Article | September 16, 2026
By Hari Prasad Govindarajan, Chief AI & Transformation Officer, DXC Technology UK&I
Two numbers should make every fintech board uncomfortable: 75% of UK financial services firms already use AI, according to a joint survey by the Bank of England and the FCA. Yet MIT researchers found that 95% of enterprise generative AI pilots deliver no measurable P&L impact. Nearly everyone is adopting; almost nobody is converting.
That gap was the unspoken theme running through FinTech North in Newcastle, and closing it has remarkably little to do with technology.
In my keynote, I called this pattern the POC trap. Pilots fail for four reasons, none of which has anything to do with the model. Fragmented data estates cannot power intelligent systems, while the lack of an operating model leaves experiments disconnected from the processes that run the business. At the same time, a governance vacuum leaves no clear ownership of AI decisions and outcomes, and chronically underfunded change management allows cultural resistance to take hold.
The research points to the same diagnosis. Gartner predicts that more than 40% of agentic AI projects will be cancelled by the end of 2027, driven by escalating costs, unclear value and inadequate risk controls, rather than technical failure. McKinsey's latest State of AI research finds that only about 6% of organisations qualify as AI high performers. Pilots don't die at the point of invention. They die at the point of integration.
The event's opening keynote put it best: innovation creates potential; trust creates adoption; and only adoption creates growth. The question that matters, the speaker argued, is not how intelligent AI will become, but how much authority we’re prepared to delegate to it.
Published just days before the event, the FCA's Mills Review describes financial services shifting from human-led, episodic interactions to AI-enabled, continuous and delegated ones. Its research also shows that just 20% of consumers (around 11 million UK adults) would let AI act autonomously on their behalf. Together, the message is clear: the technology is ready to act for customers, but four in five customers are not ready to let it.
The panel brought this tension to life. One fintech co-founder observed that a portable, tokenised digital ID could remove enormous friction from onboarding today, yet we would “be lucky to see it adopted in ten years, because of trust”. Firms still won't rely on each other's verification.
The same keynote made a distinction I would put on every boardroom wall: technology is not moving faster than regulation — it is moving faster than governance. Governance encompasses liability, accountability, standards and interoperability. Waiting for a rulebook to resolve those questions is a category error.
This is why I frame the discipline as AI resilience, the successor to the quantum resilience we spent the last decade building, but with one crucial difference: the adversary has already arrived. The scaffolding exists. The EU AI Act's transparency obligations apply from August 2026, with high-risk duties following in 2027 and 2028, while DORA already treats technology risk as a matter of operational resilience. NIST's AI RMF provides a practical framework and ISO/IEC 42001 offers a certifiable structure. The stakes are quantifiable, too: Deloitte projects that US losses from generative-AI-enabled fraud will reach $40 billion by 2027, up from $12.3 billion in 2023.
In practice, that means treating foundation models as critical suppliers, red-teaming agents as you would production systems and building kill switches and non-AI fallback paths. That is not compliance overhead. It’s what makes delegation trustworthy, which, in turn, is what makes adoption possible.
Scaling AI redraws the organisation itself. Knowledge businesses have always been structured as pyramids, with a wide junior base providing leverage. Agents are beginning to absorb much of the work performed at the base. Stanford's Digital Economy Lab has already measured a 13% relative employment decline among 22- to 25-year-olds in occupations most exposed to AI. This shift is turning pyramids into diamonds before a new kind of pyramid emerges with a digital base. A speaker discussing the northern fintech economy asked precisely the right question, “If AI takes the junior jobs, where does succession come from?” Almost nobody has budgeted for the answer.
Integration beats layering, too. A Sage executive on the panel cited the company's survey of 1,500 UK small businesses, which found that they use an average of 58 applications. The answer is not to add “an AI layer on top” of that sprawl, but to make the underlying systems work together. McKinsey's data points to the same conclusion in analyst language: workflow redesign, rather than tool adoption, is what correlates with bottom-line impact.
The next 12 months will shift the conversation from adoption to super-orchestration: fleets of specialised agents working across CRM, ERP and core platforms, with an orchestration layer routing each task to the cheapest capable model. The economics are compelling. Stanford's AI Index shows inference costs have fallen more than 280-fold in 18 months. But none of this matters without the unglamorous foundations: data readiness, an operating model, governance that earns delegation and a workforce plan for the diamond-shaped organisation.
So, I have one clear ask: Stop counting pilots and start counting the decisions your customers, and your regulator, trust your AI to make in production. Then pick a single governed workflow this quarter and take it all the way through.
Hari Prasad Govindarajan is Chief AI & Transformation Officer, DXC Technology UK&I. Connect with Hari on LinkedIn.
