EY: the AI pilots that stall were never built to scale
Enterprise AI has reached the point where the interesting question is no longer whether a model works, but whether anything downstream of it does. Across the Middle East and North Africa most large organisations now have pilots running. Comparatively few of them have become systems the business depends on.
Mazen Baroudi is MENA Chief Technology Officer and Technology Consulting Leader at EY, and is at LEAP 2026 to discuss what happens as frontier AI moves out of experimentation and into enterprise-wide use. In a written interview with The Datatech Times he set out where the work actually stalls, what clients can and cannot yet prove, and what would have to change for the region to build frontier technology rather than buy it.
His answer to why pilots stall is not about model quality. "Across MENA, the single most common reason is that pilots are built to prove a use case, not to run inside a business process," he said. He cited EY research finding that nearly nine in ten senior leaders report significant barriers to agentic AI adoption, and argued that the obstacle is structural rather than technical. The pilot, he said, "was never connected to a system of record, an owner, or a measurable process outcome. It stays a demo."
That framing shapes his account of what organisations can currently prove. Adoption and efficiency can now be measured with growing confidence: "usage rates, hours saved, cycle-time reduction and cost per transaction". What remains an article of faith is whether any of it compounds. "What many are still taking on faith is the compounding business value, specifically whether productivity gains translate into revenue growth, better decisions or sustained competitive advantage over time," he said.
EY has run the experiment on itself. Baroudi described the firm's Client Zero programme, which rolled out AI tools across its own workforce, and was candid about what it did and did not settle. "Early wins in hours saved are real and measurable," he said. "A more challenging metric is growth, that takes longer to prove." Governance and quality measures such as accuracy and safety are becoming easier to capture, he added, while board-level attribution of return on investment "continues to mature across the region".
On where the money goes, he was clear that it is not the models. "In practice, the cost rarely sits with the models themselves, as those continue to become more affordable," he said. The investment sits instead in "cleaning, connecting and governing enterprise data so AI can act on trustworthy sources of truth", in the engineering needed to integrate agents into existing systems of record, and in upskilling, change management and the new roles created to oversee what AI produces. He tied this to EY's Value Blueprints approach, which he said "treats data, systems of record and workforce as equal architectural tiers alongside AI models", on the basis that underinvesting in any one of them makes scaling more expensive later.
What happens when governance runs late
Baroudi's account of governance is the most concrete part of the interview, because he describes a failure mode rather than a principle. Where responsible AI, security and risk are run as parallel workstreams instead of being designed in from the outset, he said, "we typically see the same pattern: rapid initial pilots followed by a slowdown, as legal, security or risk teams raise concerns only when a system approaches production". Retrofitting controls "is slower and more expensive than designing them in from the beginning", and organisations that skip the step "often face longer approval cycles or systems that remain in pilot indefinitely because the necessary confidence for scaling has not been established".
Asked for a regional example of AI creating public value rather than an ambition for it, he pointed to document and case-processing automation in government services, which he said is "delivering measurable results today rather than remaining a future ambition". His qualification is about scope. Solutions built for a specific document type or process step, rather than an attempt to digitise a whole department at once, are "far more likely to reach production". Applied to permits, benefits and licensing, he said, they are "already reducing turnaround times and freeing staff to focus on judgment-based work".
Open ecosystems and the question of who builds
On open source, Baroudi described a shift away from the framing itself. Large organisations are "moving away from an 'open versus closed' debate and instead adopting a governance-led ecosystem approach", he said, setting standards for security, model evaluation, data access, monitoring and accountability while still allowing teams to experiment. "The key shift is architectural," he said. Platforms are built so that different models can be tried while control over risk, compliance and data governance stays central, which he argued reduces dependence on any single vendor. "In practice, the winners are not those choosing one technology path. They are those building the capability to safely orchestrate multiple technologies within a consistent governance framework."
The last question was whether a region often described as an adopter of AI built elsewhere can become somewhere frontier technology is deployed at scale first. Baroudi listed the assets he believes are already in place: ambitious national strategies, significant investment capacity, world-class infrastructure and strong public-sector leadership. Turning those into "globally competitive AI ecosystems" requires continued investment in talent, research, frontier computing capacity and locally relevant data assets, he said, along with closer work between governments, academia, technology companies and investors to accelerate commercialisation.
He did not argue for replication. "The opportunity is not simply to replicate technologies developed elsewhere," he said, naming Arabic-language AI, smart cities, energy, logistics, healthcare and digital government as the domains where the region holds distinctive strengths. EY has not published a breakdown of how many of its regional engagements have moved from pilot into production, and the figure on adoption barriers is the firm's own research rather than an independent survey.