Reflections from two days at the Banking Transformation Summit, London
Last week I had the opportunity to attend the Banking Transformation Summit in London. It was two intensive days of panel sessions, frank conversations, and genuinely thought-provoking debate about where financial services is heading. As Impact and Engagement Manager for the UKFin+ Network, I went in looking for signals: what are the problems that practitioners are truly wrestling with, and where does rigorous, independent research need to step in? I came away with a lot of notes, new contacts, and one line that has not left my mind since.
The Moment That Stopped Me in My Tracks
Across many of the sessions, the phrase “human in the loop” came up repeatedly as the accepted shorthand for responsible AI deployment. And then, in one session, someone said something that cut right through the consensus:
“It is not enough to just have a heartbeat in the loop. It has to be an expert.”
have been in enough rooms where this conversation happens to know that this distinction is rarely made out loud. And it should be. We cannot just put a human at a checkpoint and call it governance. We need people who understand what they are looking at – domain experts, informed practitioners, people who can interrogate an AI output, not just approve it.
In a highly regulated sector like financial services, where AI systems are increasingly involved in credit decisions, fraud detection, complaints handling, and customer communications, the difference between a heartbeat and an expert is the difference between compliance theatre and genuine accountability.
The AI Governance Conversation Is Growing Up
Much of the summit was devoted to how banks are moving from experimenting with AI to scaling it responsibly. A recurring theme was the tension between the speed of innovation and the pace of governance. One framing that resonated strongly was the idea of “Goldilocks governance”: controls that are tight enough to protect customers and institutions, but not so cumbersome that they kill the very programmes they are meant to protect. Governance, if designed well, should be an accelerator and not a friction point.
Organisations that integrate accountability from the start and embed governance into their architecture rather than bolting it on at the end are moving faster and more confidently. Those treating compliance as a checkbox are creating risk and slowing themselves down.
There was also a significant conversation about what responsible AI actually looks like in practice. Guardrails, we heard, must evolve continuously, as what was appropriate twelve months ago may already be irrelevant. And the tools to monitor AI behaviour, such as adversarial testing, logging, and performance tracking, are becoming as important as the models themselves. You cannot deploy and walk away.
Trust Is Not an Abstract Value, but an Operational Requirement
Another thread that ran through both days was trust with customers, colleagues, and regulators. Several sessions made the point that customers are adopting AI-powered experiences faster than most banks anticipated, and they are arriving with expectations, not just curiosity. They expect their data to be used ethically. They expect transparency. And when things go wrong, they expect a clear explanation rather than a letter that reads like a process map.
One example was a case in which a generative AI model was embedded in a complaints-handling process, with colleagues actively involved in reviewing and owning its outputs. What was described was not a passive sign-off arrangement: the people in the loop had become advocates for the system’s performance, stakeholders in its quality, and skilled enough to know when something was off. That is what expert-in-the-loop looks like in practice. Also, transparency about architecture, decision logic, and testing methodology builds the kind of trust that allows banks to move at the pace their customers are demanding.
The People Agenda Cannot Be an Afterthought
This brings me back to the expert-in-the-loop point, and why it connects so directly to the work UKFin+ does.
The summit made clear that the sector understands it has a skills challenge. AI is creating new roles, reshaping existing ones, and making redundant others, often simultaneously within the same organisation. One initiative mentioned aimed to reskill or upskill 100,000 individuals within banking by 2030, recognising that investing in people through this phase is as important as investing in the technology itself.
But reskilling is not the same as creating experts. There is an important distinction between building broad AI literacy across an organisation (which is essential) and cultivating the deep, domain-specific expertise required to meaningfully oversee AI systems in high-stakes environments. Financial crime detection, credit underwriting, complaints resolution, ESG reporting: these are areas where the consequences of a poorly overseen AI output can be serious and far-reaching. ‘Someone checked it’ is not sufficient. ‘Someone who understood what they were looking at checked it’ is a very different standard.
What does meaningful expert oversight of AI systems require? How do we design roles, training pathways, and governance structures that build that capacity? How do we measure whether human oversight is substantive or performative? These are complex, interdependent, and resistant to simple solutions, wicked problems, and they will not be solved by any single institution working in isolation.
These are the kind of challenges that UKFin+ is designed to address. By connecting academic researchers with industry practitioners and regulators, we can investigate these questions with the rigour and independence they deserve.
ESG: From Reporting Burden to Business Transformation
One of the most energising sessions I attended focused on the shift from ESG as a compliance exercise to ESG as a genuine business transformation. The message was consistent: sustainability cannot sit in a silo. It needs to be embedded in data infrastructure, decision-making, and day-to-day operations. And the data model required for meaningful sustainability reporting turns out to be the same joined-up, high-quality data model that effective AI deployment demands. For SMEs in particular, the conversation is evolving. The framing is moving away from ‘net zero obligations’ and towards ‘business resilience and cost reduction’. That reframe matters. It makes sustainability something SMEs can engage with on their own terms, rather than an additional overhead imposed from above.
What Banks Need to Leave Behind
One of the more entertaining sessions of the summit asked participants to nominate what banking most needs to consign to history. The answers were revealing: the gap between strategy and execution, dysfunctional silo culture, legacy end-user computing that lives in spreadsheets no one officially owns, and the tendency to launch AI initiatives without clear governance over who is responsible when something goes wrong.
What connected all of these was a simple observation: AI is powerful if you have the organisational capability to execute it. If you do not, it simply creates new versions of the same old problems – just faster, and harder to trace.
What This Means for Research, and for UKFin+
Conferences such as the Banking Transformation Summit are important for the insights they provide, and for the gaps in knowledge they reveal about the sector. The questions being asked on stage about expert oversight, responsible scaling, and the human dimensions of AI governance remain unanswered. They are being actively negotiated. And that is where independent, rigorous research has a critical role to play. Because the sector is right that we need humans in the loop. We just need to be much more honest about what ‘human’ actually means in that sentence.