The AI Centre of Excellence: A Blueprint for MENA Enterprises

# Why CoEs Work for MENA

**Fragmentation is the silent killer of AI investment in the Middle East and North Africa.** Enterprises across the GCC and broader MENA region are accumulating AI pilots, procurement contracts, and isolated data science teams. McKinsey’s 2025 MENA AI Readiness Report finds that 78 percent of large regional organisations have at least three concurrent AI initiatives. Yet only 12 percent describe these efforts as coordinated or mutually reinforcing.

This fragmentation creates predictable failure modes: duplicated tooling spend, inconsistent data governance, talent islands unable to influence enterprise architecture, and leadership unable to compare performance across business units. The financial cost is significant. BCG estimates that connected AI programmes outperform siloed pilots by a factor of 2.4 in return on investment — a gap that widens in organisations operating across multiple jurisdictions with heterogeneous regulatory environments.

**A Centre of Excellence (CoE) changes the physics of AI adoption.** Rather than treating capability as a series of local experiments, a well-designed CoE creates shared services, common standards, and governance discipline while still allowing business units the autonomy to solve domain-specific problems. For MENA enterprises, where national strategies, family-owned governance structures, and multinational compliance requirements intersect, the CoE model offers a middle path between centralised control and chaotic decentralisation.

Gartner’s 2026 CIO Survey shows that enterprises with a formal AI CoE are 3.1 times more likely to scale AI beyond pilot stages and 2.2 times more likely to report measurable revenue impact. In a region characterised by ambitious national AI strategies — from UAE AI Strategy 2031 to Saudi Vision 2030’s digital transformation pillars — the CoE is not merely an operational convenience. It is the structural mechanism through which policy aspirations become enterprise reality.

# Operating Models

**No single organisational form fits every enterprise.** MENA’s business landscape ranges from sovereign wealth fund-backed conglomerates and government-linked entities to mid-market family businesses and regional divisions of global technology firms. CoE design must reflect these realities.

Three dominant operating models have emerged across the region’s most advanced AI programmes.

### Centralised CoE

A centralised model concentrates all AI talent, budget, infrastructure, and governance within a single reporting line — typically the CIO office, chief data officer function, or a dedicated AI transformation team. All business units access AI capability through shared service agreements and standardised tooling.

**Advantages for MENA enterprises:** Consistent execution aligned to group strategy; rapid standardisation of data governance; easier compliance with cross-jurisdictional regulatory requirements; clear budget accountability. This model suits large, diversified holding companies and government-linked entities with unified governance structures.

**Risks include** perceived detachment from frontline business problems, slower response to divisional urgency, and potential bottlenecks in resource allocation.

Accenture’s 2025 Global AI Study found that centralised AI governance reduces time-to-compliance by 58 percent compared with federated models, primarily because data policies and algorithmic risk controls are defined once and applied universally.

### Federated CoE

A federated model distributes AI capability across business units or divisions while maintaining a thin coordinating layer at headquarters. Each unit retains its own data science team and technology choices, subject to minimum standards defined centrally.

**Advantages for MENA enterprises:** High responsiveness to local business context; stronger unit-level ownership of outcomes; enables parallel experimentation across unrelated sectors. This model is common among multinational banks, healthcare groups, and energy conglomerates where operating units have distinct customer bases, regulatory regimes, or legacy technology stacks.

**Risks include** drift in data quality standards, inconsistent ethical review processes, and difficulty aggregating insights at the enterprise level.

PwC analysis suggests that federated AI governance is most effective when the centre defines 20 to 30 percent of policy — data classification, risk tolerance, and procurement standards — while leaving 70 to 80 percent of execution decisions to units.

### Hub-and-Spoke CoE

The hub-and-spoke model integrates elements of both centralised and federated approaches. A compact central hub owns strategy, governance, platform architecture, and specialist talent. Local spokes — embedded within business units — handle implementation, change management, and use-case execution.

**Advantages for MENA enterprises:** Preserves business unit proximity while maintaining central standards; creates a clear talent pathway from hub to spoke and back; reduces duplication of infrastructure without removing divisional agility. This model has gained traction among fast-growing MENA banks and retail groups undergoing digital transformation.

**Risks include** role ambiguity between hub and spoke teams, potential status friction, and the need for strong matrix management capabilities.

Deliotte’s 2026 Digital Transformation Review reports that hub-and-spoke AI organisations achieve the highest combined scores on governance discipline and business impact among MENA enterprises with revenues above one billion dollars.

