Recap: The AI gap is widening and most organisations are on the wrong side of itRecap: The AI gap is widening and most organisations are on the wrong side of itRecap: The AI gap is widening and most organisations are on the wrong side of it

In the latest edition of our Impact Talks webinar series, Nearform CEO Ciaran Cosgrave convened leaders from McKinsey & Company, Sun Life, and Nearform to explore why so few organisations are turning AI adoption into meaningful business value, and where the real blockers sit.

Discussion topics included:

• What the organisations getting the most value from AI are doing differently
• Lessons from redesigning business processes and operating models inside a regulated enterprise
• Why the biggest challenge is no longer the technology itself, but how organisations design, build, and deliver AI-enabled products

The featured speakers were Ben Dimson (Partner at McKinsey & Company, who focuses on AI strategy and organisational transformation), Pierre Peiclier (Chief Information & Digital Innovation Officer at Sun Life US IT, who leads the Dental IT organisation and drives digital innovation across Sun Life U.S.), and Colin Houlihan (Chief Strategy & AI Officer at Nearform, who moves AI programmes from strategy into production).

As the AI execution gap widens, these luminaries posed a critical question: which side of it are you on?

88% of organisations are using AI. Only one percent consider themselves mature enough to actually drive meaningful value from it. That striking gap, drawn from a recent McKinsey survey, formed the centrepiece of Nearform's latest Impact Series webinar, "The AI Gap Is Widening," which convened three senior leaders to make sense of why so much investment is producing so little return.

Cosgrave, who moderated, framed the challenge directly from the outset: the problem isn't the technology. The models exist. The tools are there. Something else, whether organisational, strategic or cultural, is holding most enterprises back. With a live audience poll confirming that roughly 75% of attendees were still in an experimentation or piloting phase, the panel had no shortage of material to work with.

Dimson brought a cross-industry vantage point grounded in research across 200 organisations that had successfully scaled AI adoption, findings which are codified in McKinsey's Rewired framework. His diagnosis was unambiguous: the organisations pulling ahead aren't doing something technically different. They're doing something organisationally different. The CEO is personally accountable. AI is tied explicitly to value creation, such as revenue, cost and valuation, rather than treated as a technology initiative handed off to the tech function. And the focus is disciplined. Too many organisations fall into what Dimson called a Goldilocks problem: either deploying AI too broadly (giving everyone Copilot access with no way to measure results) or chasing point solutions that fix isolated pain points without moving the needle. The sweet spot is identifying a defined set of domains — typically ten to fifteen, with two or three prioritised for speed and two or three treated as strategic "lighthouse" bets — where AI can genuinely reshape outcomes.

From there, Peiclier offered the perspective of someone navigating exactly these challenges inside a complex, heavily regulated enterprise serving millions of customers. He described three distinct layers where AI can create value: enabling companies to do things that were previously out of scope entirely; driving team-level efficiencies; and amplifying individual employee capabilities. But he was candid that most organisations are still in the worst years of implementation, trying to overlay AI on existing processes and job definitions rather than reimagining them from scratch. The more productive frame, in his view, is to think of AI delivery not as shipping a solution but as delivering a slice of the operating model itself. That shift in framing has significant consequences for how teams are structured, how value is measured, and how enterprises move past what he called "experimentation fatigue" toward initiatives of genuine strategic significance.

On data, which was consistently ranked as a top barrier in the session's second audience poll, Peiclier offered a counterintuitive take: the goal shouldn't be perfect, comprehensive data stores. It should be leaner, more contextually connected data — what he called a "context store" — built specifically for AI solutions and tightly linked to the business domain it's meant to serve. Less data, more reliably connected, produces more predictable model outputs.

This point of view was complemented by Houlihan, who brought a ground-level perspective from working daily with engineering and AI teams actually executing these transformations. He echoed the importance of disciplined selection. The organisations that appear to have stumbled upon the magic use case have, in reality, done the unglamorous work of matching problems to technology, stress-testing business cases before scaling, and bringing compliance and security into the process from day one rather than as an afterthought. On the question of AI costs, with one audience member asking about forecasts of $1.5 trillion in AI spend by end of 2026, Houlihan was measured: token costs are rising, but organisations are already responding with greater efficiency, exploring alternative models, and rethinking architecture. Peiclier added that smarter context design can materially reduce token consumption, and that the infrastructure picture will look very different once value is proven and the optimisation phase begins.

The session closed with a note of urgency yet optimism: Gartner is predicting that 40% of agentic AI projects will be cancelled by the end of 2027 as organisations struggle with costs and misaligned foundations. The antidote, the panel agreed, is CEO-level ownership, genuine strategic grounding, and a willingness to learn from what's already working, rather than continuing to experiment for its own sake.

For enterprise leaders trying to close the gap between AI ambition and AI impact, the conversation made one thing clear: technology is no longer the hard part.

Nearform's AI-native engineering sessions offer structured 90-minute deep dives for enterprise engineering leaders on what's working in production AI today. Get in touch to find out more.