AI Operating Model — Enterprise.AI Accelerator

Operating Model Archetypes

Five approaches to organising AI capability. Click each to explore fit, trade-offs, and Gulf-specific considerations.

Hub-and-Spoke

Recommended

Central AI CoE sets standards, tooling, and governance. Business units run their own AI squads with CoE oversight and shared platform.

Best for: Gulf banks at AI maturity Level 2–3 with 3+ active use cases across divisions. This is the model we recommend for most SAMA/QCB-regulated institutions.

Strengths

Balances speed with control
Regulatory accountability is clear (CoE owns governance)
BUs retain domain expertise and agility
Scales naturally as maturity grows

Watch-outs

Requires strong CoE leadership
BU–CoE coordination overhead
Risk of shadow AI if CoE is too slow
Needs executive sponsorship to prevent turf wars

Centralised CoE

Single Centre of Excellence owns all AI development, deployment, and governance. Business units submit requests; CoE prioritises and delivers.

Best for: Institutions just starting AI (maturity Level 1). Common in Saudi banks with <3 AI use cases and limited internal talent.

Strengths

Simple accountability
Consistent standards from day one
Easier to staff (one team)
Clear regulatory reporting line

Watch-outs

Bottleneck as demand grows
Slow turnaround for BU requests
CoE disconnected from business context
Does not scale past 5–8 use cases

Federated

Each business unit builds and runs its own AI team. No central coordination — each division moves independently.

Best for: Large diversified groups (e.g. holding companies) where business units have distinct regulatory regimes and data domains.

Strengths

Maximum business unit agility
Deep domain alignment
No central bottleneck
Fast experimentation

Watch-outs

Duplicate infrastructure and spend
Inconsistent governance — regulatory risk
No knowledge sharing across units
Hard to enforce SDAIA/SAMA standards uniformly

Platform Model

Advanced

Central team builds and operates a self-service AI platform (MLOps, model registry, feature store). Business units consume the platform to build their own solutions.

Best for: Mature institutions (Level 3–4) with 10+ models in production and established data engineering. Rare in the Gulf today but the target state for 2028.

Strengths

Highest scalability
Consistent tooling and compliance
BUs self-serve — no bottleneck
Built-in model governance via platform

Watch-outs

Heavy upfront investment
Needs mature data infrastructure
Platform team must be world-class
Risk of over-engineering before product-market fit

Embedded

AI talent is embedded directly in every business and functional team. No separate AI organisation — AI is everyone's job.

Best for: Highly AI-mature organisations (Level 4–5) where AI is a core competency, not a support function. No Gulf bank is here yet.

Strengths

AI fully integrated into business
Fastest decision-to-deployment
No organisational friction
AI literacy is universal

Watch-outs

Requires massive talent pool
Governance must be embedded in culture
Hard to maintain consistency
Premature adoption creates chaos

12 Key Roles

The critical capabilities required for an enterprise AI operating model. FTE counts assume a mid-sized Gulf bank (3,000–8,000 employees).

Role	Reports To	Core Responsibility	FTEs
Chief AI Officer	CEO / CTO	AI strategy, board reporting, regulatory engagement	1
AI Strategy Lead	Chief AI Officer	Roadmap, prioritisation, business case development	1–2
AI Product Owner	Business Unit Head	Use case scoping, requirements, adoption measurement	3–5
ML Engineer	AI Platform Lead	Model development, fine-tuning, prompt engineering	4–8
Data Engineer (AI)	CDO / AI Platform	Feature engineering, data pipelines, vector stores	3–6
Model Risk Manager	CRO	SR 11-7 compliance, model validation, bias testing	2–3
AI Governance Lead	Chief AI Officer	Policy, ethics, SDAIA/SAMA compliance, audit readiness	1–2
MLOps Engineer	AI Platform Lead	CI/CD for models, monitoring, infrastructure	2–4
AI Change Manager	CHRO / Chief AI Officer	Adoption, resistance management, process redesign	1–2
AI Business Translator	AI Strategy Lead	Bridge technical teams and business stakeholders	2–3
AI Literacy Lead	L&D / Chief AI Officer	Training programmes, upskilling, AI champions network	1–2
External AI Advisor	Board / Chief AI Officer	Independent challenge, benchmarking, regulatory guidance	Retained

Total core team: 22–40 FTEs for a mid-sized Gulf bank. Most institutions start with 8–12 and scale as the pipeline grows.

Governance Operating Rhythm

AI governance is not a document — it is a rhythm. These five cadences keep models in production, regulators satisfied, and the board informed.

Daily

Model performance monitoring
Incident triage & escalation
Drift detection alerts

Weekly

AI sprint review
Use case pipeline triage
Platform health check

Monthly

AI Steering Committee
Model risk review
Adoption metrics review

Quarterly

Board AI report
Strategy alignment review
Regulatory update

Annual

AI strategy refresh
Full regulatory assessment
Operating model review

AI Delivery Pipeline

Six stages from idea to production. Click each stage to see ownership, artefacts, and governance gates.

01

Ideation

Capture & qualify

02

Prioritisation

Score & rank

03

Development

Build & iterate

04

Validation

Test & approve

05

Deployment

Ship & integrate

06

Monitoring

Observe & govern

Stage 1: Ideation

Business units, front-line staff, and leadership submit AI use case ideas through a standardised intake form. The AI Strategy Lead qualifies each idea against strategic fit, data readiness, and regulatory feasibility.

Owner

AI Strategy Lead + Business Unit Heads

Key Artefacts

Use case intake form, initial feasibility screen, data availability assessment

Gate

Strategic alignment check — does this map to a board-approved AI priority?

Stage 2: Prioritisation

Qualified use cases are scored on value (revenue impact, cost savings, risk reduction), feasibility (data readiness, technical complexity, talent availability), and risk (regulatory sensitivity, model risk tier, ethical exposure). The AI Steering Committee approves the quarterly pipeline.

Owner

AI Steering Committee

Key Artefacts

Prioritisation matrix, business case, resource allocation plan

Gate

Steering Committee sign-off with budget and timeline commitment

Stage 3: Development

Cross-functional squad (AI Product Owner, ML Engineers, Data Engineers, Business Translator) builds the solution in 2-week sprints. Follows responsible AI principles with bias testing and explainability checks built into the sprint cycle.

Owner

AI Product Owner + ML Engineering Lead

Key Artefacts

Model documentation (SR 11-7), test results, data lineage map, sprint demos

Gate

Technical review — model performance thresholds met, code reviewed, security cleared

Stage 4: Validation

Independent Model Risk Management team validates the model against SR 11-7 / SS1/23 requirements. Includes bias testing, stress testing, explainability review, and SDAIA AI Ethics assessment. This is the regulatory gate.

Owner

Model Risk Manager + AI Governance Lead

Key Artefacts

Validation report, bias audit, explainability assessment, risk tier classification

Gate

Model Risk Committee approval — mandatory before production deployment

Stage 5: Deployment

MLOps team deploys the validated model to production via CI/CD pipeline. Includes integration with enterprise systems, access controls, audit logging, and rollback procedures. Change management activates user adoption plan.

Owner

MLOps Engineer + AI Change Manager

Key Artefacts

Deployment runbook, rollback plan, user training materials, SLA definition

Gate

Production readiness review — monitoring in place, SLAs defined, rollback tested

Stage 6: Monitoring

Continuous performance monitoring against KPIs and drift thresholds. Monthly model health reviews. Annual revalidation cycle. Incident response protocol for model failures or ethical breaches.

Owner

MLOps Engineer + Model Risk Manager

Key Artefacts

Performance dashboard, drift reports, incident logs, revalidation schedule

Gate

Quarterly performance review — retire, retrain, or continue decision

90-Day Implementation Roadmap

Stand up a functioning AI operating model in three phases. Not theory — a sprint plan with deliverables, owners, and checkpoints.

Days 1–30

Assess & Design

Map the current state and design the target operating model.

AI maturity assessment across all business units
Current-state capability mapping (talent, data, infrastructure)
Select operating model archetype (Hub-and-Spoke for most Gulf banks)
Define the 12 key roles — who exists, who needs hiring, who can be redeployed
Draft governance charter and committee terms of reference
Align with SAMA/SDAIA/QCB requirements

Days 31–60

Build & Staff

Stand up the team, platform, and governance structures.

Appoint Chief AI Officer (or interim Head of AI)
Hire or redeploy core team (target 8–12 FTEs for Phase 1)
Set up AI platform environment (cloud, MLOps tooling, model registry)
Publish AI governance policy and responsible AI principles
Establish AI Steering Committee with quarterly cadence
Launch use case intake process and prioritisation framework

Days 61–90

Operationalise

Run the first use case through the full pipeline. Prove the model works.

Select one high-value, low-risk use case as the proof point
Run it through all 6 pipeline stages end-to-end
Activate governance rhythm — first Steering Committee meeting
Publish first board AI report
Launch AI literacy programme (board + ExCo first)
Measure and report on all 6 success KPIs

Success Metrics

Six KPIs that tell you whether your AI operating model is working. Track from Day 90 onward.

<12w

Time-to-Production

Idea to deployed model. Target: under 12 weeks for standard use cases.

100%

Model Risk Coverage

% of production models with completed SR 11-7 validation. Must be 100%.

>60%

BU Adoption Rate

% of business units with at least one AI use case in production.

<$150K

Cost per Use Case

Fully loaded cost to deliver one AI use case to production.

0

Regulatory Findings

Number of supervisory findings or MRAs related to AI/ML models.

>70%

AI Literacy Score

% of ExCo and senior management completing AI literacy programme.