### Mapping Operating Models to MENA Enterprise Types

No model is universally optimal. Organisations should revisit CoE design every 18 to 24 months as strategy, regulatory requirements, and competitive pressures evolve.

# Mandate

**A CoE without a clear mandate becomes a consulting shop with ambiguous accountability lines.** Mandate definition is the difference between an AI function that drives enterprise transformation and one that delivers slides.

The most effective MENA CoEs operate across four mandate dimensions: governance, enablement, delivery, and ecosystem.

**Governance.** The CoE owns enterprise AI policy: data classification frameworks, algorithmic risk registers, bias testing protocols, and regulatory compliance roadmaps aligned with UAE, Saudi, and broader regional requirements. It approves high-risk AI deployments before production release and maintains the organisation’s AI asset inventory.

**Enablement.** The CoE trains, certifies, and career-paths internal AI talent. It curates vendor relationships, negotiates enterprise licences, and maintains internal knowledge repositories. Through enablement, the CoE multiplies its impact: every external team it upskills extends the organisation’s AI capability surface.

**Delivery.** The CoE delivers complex or cross-functional AI projects that exceed unit capacity — enterprise-wide language models, customer data platforms, process automation factories, and predictive maintenance backbones. These are not merely technical builds; they are strategic assets shaped by CoE standards.

**Ecosystem.** The CoE represents the enterprise in external AI ecosystems: national AI programmes, academic partnerships, vendor early-access programmes, and sector-specific working groups. This outward-facing posture ensures the organisation benefits from regional knowledge flows rather than operating in isolation.

Gartner recommends that CoE mandates be codified in board-approved charters and reviewed annually. The charter should specify decision rights, escalation paths, resource authorities, and unambiguous performance metrics — avoiding the vague language that turns mandates into aspirations.

# Team

**The right mix of roles determines whether a CoE becomes an engine of capability or an echo chamber of theory.** MENA’s talent ecosystem — with its blend of local graduates, returning expatriate specialists, and global consultancies — offers both opportunity and complexity.

A fit-for-purpose AI CoE typically requires four interconnected roles.

**AI Strategy and Governance Lead.** This is the operational link between the CoE and the C-suite. The lead interprets business objectives into AI investment priorities, ensures regulatory alignment, and chairs governance reviews. In MENA enterprises, successful governance leads often combine technical credentials with deep familiarity with local regulatory environments — such as the UAE’s Data Protection Law or Saudi Arabia’s Personal Data Protection Law.

**AI Platform and Data Engineering Team.** This group owns the centralised infrastructure: cloud environments, data lakes, feature stores, MLOps pipelines, and API gateways. Practical platform teams in the region often operate across hybrid architectures — multi-cloud for flexibility, on-premise for regulated data, and edge deployments for operational sites. They ensure that data flowing into AI systems is trustworthy, lineage-tracked, and fit for purpose.

**AI Research and Solutions Team.** This is the applied science layer: data scientists, machine learning engineers, prompt engineers, and domain specialists. The team prototypes, validates, and optimises models before transferring proven solutions to implementation partners or business unit teams. MENA-specific challenges — multilingual NLP across Arabic dialects, geospatial analysis for logistics, computer vision for surveillance and manufacturing — demand specialised modelling skills that global generic platforms rarely provide.

**Change Management and Adoption Team.** Too many CoEs treat adoption as a training problem. In reality, adoption requires stakeholder mapping, resistance diagnosis, communication strategy, and frontline enablement. This team ensures that users trust AI outputs, understand limitations, and integrate recommendations into daily workflows. In MENA contexts, where hierarchy shapes decision-making, the team must engage middle management as active adoption sponsors rather than passive recipients.

**Sourcing strategy.** Leading MENA CoEs adopt a blended team model. Core strategic, governance, and platform roles remain permanent employees. Specialist research and delivery engagements are supplemented through managed service partnerships, university research collaborations, and secondees from technology vendors. Accenture’s model suggests that a 70/30 ratio of permanent to contracted capability balances institutional memory with intellectual freshness.

# Platform/Tooling

**Technology choices should follow mandate, not precede it.** Yet many organisations over-invest in platforms before clarifying what the CoE is meant to deliver.

### Infrastructure Layer

Enterprises operating across MENA’s varied data residency requirements often adopt a multi-region cloud strategy. Leading choices include government-aligned sovereign cloud instances — such as Azure Government regions in the UAE and AWS segregated cloud in Saudi Arabia — alongside hyperscaler commercial clouds for non-sensitive workloads.

On-premise capabilities remain essential for regulated industries and for assets where connectivity constraints exist, such as remote oil fields or logistics hubs in the GCC.

