GCC companies are all-in on AI - but scale is still out of reach
AI pilots are everywhere in the GCC. A new report shows 84% of companies have adopted AI in at least one function. Yet only 31% have managed to scale it across the business. That gap is where value is leaking.
Budgets aren't the problem. Ambition isn't the problem. Converting intent into repeatable outcomes is.
What the survey says
McKinsey surveyed senior leaders across the GCC. Adoption is up sharply from 64% in 2023 to 84% this year. But scaling is stuck: fewer than a third have moved beyond pilots into enterprise deployment.
There are heavyweight bets underway. One example: a 250-billion-parameter model trained on decades of operational data at a major energy company. Partnerships and internal builds are rising across sectors. The issue isn't activity - it's traction.
Where value is showing up today
- Service operations: Automating routine tasks, speeding resolution, and improving consistency.
- Marketing and sales: Lead scoring, segmentation, content generation, and next-best-action.
- Product and service development: The fastest-growing area since 2023, with new AI features and enhanced offerings.
Why scale breaks down
- Strategy isn't tied to value pools: Teams chase interesting use cases instead of measurable P&L impact.
- Delivery muscle is thin: Gaps in data engineering, MLOps, and domain-savvy talent stall progress.
- Change is underfunded: Workflows don't change, incentives don't change - so behaviors don't change.
- Capital and data hurdles: Automating recommendations often needs new equipment, clean data, and reliable infrastructure. Many firms lack one or all three.
The blueprint to scale
- Accountability to outcomes: Every AI initiative maps to a KPI (revenue, cost, cycle time, risk). No KPI, no build. Scale only after targets are hit in production for a sustained period.
- AI-driven development: Use AI to build AI - code generation, test creation, documentation, and workflow scaffolding. Shorten cycle time, then reinvest saved time into higher-value backlog.
- Workflow redesign: Don't bolt AI onto old processes. Redesign steps, roles, and handoffs so the model sits in the critical path - not as a sidecar.
Risk guardrails executives expect
- Cybersecurity: The top concern. Treat models, prompts, and feature stores as sensitive assets. Apply least privilege and continuous monitoring.
- Accuracy and reliability: Guard against errors and hallucinations with retrieval, grounding, human-in-the-loop, and clear fallbacks.
- Privacy: Minimize personal data use, set retention rules, and log access. Align with policy and regulation by design.
- Data quality: Fix missing values, bias, outliers, and lineage. Reliable data is the lifeblood of dependable AI outputs.
A 90-day plan to move beyond pilots
- Week 1-2: Identify three value pools with CFO sign-off (e.g., claim cycle time, sales conversion, inventory turns). Set target KPIs and thresholds to scale.
- Week 3-4: Appoint accountable owners (business lead, product manager, data lead, risk lead). Create a single backlog per value pool.
- Week 5-6: Run a data quality sprint. Define gold datasets, improve coverage, and establish monitoring for drift and bias.
- Week 7-8: Make architecture choices explicit (build vs. buy, model class, retrieval, observability, MLOps). Standardize patterns to avoid bespoke one-offs.
- Week 9-10: Redesign the workflow. Remove steps, reassign decisions, define guardrails, and bake AI into the core process.
- Week 11-12: Ship to production, measure against KPIs, and decide go/no-go for scaling. If targets are hit, templatize and replicate in the next business unit.
What to track (so scale doesn't stall)
- Value: Incremental revenue, cost per transaction, cycle-time delta, error rates, and realized savings vs. plan.
- Adoption: Percentage of decisions made with AI assistance, user NPS, and time-in-tool.
- Reliability: Model accuracy, grounding rate, escalation rate, and data drift.
- Risk: Security incidents, privacy exceptions, and policy adherence.
- Throughput: Lead time from idea to production, and weekly release cadence.
What's next
The focus has shifted from single use cases to agents that can plan, call tools, and complete multi-step work. That makes workflow redesign and guardrails even more important. The winners will pair bold bets with disciplined operating mechanics.
If you're building executive capability to lead this shift, explore curated learning paths by role at Complete AI Training.
Your membership also unlocks:
