The opinions expressed here are those of the authors. They do not necessarily reflect the views or positions of UK Finance or its members.

Cloud infrastructure, data platforms, pre-trained models, specialist talent: all available to any institution willing to invest. What differs between banks scaling AI and those cycling through pilots is not what they have deployed. It is how their leaders behave and whether AI has been integrated into the operating model or merely assigned to a data science function.

Those factors are harder to replicate than technology and take longer to build. And under SMCR, they carry accountability implications that are only now beginning to surface in supervisory conversations.

At a recent Gartner event in London, one assertion landed with particular force: agentic AI will not work until C-level executives know how to apply it. Not deploy it. Not approve it. Apply it. That distinction is the subject of this article.

The board has become the AI risk most institutions have not modelled

UK boards have grown comfortable with technology risk: cyber resilience, operational disruption, vendor dependency. AI introduces something different. Systems influencing credit decisions, customer outcomes, and fraud investigations in real time require oversight that most governance frameworks were not written to provide.

The FCA and PRA have been clear that SMCR accountability extends into AI. Boards need the literacy to interrogate how AI systems are governed and what the response would be if something went wrong at scale.

Most boards do not yet have that literacy. That is a material risk. And regulators are beginning to notice.

The most effective boards are establishing AI oversight mechanisms, setting explicit risk tolerances, and assigning clear SMCR accountability before a poor outcome forces the issue.

Leadership is where the real decisions are made

Several European banks ran agentic AI pilots in 2025 and 2026. In each case the technology performed as expected. The institutions that moved to deployment shared one characteristic: a senior sponsor had defined the governance boundaries, articulated the business value, and created a clear path for board-level approval. Those without that sponsorship produced impressive demonstrations that went nowhere. The constraint was not capability. It was board readiness.

Three behaviours that separate scaling institutions from the rest

Three leadership behaviours appear consistently in institutions that have moved beyond pilots. None are about technology choices.

They define decision rights explicitly. Business units can deploy AI within agreed risk boundaries without requiring central approval at every step. This reduces bottlenecks without removing accountability. The absence of clear decision rights is the single most effective way to ensure good AI programmes die in committee.

They fund capabilities rather than projects. Investment in shared infrastructure, a common deployment environment, or an enterprise governance framework pays back across every subsequent use case. Institutions that fund projects start from scratch every time. Those that fund capabilities compound their returns.

They treat unexpected outcomes as learning rather than failure. Institutions that create safety for teams to surface problems early are significantly more resilient than those that punish honesty. In AI, where model behaviour in production frequently surprises even the teams that built it, this is an operational necessity, not a cultural nicety.

The question is not whether your bank has the technology

Most institutions can point to a capable data platform, a growing portfolio of AI models, and a team of data scientists. What they cannot always point to is an executive team that has genuinely changed how it makes decisions, funds capability, and governs AI as a core business function rather than a technology initiative. 

Fewer still can point to a board that has updated its oversight model, assigned explicit AI accountability under SMCR, and set risk tolerances designed for AI rather than inherited from frameworks that predate it.

That is the real work of AI transformation. It does not appear in a vendor contract or a pilot result. It appears in how quickly an institution moves from idea to decision to deployment: repeatedly, reliably, at scale.

The technology is available to everyone. The organisation is not. That is the only differentiator that compounds.

How Teradata helps 

 Teradata works with ten of the top ten global banks to build the data and AI foundations that make enterprise-scale deployment possible. That means unified data pipelines that survive production, AI models that can be governed, monitored, and explained to regulators, and the agentic infrastructure that connects analytical outputs to operational decisions. We help institutions move from AI experimentation to AI at scale, with governance built in from the start rather than retrofitted under pressure. To find out more, visit teradata.com.

Read the previous blogs: