AI Centre of Excellence: Moving Beyond Shadow AI Risk to Scaled AI Adoption

Your organisation’s AI systems will make millions of decisions while you read this article. Each decision; every credit approval, pricing adjustment, or customer recommendation, carries risk and opportunity that ultimately rests with your Board. This isn’t hyperbole, it’s the mathematical reality of AI-driven operations in 2025.

The exponential increase in decision velocity from hundreds per day to millions per second has outpaced traditional governance mechanisms. Quarterly reviews and annual audits simply cannot provide oversight for systems that evolve continuously and operate autonomously. This governance gap explains why 88% of AI pilots fail to reach production and why 74% of organisations reported they had an AI breach in 2024.

The solution isn’t to slow AI adoption; that would cede competitive advantage. Instead, Boards need new governance infrastructure designed for AI’s unique characteristics. This is where the integrated approach of the AI Stages of Adoption (AISA), the Five Pillars capability areas, and the Well-Advised priorities become essential; providing the complete framework Boards need to govern AI’s complexity through an AI Centre of Excellence (AI CoE).

The Multi-Speed Challenge of AI Governance

Consider a global manufacturing company navigating their AI journey. Using AISA, we can map their stages of AI adoption: operations has reached the Optimising stage with sophisticated predictive maintenance, while finance remains at Experimenting with basic automation pilots. Meanwhile, marketing is Adopting AI for content generation, and HR hasn’t even begun their AI journey.

This multi-speed adoption pattern, where different functions naturally progress through AISA stages at different rates creates unprecedented governance challenges. Traditional IT governance assumes relatively uniform technology adoption. AI shatters this assumption. Each function faces different risk profiles, requires different capabilities, and delivers value on different timelines.

Or examine a regional financial services firm where customer service has reached the Transforming stage with AI fundamentally reshaping customer interactions, while their risk management team cautiously experiments with AI-assisted fraud detection. The governance requirements for transformative AI differ vastly from experimental applications, yet both need oversight within the same organisation.

This multi-speed reality explains why one-size-fits-all governance fails. An AI CoE provides the sophisticated governance necessary to coordinate across different AISA stages, ensuring appropriate oversight for experimental pilots while enabling scaled transformation where the organisation is ready.

Why the Five Pillars Demand Board-Level Authority

In my interactions with organisations building AI capabilities, I’ve identified five essential pillars that determine success: Governance & Accountability, Technical Infrastructure, Operational Excellence, Value Realisation & Lifecycle Management, and People, Culture & Adoption. Organisations need different maturity levels across these pillars depending on their AISA stage, and only Board-level authority can coordinate this complex capability development.

The instinct to position AI governance within IT reflects outdated thinking. When organisations adopted cloud computing, IT-led Cloud Centres of Excellence made sense because cloud was fundamentally an infrastructure evolution focused primarily on the Technical Infrastructure pillar. AI is different. It demands simultaneous capability building across all five pillars, often driven by line-of-business teams rather than technologists.

Marketing needs Governance & Accountability for ethical content generation. Operations requires Operational Excellence for reliable predictive maintenance. Finance demands Value Realisation frameworks to justify AI investments. Each function needs different pillar capabilities at different maturity levels based on their AISA stage.

This complexity makes IT governance inadequate. McKinsey’s research confirms this: 60% of organisations with successful AI governance have Board-level oversight, compared to 25% without. The correlation reflects a simple truth: effective AI governance requires authority to build capabilities across all Five Pillars simultaneously, something only Board-level positioning provides.

Addressing the Six Board Concerns Through Structured Governance

As I detailed in AI is transforming governance: Six key Boardroom priorities, Boards consistently express six critical concerns about AI adoption: Strategic Alignment, Ethical and Legal Responsibility, Financial and Operational Impact, Risk Management, Stakeholder Confidence, and Safeguarding Innovation. The AI CoE is the mechanism that comprehensively addresses all six.

Rather than treating these concerns as separate issues requiring individual solutions, the AI CoE provides integrated governance that addresses them holistically. It ensures Strategic Alignment by mapping all AI initiatives to the Well-Advised priorities, guaranteeing they advance Innovation, Customer Value, Operational Excellence, Responsible Transformation, and Revenue objectives rather than pursuing technology for its own sake.