### Data Foundation

A production AI platform rests on three data pillars: governance-quality structured data, unstructured content pipelines, and real-time streaming layers. MENA enterprises frequently underestimate the effort required to unify legacy ERP systems, customer databases, and operational telemetry. A common pattern among successful regional deployments is a six-to-twelve-month data estate modernisation sprint before production AI builds commence.

### Model Development and MLOps

Core tooling includes orchestration platforms (such as MLflow or Kubeflow), experiment tracking, model registries, and CI/CD pipelines tailored to AI workloads. For generative AI initiatives, enterprises add vector databases, retrieval-augmented generation frameworks, and prompt management systems. MENA-specific requirements — Arabic language models, right-to-left rendering, and regional content compliance — are increasingly addressed by specialised libraries and hosted model services.

### Security and Compliance

AI platforms must embed security at the infrastructure, data, and model layers. The CoE’s governance function validates model cards, monitors for concept drift, and maintains audit logs for regulatory inspection. Integration with enterprise security operations centres ensures that AI workloads are not eroded by adversarial inputs or data poisoning.

**Tooling consolidation versus best-of-breed.** McKinsey’s research shows that MENA enterprises which consolidate on a small number of tightly integrated platforms reduce overhead by 35 percent compared with those assembling heterogeneous best-of-breed stacks. The optimal approach is typically a managed platform core (cloud provider or enterprise AI vendor) supplemented by specialist point solutions where the core falls short of use-case requirements.

# Use Case Portfolio

**The CoE’s credibility is built on a visible portfolio of implemented use cases.** Portfolio composition signals both ambition and operational discipline to the board, employees, and external stakeholders.

### Portfolio Categories

**Operational excellence use cases** — predictive maintenance, supply-chain optimisation, demand forecasting, and quality inspection — often deliver the clearest financial returns. These applications leverage structured historical data and benefit from mature modelling techniques. For MENA manufacturing and logistics enterprises, operational AI can reduce asset downtime by 15 to 25 percent and improve inventory turnover by 10 to 20 percent.

**Customer and employee experience use cases** — conversational assistants, hyper-personalisation, sentiment analysis, and talent acquisition tools — extend AI’s impact beyond cost reduction into revenue growth and retention. These initiatives require investment in unstructured data handling and change management but create durable competitive differentiation.

**Strategic innovation use cases** — scenario planning, generative design, synthetic data generation, and predictive policy modelling — are exploratory by nature. They should be managed through a lightweight portfolio review process that balances experimentation discipline with tolerance for informed failure.

**Risk and compliance use cases** — fraud detection, regulatory monitoring, algorithmic auditing, and cybersecurity anomaly detection — are non-negotiable in regulated industries and offer rapid ROI by reducing loss events.

### Sequencing the Portfolio

Enterprise AI journeys benefit from a layered sequence: foundational use cases that establish data quality and build internal confidence, followed by core operational AI that delivers measurable cost or margin impact, and finally strategic and generative AI initiatives that redefine business models. BCG’s portfolio sequencing methodology has been applied successfully across MENA financial services and energy sectors, reducing implementation risk by 40 percent compared with parallel high-risk programmes.

# Fostering Adoption

**Technology deployment without adoption is expenditure without return.** MENA enterprises face particular adoption challenges: workforce heterogeneity, varying digital literacy levels, and cultural norms around authority and risk.

### Leadership Literacy First

AI adoption must start at the top. Before frontline training commences, boards and C-suite executives need practical literacy — not technical depth, but sufficient understanding to ask informed questions, allocate resources confidently, and resist vendor hype. Accenture’s 2026 Technology Vision found that enterprises investing in executive AI literacy before widespread deployment achieve adoption rates 2.3 times higher than organisations that skip this step.

### Middle-Management Engagement

In hierarchical MENA organisations, middle management acts as a cultural filter. If this cohort perceives AI as a threat to status, control, or team size, adoption stalls regardless of technical readiness. The CoE should involve managers as co-designers from the outset — integrating their operational knowledge into use-case design and positioning AI as a tool that amplifies, rather than replaces, team performance.

### Localisation of Interfaces

Arabic language support, right-to-left interfaces, and culturally relevant examples dramatically increase user comfort. Enterprises deploying English-only AI tools in Arabic-dominant business units report adoption rates 20 to 30 percent lower than those providing localised experiences.

