EY: How to Scale AI at Speed Without Impacting Innovation

Almost nine in 10 employees now use AI at work, yet only 28% of organisations are positioned to turn that activity into high-value outcomes, according to EY’s Work Reimagined Survey.

As organisations accelerate AI implementation across the enterprise, many face the same dilemma: move fast and accept fragmentation, or slow down in the name of control.

In practice, however, the organisations struggling most with AI at scale are often held back by the foundations surrounding it. 

Here, Dan Diasio, Global Consulting AI Leader at EY, discusses why architecture has become a key differentiating factor in scaling AI effectively without slowing innovation.

How has AI turned architecture into a key driver of speed?

AI's speed advantage comes from how you architect decisions. The organisations pulling ahead have built systems that know when to act, when to escalate and when human judgment is non-negotiable. They've designed trust into the architecture: clear thresholds for autonomy, explicit escalation paths and accountability frameworks that work at scale. This is human-AI orchestration for continuous delivery: systems that move at machine speed while preserving human oversight where it creates the most value.

Most organisations are stuck in what I refer to as the “model trap” where they focus on tuning accuracy, chasing marginal gains and assume a better model will unlock velocity. Speed is limited by the questions you haven't answered: who decides when the system can act alone? What triggers human review? How do you scale judgment without becoming the bottleneck? Companies that architect for these answers deliberately, embedding decision rights and trust protocols from the start, operate with a velocity and confidence that others can't match. This competitive gap really is structural.

Why does traditional AI governance slow teams down?

Traditional AI governance often slows delivery because it was built for a different era: heavy, process-centric and designed to control static systems. As AI evolves faster, that same structure accumulates friction and protection turns into drag.

The answer is smarter governance. Rigid models need to shift to lighter, adaptive guardrails that evolve with the technology. Clear ownership replaces committee paralysis; risk-based escalation protocols replace blanket approvals; accountability ties to outcomes, not layers.

Done well, governance speeds decisions while preserving trust. Teams feel empowered, adoption accelerates, delivery improves and governance becomes an enabler of responsible momentum. The goal is responsible velocity with systems that move fast because governance is designed from the jump – not tacked on.

In what ways is AI architecture about shaping people’s behaviour, not just technology?

At its core, AI architecture provides the structural foundation that enables AI to be deployed, managed and scaled across the enterprise. But structure alone does not determine success. When human factors are treated as first-class design inputs, the likelihood of meaningful value creation rises sharply. Being “data-driven” is insufficient if the context behind decisions (the why and the trade-offs) remains locked in individuals’ heads rather than captured and embedded into workflows. In this sense, AI architecture is not purely technical, it is about creating the conditions in which AI can effectively support both the enterprise and the humans within it.

Research with Oxford Saïd reinforces this view, showing transformations are 12 times more successful when leaders prioritise human-centred approaches. As AI-centric architectures mature, humans are increasingly working alongside autonomous AI agents, reshaping expectations around trust, accountability and decision-making. EY.ai Value Blueprints positions AI architecture as a foundation for shaping organisational habits, workflows and culture from the ground up.

Why is lost momentum the biggest hidden risk of fragmented AI – and how can leaders prevent it?

Losing momentum in an enterprise AI rollout doesn’t just slow progress. It creates fragmentation. Many organisations race to scale AI for quick returns, but speed without coherence comes at a cost. Real value emerges when AI is built as an enterprise capability, not a patchwork of local solutions. That’s how transformation becomes structural, not temporary and siloed across functions.

Achieving this kind of scale takes more than technology. It requires vision, leadership and a shared sense of purpose – an enterprise-wide mindset. When that energy fades, teams revert to familiar habits, AI initiatives splinter and 'bring your own AI' (BYOAI) fills the gaps. Over time, these shortcuts compound into fragmented AI that’s harder to govern, trust and scale.

At EY, we believe the opportunity is bigger than improving individual models. It’s about designing the organisation of the future, one where autonomous systems are embedded into how work actually happens. Align the toolset, skillset and mindset, and AI becomes part of daily operations, not an add-on.