The AI CoE addresses Ethical and Legal Responsibility through systematic governance across the Five Pillars, establishing clear accountability frameworks and ensuring AI decisions remain explainable and compliant. It manages Financial and Operational Impact by improving the AI pilot failure rate through structured capability building and value realisation frameworks that track benefits across all Well-Advised priorities.

For Risk Management, the AI CoE provides governance mechanisms that match AI’s operational tempo - not by reviewing every decision but by ensuring systems operate within Board-approved parameters. It builds Stakeholder Confidence through transparent practices and clear communication about AI’s role in augmenting human capability. Finally, it Safeguards Innovation by channelling creative energy within appropriate boundaries, transforming risky shadow AI into governed experimentation.

The Eighteen Essential Functions Mapped to Capability Pillars

Through my research, I’ve identified eighteen core responsibilities that an effective AI CoE must fulfil. These naturally align with the Five Pillars, ensuring comprehensive capability development:

Governance & Accountability

• Human-AI collaboration frameworks: Establishes clear protocols for when and how humans interact with AI systems, defining escalation thresholds, override authorities, and decision boundaries to ensure appropriate human oversight remains in place for critical decisions.

• AI vulnerability management: Identifies and addresses AI-specific risks such as adversarial attacks, model drift, and data poisoning, implementing continuous monitoring and rapid response mechanisms to protect AI systems from manipulation or degradation.

• Misuse and harmful content prevention: Creates safeguards against AI systems being used for unauthorised purposes or generating harmful outputs, including content filters, use case restrictions, and audit trails for sensitive applications.

• Accountability and transparency protocols: Ensures AI decisions can be explained, traced, and attributed to responsible parties, establishing clear ownership for AI outcomes and maintaining documentation standards that satisfy regulatory, ethical, and Board requirements.

Technical Infrastructure

• Data quality and governance standards: Defines and enforces standards for data collection, preparation, and management, ensuring AI systems are trained and operated on accurate, complete, and appropriately governed datasets.

• Security and privacy controls: Implements technical safeguards to protect AI systems and data from breaches, ensuring compliance with privacy regulations and preventing unauthorised access to sensitive information or model IP.

• Technical architecture guidelines: Establishes standards for AI system design, integration patterns, and technology choices, ensuring scalability, interoperability, and alignment with enterprise architecture principles.

Operational Excellence

• MLOps and model lifecycle management: Oversees the end-to-end process of developing, deploying, monitoring, and retiring AI models, ensuring consistent quality and performance throughout each model’s operational life.

• Performance monitoring and optimisation: Tracks AI system performance against defined metrics, identifying degradation or improvement opportunities and implementing continuous improvement processes for model accuracy and efficiency.

• Integration with existing operations: Ensures AI systems work seamlessly within current business processes and technology stacks, managing dependencies and designing workflows that effectively combine human and AI capabilities.

Value Realisation & Lifecycle Management

• Business case evaluation using Well-Advised dimensions: Assesses proposed AI initiatives across the Well-Advised strategic priorities: Innovation, Customer Value, Operational Excellence, Responsible Transformation, and Revenue, ensuring balanced investment decisions.

• Cost optimisation strategies: Identifies opportunities to reduce AI operational costs through efficient resource utilisation, model optimisation, and strategic vendor management while maintaining performance standards.

• Vendor and IP management: Oversees relationships with AI technology providers, negotiates contracts, manages intellectual property rights, and ensures the organisation maintains appropriate control over its AI assets and innovations.

• Benefits tracking across strategic pillars: Monitors and reports on AI value creation using comprehensive metrics that span financial returns, customer impacts, operational improvements, and innovation outcomes.

People, Culture & Adoption

• Skills assessment and development: Evaluates current AI capabilities across the organisation, identifies gaps, and implements targeted training programmes to build necessary competencies at all levels from executive to operational.

• Change management programmes: Designs and implements structured approaches to help employees adapt to AI-augmented work environments, addressing concerns about job displacement while building enthusiasm for new capabilities.

• Bias identification and mitigation: Establishes processes to detect and address bias in AI systems, including diverse team composition, systematic testing protocols, and ongoing monitoring for discriminatory outcomes.

• Stakeholder education and engagement: Creates communication and education programmes for all stakeholders—employees, customers, partners, and investors—building understanding of AI’s role and impact while managing expectations appropriately.

This structured approach ensures your AI CoE builds capabilities systematically while addressing all aspects of AI governance, from technical implementation to strategic value creation.