### Gamification and Recognition

Incentive structures shape behaviour. Public recognition of AI achievement teams, internal innovation challenges tied to AI, and career pathways for AI-certified employees create motivational momentum. The CoE should work with human resources to embed AI competencies into professional development frameworks and succession planning criteria.

# Measuring Impact

**What gets measured gets managed — and what gets managed well gets funded.** The CoE requires a measurement architecture that balances short-term execution metrics with long-term strategic indicators.

### Leading Indicators

Talent metrics — number of certified graduates, internal community engagement scores, and vacancy fill rates for AI roles — reflect capability development. Process metrics — governance review cycle times, model approval throughput, and data quality scores — reflect operational maturity. These leading indicators should be reviewed monthly and reported to the CoE steering committee.

### Lagging Indicators

Revenue metrics — AI-attributable revenue, cross-sell ratios from AI-driven personalisation, and cost avoidance from automated decisions — demonstrate commercial value. Efficiency metrics — reduction in manual processing hours, decreased defect rates, and accelerated time-to-market for AI-supported products — demonstrate operational transformation.

Combined scorecard models, adapted from PwC’s responsible AI frameworks, provide the most balanced view. They prevent CoE performance from being judged solely on the volume of pilots launched — a metric that can be gamed — while ensuring that strategic value is not lost in vague qualitative assessments.

### Benchmarking

Anonymised peer benchmarking is powerful. Enterprise AI CoEs that share operational data across sector cohorts can identify performance gaps and adopt proven practices faster than those relying solely on internal reflection. Gartner’s communities of practice research demonstrates that benchmarked organisations achieve capability maturity 18 to 24 months ahead of isolated peers.

# MENA Roadmap

**A typical AI CoE maturity journey in MENA unfolds across three to five years.** Enterprises should sequence investments according to their regulatory environment, talent availability, and strategic urgency.

### Phase 1: Foundation (Months 0–12)

Establish governance framework and policy. Recruit core CoE leadership. Conduct enterprise data and capability maturity assessments. Launch initial executive literacy programme. Select and deploy core platform infrastructure. Deliver two to three high-visibility pilot use cases that demonstrate value and build internal credibility.

### Phase 2: Scale (Months 12–24)

Expand talent pipeline through certification and university partnerships. Deploy shared data platforms and tooling standards across enterprise units. Implement governance automation — model registries, algorithmic auditing workflows, and compliance dashboards. Scale successful pilots into production across additional business units. Establish formal vendor management and commercial negotiation capabilities.

### Phase 3: Optimise (Months 24–36)

Introduce advanced analytics and generative AI programmes. Launch enterprise AI strategy reviews aligned to annual planning cycles. Embed AI KPIs into business unit scorecards. Develop advanced change management and adoption measurement capabilities. Engage external regional and global ecosystems for strategic partnership and talent pipelines.

### Phase 4: Sustain (Months 36+)

Achieve continuous AI operating model maturity. Maintain CoE as an internal consultancy and innovation laboratory. Refresh technology stack and governance in response to regulatory evolution. Mentor other enterprises through regional knowledge-sharing networks.

**Transitioning from project delivery to platform capability.** In the sustainment phase, the CoE shifts from executing discrete use cases to managing a living AI operating system. This involves continuous platform maintenance, model monitoring, re-training cycles, and governance updates. It also requires the CoE to act as an internal venture fund — allocating experimentation budgets, reviewing project portfolios, and making go/no-go decisions with disciplined stage-gate criteria.

**Institutionalising AI ethics and responsible AI.** As AI deployments mature, ethical considerations move from aspirational statements to operational requirements. The CoE develops regionally relevant ethical AI frameworks, conducts regular algorithmic impact assessments, and ensures that AI outputs remain explainable and contestable — particularly in domains such as credit scoring, recruitment, and public-service delivery where life outcomes are directly affected.

**Building MENA AI talent ecosystems.** The most advanced CoEs actively shape regional talent markets. Through university partnerships, internship programmes, public research contributions, and international researcher exchanges, leading organisations transform from talent consumers into talent creators — reducing external dependency and contributing to broader regional AI capability development.

The roadmap should be treated as a living reference, revisited quarterly against actual delivery velocity, leadership commitment, and external regulatory or competitive changes.

—

*Organisations commencing their AI Centre of Excellence journey should begin with a candid readiness assessment: data estate evaluation, leadership literacy audit, talent gap analysis, and regulatory risk review. This diagnostic forms the evidence base for mandate design, investment sequencing, and stakeholder communication — ensuring that the CoE is built on operational reality rather than aspiration alone.*

More From APH Insights

Need to talk sooner?

Book a Discovery Call

Thank You!