confidential processing or non-consequential customer advisory.
TIER 3
Consequential / Autonomous
Regulated processes, external actions, or significant impacts on individuals.
*Internal triage tiers, distinct from statutory EU AI Act classifications.
07 / GO-TO-MARKET & COMMERCIALS

Productised Services & Commercial Model

Predictable delivery economics built around high-margin governance baselines, foundations, and recurring managed board services.

Executive GTM Wave Art
Strategic Economics • 4 Productised Service Packages First Year Target: £258,000 (48.8% Gross Margin)

Illustrative Client Value Model

Realising tangible management capacity and licensing cost containment.

Monthly manual audit prep: 10 Client Days
Automated reduction by service: -6 Days
Validated rate (£600/day): £3,600 / month
Licence & consumption optimisation: £2,000 / month
Direct Quantified Monthly Value: £5,600 / month

*The £2,400 monthly difference to the £8k core fee represents risk containment, regulatory defensibility, and executive acceleration. Avoid asserting unsubstantiated fine avoidance calculations.

First Year Planning Scenario

Pipeline and revenue model under conservative delivery assumptions.

4 × Baseline Engagements (£18k) £72,000
2 × Foundation Conversions (£45k) £90,000
2 × Managed Clients (6 Months @ £8k) £96,000
Total Recognised Service Revenue £258,000
Direct Cost: £132,000 (168 Days) Gross Profit: £126,000 (48.8%)

90-Day Go-to-Market Execution Plan

Authoritative References & Standards Alignment

S1 & S2 Microsoft Learn

Purview security & compliance for M365 Copilot; Copilot Studio security, DLP connectors, and agent authentication.

S3 & S4 Microsoft Foundry

Observability, tracing, and automated pre-production evaluation benchmarks for custom generative AI applications.

S5 NIST AI RMF

NIST AI Risk Management Framework 1.0 (MAP, MEASURE, MANAGE, GOVERN). Voluntary reference framework.

S6 UK ICO Guidance

Information Commissioner's Office AI & data protection guidance; individual rights and accountability lineage.

S7 European Commission

EU AI Act regulatory timeline and high-risk conformity assessment obligations.

Model Provenance

Design baseline v1.0 • 22 September 2026. The 6-layer 30-box structure provides the foundational taxonomy.