Creating Value Across the Well-Advised Priorities

An AI CoE doesn’t just manage risk, it ensures AI investments create balanced value across all strategic dimensions. Too often, organisations pursue AI initiatives that deliver narrow benefits, typically focused on cost reduction. The AI CoE uses Well-Advised to ensure more comprehensive value creation.

For Innovation and New Products/Services, the AI CoE identifies opportunities where AI can enable entirely new business models. Our manufacturing example discovered their predictive maintenance capability could transform them from equipment seller to uptime-as-a-service provider.

For Customer Value and Growth, the AI CoE ensures AI enhances rather than diminishes customer experience. The financial services firm’s AI CoE helped them recognise that while chatbots reduced costs, their real value lay in providing 24/7 support that improved customer satisfaction.

For Operational Excellence, the AI CoE goes beyond simple automation to fundamental process transformation. For Responsible Business Transformation, it ensures AI adoption considers ethical implications and stakeholder impact. For Revenue, Margin and Profit, it tracks not just cost savings but revenue enhancement and new value streams.

This balanced approach, operationalised through the AI CoE, transforms AI from a collection of point solutions into a coherent capability that advances multiple strategic objectives simultaneously.

From Recognition to Implementation: Your Path Forward

Establishing an AI CoE requires more than recognition, it demands formal Board action with a clear mandate, appropriate authority, and sufficient resources. The charter must acknowledge your organisation’s multi-speed reality, with different functions at different AISA stages. It must provide authority to build capabilities across all Five Pillars systematically. And it must ensure AI initiatives create value across all Well-Advised priorities.

Your Board resolution should establish direct reporting to the risk committee, ensuring independence from operational pressures. It should provide authority to govern AI initiatives across all AISA stages, from early experiments to scaled transformations. Resources must be sufficient to build capabilities across all Five Pillars, not just technical infrastructure. Success metrics should reflect balanced value creation across Well-Advised dimensions, not just cost savings or risk mitigation.

Your Eight-Week Journey to AI Governance Excellence

In this series of articles, I’ll provide a practical roadmap for establishing your AI CoE using an approach proven with organisations in all parts of the world. Over the next seven weeks, we’ll explore:

• Week 2: Deep dive into the eighteen AI CoE functions and how they build Five Pillars capabilities
• Week 3: Assessing your AI landscape using AISA to understand your multi-speed reality
• Week 4: Designing your AI CoE structure to govern across different AISA stages
• Week 5: Building essential capabilities using the Five Pillars framework
• Week 6: Launching your first 90 days with Well-Advised value focus
• Week 7: Scaling beyond pilots by leveraging integrated governance
• Week 8: Future-proofing your AI CoE as you progress through AISA stages

Each article builds practical implementation guidance using these proven mechanisms, showing how they work together to create comprehensive AI governance.

The Imperative for Integrated Governance

The question facing Boards isn’t whether to establish an AI CoE, but how quickly they can implement one using a proven approach. The integrated framework of AISA, Five Pillars, and Well-Advised provides the complete structure needed to govern AI’s complexity while ensuring strategic value creation.

Forward-thinking Boards are already establishing AI CoEs that operationalise these frameworks, transforming AI from disconnected experiments into coordinated capabilities. They’re building systematic governance that adapts to multi-speed adoption, develops capabilities comprehensively, and ensures balanced value creation.

Your next Board meeting presents an opportunity. Will you continue accepting the risks of ungoverned AI adoption? Or will you take the essential step of establishing an AI CoE that implements proven frameworks for comprehensive governance?

The journey begins with understanding not just why an AI CoE is essential, but how proven frameworks guide its implementation. That understanding, backed by board action, transforms AI from your greatest ungoverned risk into your most powerful competitive advantage.

About the Author

Mario Thomas is a transformational business leader with nearly three decades of experience driving operational excellence and revenue growth across global enterprises. As Head of Global Training and Press Spokesperson at Amazon Web Services (AWS), he leads worldwide enablement delivery and operations for one of technology's largest sales forces during a pivotal era of AI innovation. A Chartered Director and Fellow of the Institute of Directors, and an alumnus of the London School of Economics, Mario partners with Boards and C-suite leaders to deliver measurable business outcomes through strategic transformation. His frameworks and methodologies have generated over two-billion dollars in enterprise value through the effective adoption of AI, data, and cloud technologies.