In the debate on artificial intelligence in finance, one confusion keeps resurfacing: equating computational speed with decision quality. AI can read financial statements, detect patterns, flag anomalies and accelerate pre-screening and other repetitive tasks. But when decisions become genuinely structural, complex corporate credit, structured transactions, capital market dynamics and extraordinary finance operations, accountability, transparency of assumptions, reputation and the ability to defend an investment thesis in front of sophisticated counterparties remain central.
In this interview, Mario Nuti, finance professional, Executive Director and Senior Financial Advisor specialising in Corporate & Investment Banking, M&A, Debt and Equity Capital Markets, as well as director of a credit mediation company registered with the OAM, draws a clear line: yes to AI as an operational facilitator in ordinary finance; no to the idea that it can replace judgement, accountability and the “signature” in extraordinary transactions.
Juliette Haller : Let’s start with the basics. When we talk about AI in banking, what is the most common promise?
Mario Nuti: The recurring promise is to reduce time and costs and – at least in part – to make risk assessment more consistent. This promise is relevant when AI is applied to repetitive and standardised tasks. The problem begins when one attempts to extend this automation to decisions that are intrinsically non-linear, which cannot be reduced to a stable set of variables.
Where do you see the first genuinely effective field of application?
Mario Nuti: Retail credit and ordinary finance: preliminary file analysis, document collection and verification, consistency checks, ratio analysis, anomaly alerts. AI can save time and free up operational capacity. But it remains a support tool, not a substitute for accountability.
Concretely, what does “support tool” mean in a banking context?
Mario Nuti: It means that AI can produce a preliminary analysis, a classification, a list of warning signals. But someone must always assume responsibility for the final decision. In banking, accountability is not a detail – it is the backbone of the system. If something goes wrong, one cannot simply say: “the algorithm told me so.”
You often distinguish between ordinary finance and extraordinary finance. Why is this distinction so important?
Mario Nuti: Because in ordinary finance complexity is more limited and many assessments rest on similar structures: processes, risk models, parameters, collateral logic. In extraordinary finance – complex corporate transactions, structuring, market and negotiation dynamics, elements come into play that an algorithm captures poorly: context, reputation, management quality, counterparty behaviour, strategic objectives, timing and market sentiment. It is a different profession.
In short: ordinary finance “yes”, extraordinary finance “no”?
Mario Nuti: Yes. In ordinary finance, AI can play a significant role. In extraordinary finance, the idea of AI becoming the “decision-maker” is misleading. It can help with technical components, but it does not replace what makes a transaction credible and executable.
What is the main obstacle to AI becoming a true decision engine?
Mario Nuti: Transparency and governance. A simple question: who built the algorithm, with what assumptions, what data, and under what controls? If this is not clear, it becomes difficult to trust the result, especially when the stakes are high.
This naturally leads to regulation. Is the absence of a clear framework a concrete problem?
Mario Nuti: Yes, because without rules and without clearly assigned accountability, the algorithm remains a black box. In finance, a black box is a structural problem – not only for compliance, but for the quality of risk taken. If one cannot explain how a judgement was produced, one cannot defend it.
Does this create a paradox where AI adds a layer of control instead of simplifying?
Mario Nuti: Exactly. If AI produces a result but the bank must redo human checks because it cannot explain or justify the output, then nothing has been simplified. A step has been added. This is one of the most frequent paradoxes.
When it works well, what kind of output should AI provide?
Mario Nuti: Interpretable results. It is not enough to indicate “high risk” or “low risk”. Traceability is required: which variables mattered, which signals oriented the analysis, which data were used, and what is their quality. If the result cannot be explained, it is fragile.
From a company’s perspective, how will AI transform the bank–company relationship?
Mario Nuti: It will transform the operational layer for standard cases: speed, document flows, first analysis. But for companies seeking significant credit lines or structured operations, the fundamentals remain: credible governance, consistent and readable figures, planning capacity, reputation and relationship management. An algorithm does not replace perceived reliability.
You often mention reputation and networks. Some will say this is unmeritocratic.
Mario Nuti: It is uncomfortable, but realistic. Finance is risk management, and risk is not purely numerical. Reputation is a condensed signal of long-term reliability: if someone is known for honouring their commitments and managing crises properly, that reputational capital carries weight. It is not the only factor, but ignoring it amounts to misunderstanding how markets actually work.
In a complex corporate transaction, what makes a project “financeable” beyond the numbers?
Mario Nuti: The solidity of the narrative and the credibility of the person endorsing it. A model is not enough: one needs a logic that withstands due diligence, stress tests, negotiation and tough questions. One needs someone capable of sitting at the investors’ table and fully assuming the claims put forward.
Even in extraordinary finance, are there phases where AI can intervene without encroaching on judgement?
Mario Nuti: Yes: preliminary document review, data comparison, preparation of materials, detection of inconsistencies between sources. Everything that falls under operational excellence can be enhanced. But the transition from technical support to strategic decision remains human.
Some claim that AI will replace junior analysts. Do you share this view?
Mario Nuti: AI will reduce some of the repetitive work, that is certain. But be careful: professional training is also built through that work. If it is eliminated without rethinking how judgement and competence are formed, one risks obtaining organisations that are faster in the short term but less robust over time.
Is there a risk of “false security”, believing the algorithm makes the bank more prudent, when in fact it makes it simply more uniform?
Mario Nuti: Yes. Uniformity is not synonymous with quality. If everyone uses the same logic, the system can become more procyclical: too open in favourable periods, too restrictive in unfavourable ones. Human oversight is always needed – people capable of interpreting context and knowing when a rule should be adapted rather than applied mechanically.
Does AI genuinely reduce information asymmetry?
Mario Nuti: It reduces it when it aggregates data and makes them readable. But in complex operations, the asymmetry is not only informational: it is also strategic and behavioural. Having more data is not enough; one must understand what they do not say.
Looking a few years ahead, what evolution do you anticipate?
Mario Nuti: More automation in standard processes, especially in retail and ordinary finance. More mature tools in large organisations capable of investing in governance, control and data quality. But I do not believe AI will replace decision-making in extraordinary transactions. There, competence, reputation, negotiation and accountability will remain decisive.
If you were to leave a final message to those who present AI as a total solution for finance?
Mario Nuti: Finance is not a spreadsheet. Data are essential, but the decision is an act of responsibility. AI is a useful accelerator, as long as it remains in its place: supporting processes and analysis. When one claims it can replace judgement, credibility and the signature, one oversimplifies, and in finance, oversimplification has a cost.
Editorial note: Interview transcribed and lightly edited for readability
Une version française de cette interview est disponible pour nos lecteurs francophones.
Master’s student in Digital Economy Law | M2 Droit de l’Économie Numérique