My Advice to Banks on AI: Gianluca Berghella of Armundia Group

I spoke with Gianluca, Group Chairman & CEO of Armundia Group, a techfin company building modular platforms for banks, asset managers and insurers. With over 30 years in financial services technology, Gianluca shares practical advice on the AI and data decisions that will define competitive advantage in banking over the next 12–18 months.

Over to you Gianluca - my questions are in bold:

Can you give us an introduction to you and an overview of your organisation?

My background is in economics, but I have spent my career thinking about technology applied to financial services. I co-founded Armundia in 2007 with a clear conviction: that the industry needed to support and evolve its processes through technology built on a deep understanding of its operational and strategic logic, with a medium-to-long-term view.

Since then, the group has grown into an international organisation with a presence in Italy, Luxembourg, the United Kingdom, Albania and Asia.

Armundia Group is what we call a techfin — a technology company that has built its entire capability around financial and insurance services as a vertical market. We work with banks, asset managers, wealth managers and insurers, supporting them in evolving their operating models through the 3SIXTY Finance and 3SIXTY Insurance frameworks: modular, customisable architectures that allow institutions to innovate gradually and in a controlled manner, without having to stake everything on radical transformations whose timelines and costs are notoriously difficult to predict. After more than 30 years in this industry, I remain convinced that sustainable change in complex, heavily regulated systems must be incremental, always grounded in operational continuity — and that technology only truly delivers when it addresses the real problems of the people who use it.

If you were advising a bank CEO today, what would you say is the single biggest mistake they're making with data and AI?

Treating data quality as someone else's problem. I speak with many bank executives who are genuinely excited about artificial intelligence — and rightly so — but who have not yet come to terms with the fact that AI is only as reliable as the data it operates on. You cannot build trustworthy, regulation-compliant outputs on top of fragmented, inconsistent, poorly governed data. Yet this is precisely what many institutions are attempting to do.

The consequence is that AI initiatives either fail quietly — producing results that no one fully trusts — or they succeed technically but cannot be scaled, because the underlying data infrastructure is not capable of supporting them.

There is, however, a second mistake, often connected to the first: approaching AI as a standalone project, separate from the institution's broader operational strategy. AI is not a technology you install — it is a capability you build over time, starting from clear processes, reliable data and people who know how to use it. Before investing seriously in this space, every bank should ask itself a harder question: do we actually know what our data looks like, where it comes from, and who is responsible for it?

What's one AI or data capability banks should prioritise in the next 12–18 months, and why?

Agentic AI — the ability to deploy AI systems capable of executing complex, multi-step tasks autonomously, within defined boundaries and under human supervision. We are moving beyond the phase in which generative AI was primarily a content-production tool, and into a phase where AI agents interact with live systems, orchestrate operational workflows and produce outputs with real consequences for business processes.

The banks that address this transition now — with a gradual approach and control mechanisms built in from the outset — will find themselves in a considerably stronger position than those who arrive late. The key word, however, is "supervision": in a heavily regulated environment such as banking, the value of agentic AI depends entirely on the institution's ability to maintain accountability for every decision the system produces. Autonomy without governance is not an advantage — it is a liability.

The most concrete starting point for many banks is not necessarily the most visible one: building agents capable of operating reliably on internal data — verifying, reconciling, flagging anomalies — before extending that capability to more complex, high-impact processes.

Where do you see banks overestimating AI, and where are they underestimating it?

Overestimation tends to manifest at the strategic level. There is a tendency, among boards and executive committees, to treat AI as a solution in itself — as though adopting the technology were sufficient on its own to generate a competitive advantage. It is not. AI amplifies the quality of what already exists: if processes are poorly structured and data is unreliable, an AI layer does not fix the problem — it accelerates it. The real risk is not that AI will fail to work. It is that it will work exactly as intended, on foundations that should never have been left unchanged.

Underestimation, paradoxically, tends to occur at the opposite level. Those who work daily with data — reconciliation teams, compliance analysts, back-office staff — often regard AI with scepticism, because the applications they have seen feel remote from the concrete problems they face every day. That scepticism is understandable, but in many cases unfounded: the practical benefits of well-targeted AI in areas such as data validation, anomaly detection and document processing are significant, achievable with contained investment and measurable within a reasonable timeframe.

The gap between boardroom enthusiasm and operational scepticism is, in my view, one of the most underappreciated challenges banks face today. Closing it requires concrete use cases, demonstrable results and a degree of internal honesty that is rarer than it should be.

What does "good" actually look like when AI and data are working well inside a bank?

The most honest answer is: you don't see it. When AI and data are working well, what becomes visible is not the technology — it is the quality of decisions and the fluidity of processes. People work with more reliable information, arrive at meetings with analysis that has already been verified, spend less time gathering and reconciling data and more time thinking about what to do with it. Technology becomes infrastructure, in the most positive sense of the word: present, necessary, invisible.

In concrete operational terms: a relationship manager who previously spent hours consolidating a client's portfolio data can arrive at a meeting with that analysis already prepared, updated in real time and presented in an immediately useful format. A compliance team that previously reviewed dozens of documents manually can focus on the exceptions that genuinely require human judgement, because routine verification has been automated. An operation that once generated queues of emails and manual reconciliations can be handled by an agent working across internal systems, flagging anomalies and leaving the operator only the decisions that actually matter.

The measure I use to assess the success of these implementations is not the sophistication of the technology deployed. It is a much simpler question: can the people using it do their jobs better, with greater confidence in the information they are acting on? If the answer is yes, we are moving in the right direction.

What's the hardest AI or data decision bank executives are avoiding right now, and why?

The question of ownership. Someone in the organisation must be unambiguously responsible for the quality, governance and evolution of the institution's data — and that person must have genuine authority, not merely a job title. It is an organisational decision before it is a technological one, and that is precisely why it is so consistently deferred: it disturbs the internal status quo, redraws boundaries of responsibility between functions that rarely communicate, and produces no immediately visible results.

Most banks have distributed data responsibilities across business lines, IT, compliance and risk in ways that have accumulated over time and are difficult to trace back to any coherent design. The result is that when it comes to acting — on a quality problem, an architectural decision, a new AI initiative — it is rarely clear who has the final word. And in the absence of clarity, things are postponed.

The same problem surfaces with AI: who is accountable when a system produces an output that turns out to be wrong? In a regulated environment, "the algorithm decided" is not an acceptable answer — not for regulators, and ultimately not for clients either. Establishing clear lines of ownership — over data, over AI outputs, over model behaviour over time — is unglamorous, politically complex work. It is also, in my view, the necessary condition for everything else to function. The banks that have addressed this issue, even partially, are already operating with a tangible advantage over those that continue to avoid it.

Thank you Gianluca! You can connect with Gianluca on his LinkedIn Profile and find out more about the company at www.armundia.com.