If David Billings asked the room to lift its head at Innovation Day 2026 in London, the panel that followed grounded the conversation in operational reality.
Moderated by Amit Singhal (EPAM), the discussion brought together leaders working at the sharp end of AI adoption across financial services, banking, insurance and energy – including Abhay Pradhan from LSEG, Vincent McElwain from the Bank of Ireland, Mark Mamone from Willis Towers Watson (WTW) and Oliver Blackwood from Ensek.
Different sectors. Different constraints. But a shared reality: AI is already in motion, and organizations are under pressure to make it deliver.
The gap is real and it’s not closing on its own
The panel opened with a familiar tension: the growing gap between AI ambition and actual impact. Across organizations, investment is rising, experimentation is widespread, and expectations are high. But value remains uneven. Much of what is labelled “AI adoption” still sits on top of existing systems rather than reshaping them. For Abhay Pradhan (LSEG), the shift is already clear:
“We don’t want to build the model. Our strength is trusted data and making it accessible wherever our clients are.”
The implication is subtle but important. The race is no longer just about capability. It is about where that capability sits and how effectively it connects.
Across the panel, it was clear there is little appetite for model‑centric thinking. Instead, value is moving toward data, context and integration. Being able to place your capabilities inside the environments where decisions are already being made is becoming more important than owning the underlying technology.
Regulation reshapes how certain companies move
In highly regulated industries, AI adoption carries a higher burden of proof. But rather than acting purely as a constraint, regulation is starting to function as a forcing mechanism. Vincent McElwain (Bank of Ireland) described the reality: “We are highly regulated, that slows lift‑off. But once something works, it delivers real, measurable value.” Oliver Blackwood (Ensek) framed it differently, pointing to the structural upside:
“Regulation gives you clear guardrails and those are easier to embed into systems than into human processes.”
The result is a shift in where quality happens: where it used to sit at the end of the process, it is now designed in from the start.
Bottlenecks reveal where the real issues are
One of the most consistent insights from the panel was that AI accelerates parts of the system, but not the system as a whole. Coding is faster. Prototyping is easier. Engineering throughput is improving. But end‑to‑end delivery is not accelerating at the same rate. Abhay Pradhan (LSEG) named the constraint directly:
“Coding is not the problem. The real blocker is evaluation: how do you make sure it’s right?”
As one bottleneck is removed, another becomes visible. Data quality. Validation. Governance. Alignment. Instead of simplifying organisations, AI makes their dependencies harder to ignore.
From users to agents: the next interface shift
The panel also made clear that the next transformation is not just about capability, but about interaction.
Instead of interacting directly with systems, users are increasingly delegating that interaction to AI agents. Today, most interactions still happen between humans and digital platforms, with AI supporting parts of the experience. That is already shifting toward human‑to‑agent interaction, where people rely on AI to interpret intent, navigate systems and act on their behalf. The next step, which several panelists pointed to, is a move toward agent‑to‑agent interaction where systems communicate directly with each other, negotiate, transact and execute tasks without a human in the loop.
Vincent McElwain pointed to what this means for banking:
“People will want to bank through Claude. The question is how we enable that without removing ourselves from the journey.”
Oliver Blackwood extended that idea further: “The real shift happens when there are no humans in the loop, when services talk directly to each other.” Interfaces are becoming less visible. But the underlying systems (and the responsibility that comes with them) are becoming more complex.
The hardest part is still the human part
For all the focus on technology, the most complex challenge remains organisational. Teams are navigating mixed reactions: curiosity, skepticism, concern. Adoption is uneven. Confidence builds through experience, not messaging. Mark Mamone (Willis Towers Watson) captured the core issue:
“The human aspect of transformation is often missed. With AI, it’s the most critical part.”
Across the panel, the dominant narrative was not replacement, but redistribution. Less repetition. More judgment. More ownership of outcomes.
What not to do
The panel closed with a question that cut through the noise: what should others avoid? The answers were immediate and consistent.
“Watch your token spend.”
Abhay Pradhan, Group Head of Analytics & AI, LSEG
“Don’t just spread tools and expect magic – bring people on the journey.”
Vincent McElwain, Head of Tribe Technology, Bank of Ireland
“Start from the value you want – not from the technology you’re excited about.”
Mark Mamone, CTO ICT, Willis Tower Watson
“If you don’t have high‑performing teams, throwing AI at the problem won’t fix it.”
Oliver Blackwood, CTO, Ensek
What this panel made clear
Though they presented different angles, each panelist had the same underlying message: AI is no longer experimental... but it is not yet structural.
Organizations are caught in the transition. They are moving beyond isolated pilots, but have not yet fully reconfigured how their systems, teams and decisions connect. That gap is where most of the friction (and most of the opportunity) currently sits.
The companies that move ahead will not be those that adopt AI the fastest. Nor those that experiment the most visibly. They will be the ones that resolve these internal tensions early: between speed and control, between capability and coordination, between technology and the people expected to use it.
Because in the end, AI does not create alignment. It exposes whether it was there to begin with.
Contributor in this article
